ISBN 978-9915-698-43-4
Scientific research methodology
applied to artificial intelligence and
data science: General approach
García Cruz, Josefina Arimatea; Camac Tiza,
Maria Maura; Asián Quiñones, Carlos
Alberto; Piedra Isusqui, José César;
Juscamayta Ramírez, Luis Alberto; Huaman
Fernandez, Jackeline Roxana; Rosas
Cajahuanca, Judith del Rosario
© García Cruz, Josefina Arimatea; Camac Tiza,
Maria Maura; Asián Quiñones, Carlos
Alberto; Piedra Isusqui, José César;
Juscamayta Ramírez, Luis Alberto; Huaman
Fernandez, Jackeline Roxana; Rosas
Cajahuanca, Judith del Rosario, 2025
First edition (1st ed.): October, 2025
Edited by:
Editorial Mar Caribe ®
www.editorialmarcaribe.es
Av. Gral. Flores 547, 70000 Col. del
Sacramento, Department of Colonia,
Uruguay.
Cover design and illustrations: Isbelia
Salazar Morote
E-book available at:
https://editorialmarcaribe.es/ark:/10951/is
bn.9789915698434
Format: Electronic
ISBN: 978-9915-698-43-4
ARK: ark:/10951/isbn.9789915698434
Editorial Mar Caribe (OASPA): As a member of the
Open Access Scholarly Publishing Association, we
support open access in accordance with OASPA's
code of conduct, transparency, and best practices for
the publication of scholarly and research books. We
are committed to the highest editorial standards in
ethics and deontology, under the premise of "Open
Science in Latin America and the Caribbean"
Editorial Mar Caribe, signatory No. 795 of 12.08.2024
of the Declaration of Berlin
"... We feel compelled to address the challenges of the
Internet as an emerging functional medium for the
distribution of knowledge. Obviously, these advances can
significantly modify the nature of scientific publishing, as
well as the current quality assurance system...." (Max
Planck Society, ed. 2003., pp. 152-153).
CC BY-NC 4.0
Authors may authorize the general public to reuse
their works solely for non-profit purposes, readers
may use one work to generate another, provided
credit is given to the research, and grant the publisher
the right to first publish their essay under the terms
of the CC BY-NC 4.0 license.
Editorial Mar Caribe adheres to the UNESCO
"Recommendation concerning the Preservation of and
Access to Documentary Heritage, including Digital
Heritage" and the International Reference Standard
for an Open Archival Information System (OAIS-ISO
14721). This book is digitally preserved by
ARAMEO.NET
Editorial Mar Caribe
Scientific research methodology applied to
artificial intelligence and data science: General
approach
Colonia, Uruguay
2025
Scientific research methodology applied to
artificial intelligence and data science: General
approach
Index
Page
Introduction
7
Chapter 1
The Fifth Paradigm:
Redening Scientic
Inquiry through Articial
Intelligence and Big Data
Methodologies
10
Chapter 2
The Epistemology of
Algorithms: A
Comprehensive
Framework for Scientic
Methodology in Articial
Intelligence and Data
Science
36
Chapter 3
Beyond the Binary: The
Convergence of
Qualitative Inquiry and
Articial Intelligence
60
Chapter 4
Quantitative Research
Methodology in Articial
Intelligence and Data
Science: A Comprehensive
Framework for Empirical
Analysis
89
Conclusion
118
Bibliography
120
Index of Tables
Page
Table 1: AI-Enhanced Research Platforms
12
Table 2: Comparative Analysis of Research
Methodologies
27
Table 3: Comparative Analysis of Research Lifecycles
39
Table 4: Selection Matrix for Statistical Tests in AI
46
Table 5: Distant, close and blended reading
64
Table 6: Tool category of key platforms
69
Table 7: Key Methodological Comparisons
78
Table 8: Common causes of reproducibility failure in AI
94
Table 9: Summary of Quantitative Metrics and Methods
by Domain
109
Dedication
“… To the researchers of tomorrow, who will inherit these tools to decipher the
mysteries that we can barely glimpse today. May the Fifth Paradigm be the
compass that guides your curiosity towards some knowledge without borders.
And, for those who look for paerns in chaos and beauty in data..."
7
Introduction
The history of science has gone through various paradigms: from the
empirical observation of natural phenomena and the theoretical formulation of
laws to the computational simulation of complex systems. Today, we are
immersed in what Jim Gray called the "fourth paradigm": data-intensive scientic
discovery. In this new scenario, Articial Intelligence (AI) and Data Science are
not mere technical engineering tools; they have become the fundamental lens
through which we interrogate reality.
However, the dizzying advance of these technologies has brought with it
an epistemological challenge: the gap between predictive capacity and scientic
validity. In the race to optimize hyperparameters and reduce mean square error,
it is often forgoen that an AI model, in a research context, is not just a software
product, but a mathematically formalized hypothesis that must be tested,
disproved, and explained.
There is a latent tension in current practice. On the one hand, software
engineering seeks to make the system "work"; on the other, the methodology of
research requires understanding "why it works" and under what conditions it is
reproducible.
This book, Scientic research methodology applied to articial intelligence and
data science: General approach, was born from the imperative need to systematize
research in this eld. It seeks to answer critical questions that many students and
professionals face: How do you formulate a valid research question in a Machine
Learning project?, What dierentiates a technology implementation project from
a scientic research thesis?, How do we address the reproducibility crisis in
8
"Black Box" models?, What are the ethical and bias standards that should govern
data collection?.
The central objective of this text is to serve as a bridge that connects the
rigor of the traditional scientic method—with its emphasis on hypothesis,
controlled experimentation, and causal inference—with the exibility and power
of modern Data Science workows (CRISP-DM, KDD, etc.).
It is not a programming manual in Python or R, nor a compendium of
neural network architectures. It is a guide to scientic thinking applied to data.
Here, the reader will learn how to structure big data chaos within a
methodological framework that ensures robust, generalizable, and ethically
responsible ndings.
Through the four chapters, we will break down the AI research lifecycle
from a methodological perspective:
- Epistemological Foundations: We will explore the nature of knowledge
generated by inductive and deductive algorithms.
- Research Design: Denition of scope, selection of variables and the
dichotomy between explanatory and predictive studies.
- Data Management as Evidence: Processing, cleansing, and the importance of
data quality beyond volume.
- Validation and Metrics: Going beyond accuracy as the only metric;
sensitivity analysis, robustness, and rigorous cross-validation.
- Ethics and Communication: The Researcher's Responsibility in the Face of
Algorithmic Bias and How to Write Technical Findings for a Scientic
Audience.
Ultimately, this book proposes that the best Articial Intelligence is not the
one that simply processes data faster, but the one that helps us beer understand
9
the world with greater rigor and truth. Welcome to the study of scientic
methodology in the age of data.
10
Chapter 1.
The Fifth Paradigm: Redefining Scientific
Inquiry through Artificial Intelligence and
Big Data Methodologies
1. The Epistemological Shift in Research Methodology
The history of scientific discovery is often categorized into distinct paradigms, each
defined by the primary instrument of inquiry. The first paradigm was empirical, driven by
direct observation of natural phenomena. The second was theoretical, characterized by the
formulation of laws and generalizations, such as Maxwell’s equations or Newton’s laws of
motion. The third, emerging in the mid-20th century, was computational, utilizing
simulations to model complex systems that were analytically intractable. The fourth
paradigm, articulated by Jim Gray in the early 2000s, was data-intensive, predicated on the
analysis of massive datasets generated by instruments and sensors. Today, we stand at the
precipice of a "Fifth Paradigm".1 This new era is not merely an extension of the data-
intensive fourth paradigm; it represents a fundamental qualitative shift driven by the
integration of Artificial Intelligence (AI) and Big Data into the very fabric of the scientific
method.
In this Fifth Paradigm, the relationship between the researcher and the object of
study is mediated—and in some cases, fully managed—by intelligent algorithms. We are
witnessing the transition from "human-generated hypotheses tested on data" to "AI-
generated hypotheses validation by humans".3 The epistemological implications are
profound. Traditional research methodologies, grounded in the scarcity of data and the
necessity of sampling, are being upended by an abundance of data and the capability for
"total population analysis".4 The linear progression of the scientific method—observation,
hypothesis, experimentation, analysis, conclusion—is being compressed into iterative, high-
11
velocity loops executed by autonomous agents.6
This report provides an exhaustive analysis of these emerging methodologies. It
explores how Large Language Models (LLMs) are revolutionizing literature synthesis,
transforming it from a manual bottleneck into an automated, semantic reasoning process. It
examines the "Causal Revolution," which seeks to move AI beyond mere correlation to the
understanding of cause-and-effect relationships essential for policy and scientific
intervention.8 It investigates the rise of "Generative Social Science," where synthetic
populations mimic human behavior, allowing for in silico sociological experiments.10
Finally, it addresses the critical challenges of this new era: the reproducibility crisis
exacerbated by data leakage, the ethical imperatives of algorithmic fairness, and the
changing role of the human scientist in a loop increasingly dominated by silicon intelligence.
2. The Revolution in Literature Review and
Knowledge Synthesis
The initial phase of any research project—the literature review—has traditionally
been a labor-intensive process of manual discovery, reading, and synthesis. The exponential
growth of scientific publishing, with millions of papers published annually, has made
comprehensive manual review nearly impossible. The integration of AI, specifically
through semantic search and Retrieval-Augmented Generation (RAG), has fundamentally
altered this landscape, transitioning the researcher's role from a consumer of text to an
architect of inquiry.
2.1 From Lexical Search to Semantic Reasoning
For decades, information retrieval in academia relied on lexical or keyword-based
search. This methodology is inherently brittle; a search for "cardiovascular disease" might
miss relevant papers that strictly use the term "heart failure" unless complex Boolean strings
are constructed. The Fifth Paradigm introduces Semantic Search, powered by vector
12
embeddings. In this model, text is converted into high-dimensional vectors, and retrieval is
based on the proximity of these vectors in semantic space rather than exact word matching.11
Semantic Scholar, a pioneer in this field, utilizes machine learning to understand the context
of citations. It distinguishes between a citation that merely mentions a prior work and one
that is foundational to the methodology, classifying them as "highly influential citations".12
This moves the metric of scientific impact from raw counts to semantic relevance.
Furthermore, tools like Elicit and Consensus leverage LLMs to perform "automated
extraction." Upon receiving a natural language query (e.g., "What is the impact of
continuous glucose monitors on long-term diabetes complications?"), these systems do not
just return a list of links. They scan the full text of relevant papers, extract the specific
findings, sample sizes, and methodologies, and present them in a structured matrix.13
This capability represents a methodological leap: the instantaneous generation of a
"Matrix of Evidence." A task that previously required weeks of manual coding—extracting
P, I, C, O (Population, Intervention, Comparison, Outcome) elements from dozens of
papers—can now be performed in minutes (see Table 1).
Table 1: AI-Enhanced Research Platforms
13
2.2 Automated Systematic Reviews and Screening Efficiency
The systematic review is the gold standard of evidence synthesis, particularly in
medicine and the social sciences. However, it is resource-intensive, often requiring teams of
reviewers to screen thousands of titles and abstracts to identify the few dozen that meet
inclusion criteria. New methodologies employ LLMs to automate this screening phase,
acting as a "second screener" or even a primary filter.
Recent studies comparing LLM performance to human reviewers have
demonstrated transformative results. In one comprehensive evaluation, an LLM-assisted
workflow reduced the total screening time from 54.7 hours (using traditional semi-
automated software like Rayyan) to just 25.5 hours—a time saving of over 50%.15 More
strikingly, when the LLM was used for the initial title/abstract screening, it achieved a
Negative Predictive Value (NPV) of 100%, meaning it successfully excluded irrelevant
studies without discarding a single relevant one.15 Other research indicates that LLM-
assisted screening can reduce manual effort by up to 95% in specific contexts.16
The methodology for systematic reviews is thus evolving from a "dual-human"
process to a "Human-AI" hybrid model. In this workflow, the AI serves as a high-recall filter,
flagging potential studies and providing rationales for exclusion, while the human expert
focuses on the final inclusion decisions and the subtle interpretation of complex findings.
This shift allows researchers to conduct "living systematic reviews"—reviews that are
continuously updated as new literature is published—rather than static snapshots that
become outdated within months.
14
2.3 Knowledge Graphs and GraphRAG: Solving the Hallucination
Problem
A critical limitation of pure LLMs in research is "hallucination"—the generation of
plausible but fictitious text, citations, or data points. In scientific writing, a hallucination rate
of even 1% is unacceptable. To mitigate this, advanced methodologies are moving beyond
simple "chat" interfaces to Retrieval-Augmented Generation (RAG) systems grounded in
Knowledge Graphs (KGs).17
A Knowledge Graph is a structured representation of facts, where entities (e.g.,
"Protein A," "Algorithm B," "Author C") are nodes and their relationships (e.g.,
"Upregulates," "Developed by," "Cites") are edges. The integration of KGs with LLMs,
known as GraphRAG, allows for multi-hop reasoning.
Consider a research question: "What is the connection between Gene X and Disease
Z?" A standard LLM might hallucinate a connection based on statistical probability in its
training data. A GraphRAG system, however, will traverse the knowledge graph:
1. Paper A links Gene X to Enzyme Y.
2. Paper B links Enzyme Y to Disease Z.
3. The system infers a potential pathway: Gene X -> Enzyme Y -> Disease Z.
The LLM then generates a response citing Paper A and Paper B as the evidence for
this inferred link.19 This methodology transforms the literature review from a retrieval task
to a logic discovery task, allowing researchers to uncover latent connections in the scientific
corpus that no single paper has explicitly articulated.20
2.4 The Risk of Bibliographic Simulacra and Algorithmic Literacy
Despite these advances, the "hallucination rate" (HR) remains a critical
methodological variable. Research evaluating tools like GPT-4 and Claude 2 in scientific
contexts mandates the reporting of HR and "Prompt Sensitivity" (PS).21 While models have
15
improved, they still occasionally fabricate citations or misattribute findings, particularly in
niche fields where training data is sparse.22
Furthermore, the ease of generating summaries poses a risk of "bibliographic
simulacra"—a state where researchers interact primarily with AI-generated syntheses rather
than the primary texts. This can lead to the propagation of subtle misinterpretations or the
loss of nuance. The Fifth Paradigm demands that researchers develop algorithmic literacy:
the ability to audit AI outputs, verify citations against primary sources, and understand the
provenance of the information presented.24 We are moving toward a methodology of "Trust
but Verify," where AI acceleration is balanced by rigorous human validation.
3. Big Data Methodologies: Beyond Sampling and
Significance
The traditional research paradigm was constrained by the high cost and logistical
difficulty of data collection. This constraint necessitated the development of sampling
theory—the mathematical framework for inferring the properties of a whole population
from a small, representative part. Big Data, characterized by the "Three Vs" (Volume,
Velocity, Variety) and now expanded to dimensions including Veracity, Value, and
Variability 25, fundamentally challenges the necessity of sampling.
3.1 "N=All": Total Population Analysis
In the digital age, researchers often have access to the entire population of interest—
every transaction in a financial market, every click on a website, or every genomic sequence
in a biobank. This shift to "N=All" renders traditional notions of statistical significance (p-
values) less relevant.4
When analyzing a total population, the concept of sampling error—the error
introduced by observing only a subset—vanishes. Differences observed in the data are, by
16
definition, real differences in that population. The methodological challenge shifts from
calculating sampling error to managing measurement error and model error. As noted by
Mayer-Schönberger and Cukier, "N=All" allows researchers to embrace the messiness of
real-world data rather than curating pristine, small samples.26
However, "N=All" is not a panacea. The phenomenon of "Big Data Hubris"—the
assumption that massive datasets automatically yield truth—remains a risk. The failure of
Google Flu Trends, which attempted to predict flu outbreaks based on search queries (N=All
searches), demonstrated that a massive dataset can still be biased if the underlying data
generation process (user search behavior) drifts over time or is influenced by media
coverage.5 Thus, the new methodology emphasizes representativeness over volume. A
dataset of 100 million tweets is an "N=All" of Twitter users, but it is not an "N=All" of the
global population. Researchers must now account for algorithmic selection bias inherent in
the platforms generating the data.27
3.2 Real-Time Streaming Analytics: Lambda and Kappa Architectures
Traditional research is retrospective: data is collected over a period, cleaned, and
then analyzed in batches. The "Velocity" of Big Data necessitates streaming analytics, where
data is processed in motion. This is particularly transformative in fields like public health
(epidemic tracking), sociology (sentiment analysis), and finance.28
Methodologies utilizing Apache Kafka, Flink, or Spark Streaming allow researchers
to analyze data streams in real-time.29 This requires a shift in data architecture.
● Lambda Architecture: A hybrid approach that uses a "speed layer" for real-time views
and a "batch layer" for comprehensive, high-latency analysis. This allows researchers
to see immediate trends while correcting for errors later.28
● Kappa Architecture: A simplified model that treats everything as a stream, allowing
for a single processing framework.
17
In this context, the research "instrument" is no longer a static survey but a dynamic
algorithm. Anomaly detection algorithms run continuously on these streams, identifying
outliers (e.g., a sudden spike in emergency room visits) that would be smoothed out in
aggregate batch analysis.30 This requires defining "windows" of analysis—sliding windows
or tumbling windows—rather than static cross-sections.
3.3 The Integration of Unstructured Data via Multimodal Analysis
The "Variety" of Big Data refers to the influx of unstructured data—images, video,
free text, and sensor logs. Traditional methodology required this data to be manually coded
before analysis. AI, particularly Deep Learning (CNNs and Transformers), allows for the
direct analysis of unstructured data.31
For example, in migration studies, researchers traditionally relied on lagged
government statistics. New methodologies integrate Call Detail Records (CDRs) from
mobile phones, satellite imagery indicating settlement growth, and social media activity to
map migration flows in real-time.33 This triangulation of disparate data sources—multi-
modal analysis—provides a holistic view but introduces complex ontological challenges.
How does a researcher reconcile the "truth" of a satellite pixel with the "truth" of a tweet?
The methodology requires a robust framework for data fusion, assigning confidence
weights to different modalities based on their known reliability and bias profiles.
4. Generative Social Science and Synthetic
Populations
One of the most avant-garde developments in research methodology is the use of
Generative AI to simulate human subjects. This field, termed Generative Social Science,
leverages the capabilities of LLMs to create "silicon subjects" or "generative agents" that
mimic human attitudes, behaviors, and social interactions.10
18
4.1 Generative Agents: The Silicon Subject
Traditional social science relies on surveys and human-subject experiments, which
are expensive, slow, and often difficult to replicate (WEIRD—Western, Educated,
Industrialized, Rich, Democratic—bias is a known issue). Generative Agents offer a radical
alternative. These are computational agents instantiated with specific personas (e.g., "a 35-
year-old nurse with two children and conservative political views") and placed in a
simulated environment.35
Research by Stanford and others has shown that these agents can replicate the
responses of real human populations on surveys like the General Social Survey (GSS) with
high fidelity, achieving up to 85% correlation with human responses.36 This allows
researchers to perform in silico experiments: testing the impact of a policy intervention, a
marketing campaign, or a social rumor on a synthetic population before deploying it in the
real world.
The methodology involves "Inverse Generative Social Science" (iGSS). Instead of
designing agents with simple, hand-crafted rules (as in traditional Agent-Based Modeling),
researchers use evolutionary computing or LLMs to evolve agents that can reproduce a
known macro-phenomenon.34 This shifts the focus from "what happens if rules X apply?" to
"what micro-rules must exist to explain macro-phenomenon Y?"
4.2 Synthetic Data in Medicine: Balancing Privacy and Utility
Beyond simulation, synthetic data is revolutionizing privacy-sensitive research,
particularly in healthcare. Access to patient records is often restricted by regulations like
HIPAA or GDPR. Synthetic data generation involves training an AI model (often a
Generative Adversarial Network or GAN) on real patient data to learn the statistical
distributions and correlations. The model then generates a new, artificial dataset that
preserves these statistical properties without containing any real individuals.38
19
This approach allows for Total Population Analysis without privacy compromise.
Researchers can share synthetic datasets openly, enabling reproducibility and collaboration
that was previously impossible. However, the methodology requires rigorous validation.
Researchers must perform:
● Utility Evaluations: Does the synthetic data yield the same regression coefficients and
predictive power as the real data?
● Privacy Evaluations: Is it possible to re-identify real individuals from the synthetic set
via linkage attacks?.40
Recent studies propose metrics to quantify "privacy leakage" in synthetic datasets,
ensuring that the trade-off between data utility and privacy is transparent. This "tiered
access" model—open synthetic data for exploration, restricted real data for final
validation—is becoming a standard workflow in medical research.41
4.3 The "Echo Chamber" Risk and the Limits of Simulation
While promising, Generative Social Science faces the "Echo Chamber" problem.
LLMs are trained on internet text, which reflects specific biases, cultural norms, and
linguistic patterns. A society of generative agents might simply reproduce the stereotypes
present in the training data rather than the nuanced reality of human behavior.27
There is a risk that researchers might mistake the simulation for the territory. The
methodology requires a continuous loop of validation: real-world data calibrates the
simulation, the simulation generates hypotheses, and the hypotheses are tested back in the
real world. Critics argue that relying too heavily on generative agents could lead to a
"positivist" drift in qualitative research, imposing rigid categories on the fluid nature of
human experience.42
5. The Causal Revolution: Beyond Correlation
For decades, the mantra of statistics has been "correlation does not imply causation."
20
Traditional machine learning excels at pattern recognition—identifying correlations in high-
dimensional space—but often fails to understand why those patterns exist. This limitation is
critical in fields like medicine and policy, where the goal is intervention, not just prediction.
The Causal AI movement, spearheaded by Judea Pearl and others, is introducing
methodologies that allow researchers to infer causality from observational data.8
5.1 Structural Causal Models (SCMs) and Directed Acyclic Graphs
(DAGs)
The core of this methodology is the Structural Causal Model (SCM). Unlike a neural
network, which is a "black box" of weights, an SCM is explicitly represented by a Directed
Acyclic Graph (DAG)—a visual map where nodes are variables and arrows represent causal
influence.9
Researchers use tools like DoWhy (Microsoft), CausalNex, and Pyro to build these
models. The methodology involves a distinct two-step process:
1. Causal Discovery: Algorithms analyze the data to suggest potential causal structures
(e.g., "does X cause Y, or does confounder Z cause both?").
2. Causal Inference: The researcher encodes their domain knowledge into the DAG and
uses it to estimate the effect of an intervention.44
This approach enables counterfactual reasoning—asking "What would have
happened if?" questions. For example, in a clinical trial, Causal AI can estimate the outcome
for a specific patient had they received the alternative treatment, a calculation impossible with
standard statistics that only observe the realized outcome.8
5.2 Prescriptive vs. Predictive AI
The shift from Predictive AI (what will happen?) to Prescriptive AI (how can we
make it happen?) is the hallmark of causal methodologies. In manufacturing, Causal AI is
used for Root Cause Analysis, distinguishing between a symptom (correlation) and the
21
actual defect source (causation).45
This methodology is crucial for mitigating spurious correlations. In Big Data, the
probability of finding a statistically significant but meaningless correlation increases with
the size of the dataset (the "Look-elsewhere effect"). Causal AI acts as a filter, rejecting
correlations that do not fit the causal structure of the world, thus enhancing the robustness
and generalizability of research findings.46 For instance, an AI might find a correlation
between ice cream sales and drowning deaths; a Causal AI model would identify
"Temperature" as the confounder causing both, preventing a policy recommendation to ban
ice cream to save swimmers.
5.3 Tools for Causal Inquiry
The ecosystem of tools supporting this methodology is maturing rapidly:
● DoWhy: A Python library that unifies causal inference under a single API (Model,
Identify, Estimate, Refute). It is particularly strong on the "Refute" step, allowing
researchers to stress-test their causal assumptions.44
● CausalNex: Allows researchers to use Bayesian Networks to infer causality, offering a
"Human-in-the-loop" interface where experts can manually correct the edges of the
learned graph.47
● TensorFlow Causal: Brings causal inference to the scale of deep learning, allowing for
causal discovery on massive datasets.47
6. Autonomous Experimentation: The Rise of the AI
Scientist
Perhaps the most futuristic development in research methodology is the automation
of the scientific process itself. Self-Driving Laboratories (SDLs) and AI Scientists are systems
capable of planning, executing, and analyzing experiments with minimal or no human
intervention.
22
6.1 Materials Acceleration Platforms (MAPs)
In materials science and chemistry, the search space for new molecules is practically
infinite (estimated at 10^{60} small molecules).48 Traditional trial-and-error is too slow.
Materials Acceleration Platforms (MAPs) combine robotics, AI, and high-throughput
screening to close the loop of experimentation.49
An SDL operates in cycles, often referred to as "Level 3" autonomy:
1. Design: An AI (often using Bayesian Optimization or Active Learning) selects the next
best experiment to run based on previous results, balancing exploration (trying new
things) and exploitation (refining known hits).
2. Synthesize: Robotic arms mix reagents and manage the reaction.
3. Characterize: Sensors measure the properties of the new material (e.g., conductivity,
absorbance).
4. Learn: The results update the AI's model, and the cycle repeats.49
These systems can run 24/7, accelerating discovery by orders of magnitude. The
methodology shifts the researcher's role from "bench scientist" to "system architect"—
designing the search space and the optimization parameters rather than pipetting liquids.51
Notable examples include setups at NC State and BU, where "dynamic flow experiments"
redefine data utilization in fluidic labs.52
6.2 "The AI Scientist": Automated Paper Generation
Beyond physical experiments, AI agents are now capable of conducting end-to-end
computational research. Sakana AI recently introduced "The AI Scientist," a comprehensive
system that can generate novel research ideas, write the necessary code, execute the
experiments, visualize the results, and write a full scientific paper—all autonomously.6
The workflow of such an agentic system is:
1. Ideation: The LLM brainstorms research directions and checks them against existing
23
literature (via Semantic Scholar) to ensure novelty.
2. Coding: An agent (like Aider) writes the experiment code (e.g., Python/PyTorch).
3. Execution: The code is run, and logs/results are captured.
4. Drafting: The LLM interprets the results and drafts a paper in LaTeX, generating
figures and citations.
5. Reviewing: An "Automated Reviewer" module critiques the paper, prompting the
scientist agent to revise the text or run additional experiments.53
While currently limited to computational domains, this challenges the definition of
authorship. It raises the prospect of "Recursive Scientific Improvement," where AI systems
design better AI systems. However, critics note that these systems currently struggle with
deep methodological innovation, often producing derivative or incrementally novel work,
and can suffer from "hallucinated" correctness where the code runs but the methodology is
flawed.54
7. Methodological Integrity and Ethics in the AI Era
The integration of AI and Big Data into research methodology is not a panacea; it
introduces systemic risks that, if unaddressed, threaten the integrity of science.
7.1 The Reproducibility Crisis and Data Leakage
Machine Learning-based science is facing a reproducibility crisis. A major driver is
Data Leakage—the accidental inclusion of information from the test set into the training
process. This leads to overly optimistic performance estimates that fail to generalize to new
data.55
Methodologists have identified profound taxonomies of leakage, including:
● Temporal Leakage: Using future data to predict the past (e.g., using a patient's
discharge diagnosis to predict their admission risk).
● Feature Leakage: Including variables that are proxies for the target label.
24
● Distributional Leakage: When the test set is not drawn from the same distribution as
the deployment environment.
To combat this, new methodologies mandate the use of Model Info Sheets and
rigorous Train-Test-Validation splits. The concept of reproducibility is expanding to include
"Algorithmic Reproducibility"—can another researcher, using the same code and data,
generate the exact same model? This is often difficult due to the stochastic nature of GPU
training and random seeds.55
7.2 Algorithmic Bias and Ethics
Data is never neutral; it is a social artifact. Algorithmic Bias occurs when AI models
amplify existing societal prejudices present in the training data. In medical research,
algorithms trained on predominantly white populations have been shown to misdiagnose
skin conditions in patients with darker skin tones.27
Methodologies for Fair AI are emerging to address this. These include:
● Pre-processing: Re-weighting the data to ensure balanced representation.
● In-processing: Adding fairness constraints to the loss function of the model.
● Post-processing: Adjusting the model's outputs to equalize error rates across
demographic groups.43
Ethical research in the Fifth Paradigm requires a "Data Nutrition Label" approach—
transparently documenting the provenance, composition, and limitations of the datasets
used.57
8. Human-AI Collaboration Models
As AI systems become more autonomous, the nature of human oversight is evolving.
The methodology now explicitly defines the human's role in the loop.
25
8.1 Human-in-the-Loop (HITL)
In HITL models, the human actively makes decisions at key points. The AI provides
a recommendation, but the human must "sign off" for the process to proceed. This is the
standard for high-stakes research, such as clinical trials or weapon systems research.58 For
example, in automated systematic reviews, the human is the final arbiter of inclusion.
8.2 Human-on-the-Loop (HOTL)
In HOTL models, the human plays a supervisory role. The system operates
autonomously, but the human monitors the performance metrics and intervenes only when
anomalies occur or confidence scores drop below a threshold. This is the typical model for
Self-Driving Laboratories and high-frequency trading algorithms.60
8.3 Human-in-Command (HIC)
HIC emphasizes that while AI may execute tasks, the strategic direction and ethical
responsibility remain solely with the human. This model is gaining traction in defense and
policy research, ensuring that "agentic" workflows do not drift from their intended human-
aligned goals.58
The consensus in methodological ethics is that for scientific discovery, HOTL is the
minimum standard. The "black box" problem means that even if an AI discovers a truth,
human validation is required to integrate that truth into the broader scientific canon.
9. Future Frontiers: Quantum Computing and the 2030
Outlook
Looking ahead to 2025 and beyond, the intersection of Quantum Computing and Big
Data promises to shatter current computational limits.
9.1 Quantum Machine Learning (QML)
26
Quantum computers, utilizing qubits and superposition, can solve optimization
problems (like protein folding or supply chain logistics) exponentially faster than classical
computers.62 Quantum Machine Learning (QML) will allow researchers to analyze datasets
of a complexity that is currently intractable. For instance, simulating the precise quantum
mechanics of a chemical reaction is impossible for classical supercomputers but native to
quantum devices. This will likely lead to a "quantum leap" in pharmacology and materials
science.63 By 2025, we expect to see the first "Quantum Advantage" in specific research
niches, such as optimizing catalysts for carbon capture.
9.2 The "Fifth Paradigm" by 2030
By 2030, predictions suggest that AI Agents will be proactive research partners. They
will not just answer questions but independently monitor the literature, identify gaps, and
propose hypotheses.65 The global market for these AI-driven research tools is projected to
reach hundreds of billions of dollars.65 The definition of a "scientist" will expand to include
those who experts in are orchestrating these silicon intelligences—"Prompt Engineers" of
the physical world.
The integration of Artificial Intelligence and Big Data into research methodology is
not merely an upgrade in tools; it is a fundamental restructuring of how we ask and answer
questions. We are moving from a scarcity-based epistemology—defined by sampling, linear
regression, and manual synthesis—to an abundance-based epistemology—defined by total
population analysis, deep learning, and autonomous generation.
This transition empowers researchers to tackle "wicked problems" of organized
complexity—climate change, pandemics, metabolic networks—that were previously
beyond reach. The cost of drug discovery could plummet, the speed of materials innovation
could skyrocket, and our understanding of social dynamics could become predictive rather
than merely descriptive.
However, this potential comes with the responsibility of rigour. The researcher of
27
the future must be a hybrid scholar: part domain expert, part data scientist, and part ethicist.
We must guard against the seduction of "easy answers" provided by black-box algorithms.
Our methodologies must remain anchored in causal reasoning, robust validation, and an
unwavering commitment to human interpretability. As we deploy agents that can read,
reason, and experiment, the ultimate goal of research remains unchanged: not just to process
data, but to generate meaning (see Table 2).
Table 2: Comparative Analysis of Research Methodologies
28
Chapter bibliographies
1. Abisoye, A. (2024). A Conceptual Framework for Integrating Artificial Intelligence
into STEM Research Methodologies for Enhanced Innovation. International Journal of
Future Engineering Innovations, 1(1). https://doi.org/10.54660/ijfei.2024.1.1.56-66
2. Abu Sharshouh, A. (2025). The Use of Artificial İntelligence in Accounting and
Auditing TT - Muhasebe ve Denetim Alanında Yapay Zekâ Kullanımı. Karadeniz
Ekonomi Araştırmaları Dergisi, 6(1).
3. Arslantaş, S. (2024). Artificial intelligence and big data from digital health
applications: publication trends and analysis. Journal of Health Organization and
Management. https://doi.org/10.1108/JHOM-06-2024-0241
4. Axiotis, E., Kontogiannis, A., Kalpoutzakis, E., & Giannakopoulos, G. (2021). A
personalized machine-learning-enabled method for efficient research in
ethnopharmacology. The case of the southern balkans and the coastal zone of asia
minor. Applied Sciences (Switzerland), 11(13). https://doi.org/10.3390/app11135826
5. Benedeti Rosa, M., & Kubota, L. C. (2025). Artificial intelligence: where does Brazil
stand in global scientific production and what are the main technical determinants
of adoption by Brazilian companies. EconomiA. https://doi.org/10.1108/ECON-01-
2025-0010
6. Bhardwaj, M., Kumar, P., & Singh, A. (2025). Bibliometric review of digital
transformation in agriculture: innovations, trends and sustainable futures. Journal of
Agribusiness in Developing and Emerging Economies. https://doi.org/10.1108/JADEE-
02-2025-0083
7. Budiman, A. S., Putri, D. N. N., Candra, H., Djuana, E., Sari, T. K., Aji, D. P. B., Putri,
L. R., Sitepu, E., Speaks, D., & Pasang, T. (2024). Crack Catcher AI – Enabling smart
fracture mechanics approaches for damage control of thin silicon cells or wafers.
Solar Energy Materials and Solar Cells, 273.
https://doi.org/10.1016/j.solmat.2024.112927
29
8. Cheng, J. C. P., Chen, W., Chen, K., Wang, Q., Sadri, H., Karki, B. R., Basnet, S., Xiang,
J., Montoya, J., Porras, J., Ucar, A., Karakose, M., Kırımça, N., Abdelalim, A. M.,
Essawy, A., Alnaser, A. A., Shibeika, A., Sherif, A., van Roosmale, S., … Liu, G.
(2024). applied sciences Artificial Intelligence for Predictive Maintenance
Applications: Journal of Building Engineering, 12(1).
9. Colina Vargas, A. M., Espinoza Mina, M. A., Catalán, L. L., & Catalán, B. L. (2024).
Computational intelligence, educational robotics, and artificial intelligence in the
educational field. A bibliometric study and thematic modelling. International Journal
of Educational Research and Innovation, 2024(22). https://doi.org/10.46661/ijeri.10369
10. Cricchio, J. (2025). Balancing openness and ownership: open innovation strategies
for AI development. European Journal of Innovation Management.
https://doi.org/10.1108/EJIM-04-2024-0470
11. Demidenko, S. Y., & Romanovskiy, N. v. (2023). THEORETICAL
UNDERSTANDING OF SOCIAL REALITY: THE 25th ANNIVERSARY OF
KHARCHEV READINGS. Sotsiologicheskie Issledovaniya, 2023(9).
https://doi.org/10.31857/S013216250027774-4
12. Educause. (2018). Horizon report: 2018 higher education edition. Retrieved from
EDUCAUSE Learning Initiative and The New Media Consortium. Liila.
13. Evolution Analysis and Prediction of Scientific Topic Popularity in the Field of LIS.
(2021). Documentation, Information and Knowledge, 38(2).
https://doi.org/10.13366/j.dik.2021.02.035
14. Gong, J., Zhao, Z., Niu, X., Ji, Y., Sun, H., Shen, Y., Chen, B., & Wu, B. (2025). AI
reshaping life sciences: intelligent transformation, application challenges, and future
convergence in neuroscience, biology, and medicine. In Frontiers in Digital Health
(Vol. 7). https://doi.org/10.3389/fdgth.2025.1666415
15. Govindan, K., Cheng, T. C. E., Mishra, N., Shukla, N., Pei-Chen, L., Arya, V., Sharma,
P., Singh, A., de Silva, P. T. M., Chatfield, A. T., Reddick, C. G., Karlsson, P.-M., Lal,
S., Rexhausen, D., Ajak, A. D., Lilford, E., Topal, E., Zhang, Y. Y., Fu, Z., … Perez-
Franco, R. (2018). Industry Surveys IT Consulting & Other Services. Journal of Cleaner
30
Production, 25(1).
16. Guo, Z. (2024). Innovative Functionality of the Scientific and Technical Information
System of China. Visnyk of Kharkiv State Academy of Culture, 65.
https://doi.org/10.31516/2410-5333.065.04
17. Harahulia, S., & Symonenko, T. (2025). RESEARCH SUPPORT SERVICES IN
DIGITAL HUMANITIES:APPROACHES, TECHNOLOGIES, TOOLS. Manuscript
and Book Heritage of Ukraine, 2025(2). https://doi.org/10.15407/rksu.37.189
18. Irtyshcheva, I., Kim, V., & Kletsov, Y. (2022). MANAGING BUSINESS PROCESSES
IN THE CONDITIONS OF THE DEVELOPMENT OF THE DIGITAL ECONOMY:
GLOBAL AND NATIONAL EXPERIENCES. Baltic Journal of Economic Studies, 8(5).
https://doi.org/10.30525/2256-0742/2022-8-5-101-107
19. Iurasova, O., Ivashko, L., Maksymova, J., & Maksymov, O. (2025). INNOVATION
TRENDS: IMPACT ON DEVELOPMENT OF FINANCIAL SECTOR. Смарт-
Економіка, Підприємництво Та Безпека, 3(1). https://doi.org/10.60022/sis.3.(01).1
20. Jeon, Y. (2021). A Study on the Ethics of the Concept and Purpose of Culture
Technology in the Post-COVID Era. Society for International Cultural Institute, 14(1).
https://doi.org/10.34223/jic.2021.14.1.1
21. Jirandehi, A. B., & Sajadi, S. M. (2024). Co-Citation Analysis and Co-Occurrence
Analysis of Resilience Keywords in StartUps. Scientometrics Research Journal, 10(2).
https://doi.org/10.22070/rsci.2024.17648.1667
22. Khabsa, M., Elmagarmid, A., Ilyas, I., Hammady, H., Ouzzani, M., Kouznetsov, A.,
Japkowicz, N., Visser, E., Felizardo, K. K. R., Souza, S. R. S., Maldonado, J. C. J. C. J.
C. J. C., Salleh, N., Martins, R. M., Mendes, E. E., MacDonell, S. G., Maldonado, J. C.
J. C. J. C. J. C., Li, D., Wang, Z. Z., Wang, L., … Demner-Fushman, D. (2020). A New
Method to Enhance. Systematic Reviews, 9(1).
23. Khmelkov, A. (2023). INSTITUTIONAL DESIGN OF ANALYTICAL TOOLS OF
THE FINANCIAL CONTROL INSTITUTION. Social Economics, 66.
https://doi.org/10.26565/2524-2547-2023-66-08
24. Kim, T. Y., Gang, J. Y., Oh, H. J., Ahmad, K., JianMing, Z., Rafi, M., Allen, G. G.,
31
Travis, T. A., Ramirez, C., , . . . . . ., Kyrillidou, M., Manifesto, L. S.,
Johnson, L., Smith, R., Willis, H., Levine, A., Haywood, K., Vice-chancellor, S. D.,
Vice-chancellor, D. V.-S. D., … Briggs, T. (2019). Using Data Visualization and
Dashboards to Support the Decision Making of Public Library Branch Managers.
Library Hi Tech, 38(1).
25. Lausiri, N., & Nahda, K. (2022). Pengaruh Rasio Likuiditas, Leverage Dan
Profitabilitas Terhadap Kondisi Financial Distress Dengan Kepemilikan Manajerial
Sebagai Variabel Moderasi. Selekta Manajemen: Jurnal Mahasiswa & Manajemen, 1(5).
26. Litynska, V. (2025). MODERN METHODS OF CONDUCTING MARKETING
RESEARCH. Social Economics, 70. https://doi.org/10.26565/2524-2547-2025-70-06
27. Luo, J., Zhuo, W., Liu, S., & Xu, B. (2024). The Optimization of Carbon Emission
Prediction in Low Carbon Energy Economy under Big Data. IEEE Access, 12.
https://doi.org/10.1109/ACCESS.2024.3351468
28. Mekhovych, K. (2025). TRANSFORMATION OF THE THEORY OF INNOVATIVE
MARKETING EFFICIENCY IN THE CONDITIONS OF AN UNDETERMINED
FUTURE. Energy Saving. Power Engineering. Energy Audit., 1 (204).
https://doi.org/10.20998/2313-8890.2025.01.07
29. Mi, S., Dong, C. Q., & Liu, Y. (2025). Data insights and talent portraits of science and
technology talents: Principles and applications. Studies in Science of Science, 43(6).
30. Morozova, I. A., Smetanin, A. S., & Smetanina, A. I. (2024). Management of Business
Competitiveness Based on Artificial Intelligence and Big Data for Its Sustainable
Development. Journal of Modern Competition, 18(1). https://doi.org/10.37791/2687-
0657-2024-18-1-29-40
31. Mosleh, G., Hemmati, S., Mohagheghzadeh, A., & Zarshenas, M. M. (2023).
Integrating Systems Biology and Artificial Intelligence in Traditional Pharmacy
Research: Advancements, Challenges, and Opportunities. Trends in Pharmaceutical
Sciences, 9(3).
32. Mossalanejad, A. (2025). Artificial Intelligence Policymaking and the Power
Equation Priorities. Geopolitics Quarterly, 21(3).
32
https://doi.org/10.22034/igq.2025.511096.2023
33. Nazari, M., Ansari, M., & Pour, M. J. A. (2025). Bibliographic Analysis and Future
Research Directions in the Field of Sustainable and Innovative Business Models.
Scientometrics Research Journal, 11(2). https://doi.org/10.22070/rsci.2024.19650.1759
34. Nedoshytko, I., & Patriak, O. (2022). Electronic Document Management and Its
Value for Business. Digital Platform: Information Technologies in Sociocultural Sphere,
5(2). https://doi.org/10.31866/2617-796x.5.2.2022.270142
35. Nyomba, Muh. A. (2021). Faktor Yang Berhubungan Dengan Kejadian Infeksi
Saluran Pernapasan Akut (ISPA) Pada Balita di Sekitar Wilayah Tempat
Pembuangan Akhir (TPA) Sampah Antang Kota Makassar Tahun 2021. Factors
Associated of The Incidence of Acute Respiratory Infection (ARI) in. Universitas
Hasanuddin., 9.
36. Pan, C., Bowen, W., Peng, Z., & Shao, L. (2024). Biological network-based exploration
of the principles of traditional Chinese medicine. Kexue Tongbao/Chinese Science
Bulletin, 69(1). https://doi.org/10.1360/TB-2023-0730
37. Perano, M., Cammarano, A., Varriale, V., del Regno, C., Michelino, F., & Caputo, M.
(2023). Embracing supply chain digitalization and unphysicalization to enhance
supply chain performance: a conceptual framework. International Journal of Physical
Distribution and Logistics Management, 53(5–6). https://doi.org/10.1108/IJPDLM-06-
2022-0201
38. Pysmennyi, V., & Rudan, V. (2024). Methods of financial management in local
communities under the conditions of the russian-Ukrainian war. Economic Analysis,
34(3). https://doi.org/10.35774/econa2024.03.351
39. RAHMAN, G. O. H. S. E. Y. T. A. (2020). DETERMINANTS OF FOOD DELIVERY
PLATFORMS THROUGH CUSTOMER LOYALTY TOWARD SUSTAINABILITY
OF BUSINESS. Human Relations, 3(1).
40. Rejter, A. (2024). STUDIES IN ONOMASTICS AND THE NEW HUMANITIES.
Onomastica, 68. https://doi.org/10.17651/ONOMAST.68.2
41. Shakil, M., Ali, M., Illahi, T., & Ahmed, F. (2025). FinTech 5.0 and the Future of Global
33
Finance: Harnessing Artificial Intelligence, Blockchain, and Big Data to Build
Inclusive, Sustainable, and Resilient Financial Ecosystems. Inverge Journal of Social
Sciences, 4(4). https://doi.org/10.63544/ijss.v4i4.179
42. Sipa, M. (2023). Trends and challenges in talent management – mind map. Scientific
Papers of Silesian University of Technology Organization and Management Series,
2023(164). https://doi.org/10.29119/1641-3466.2022.164.34
43. Skoczkowski, T., Bielecki, S., Wołowicz, M., & Węglarz, A. (2025). Redefining Energy
Management for Carbon-Neutral Supply Chains in Energy-Intensive Industries: An
EU Perspective. Energies, 18(15). https://doi.org/10.3390/en18153932
44. Skrypnyk, S. v. (2025). Key Concepts and Models of Strategic Innovation
Management in Ukraine: Modern Approaches. The Problems of Economy, 2(64).
https://doi.org/10.32983/2222-0712-2025-2-77-85
45. Soni, A., Jain, A., & Kaur, B. (2022). A bibliometric analysis of effect of artificial
intelligence on customer satisfaction in banking sector. International Research Journal
of Modernization in Engineering, 4(3).
46. Sydorovych, O., Bukovskyi, M., & Yaroshovych, A. (2024). DIGITAL TOOLS OF
TAX ADMINISTRATION: CURRENT TRENDS AND PERSPECTIVES.
INNOVATIVE ECONOMY. https://doi.org/10.37332/2309-1533.2024.4.26
47. Torres-Sainz, R., de-Zayas-Pérez, M. R., Trinchet-Varela, C. A., Pérez-Vallejo, L. M.,
& Rodríguez, R. P. (2025). INTELLIGENT PREDICTIVE MAINTENANCE: A
BIBLIOMETRIC APPROACH THROUGH THE SCOPUS DATABASE. Bibliotecas,
Anales de Investigacion, 21(1).
48. Vollmar, H. C., Kramer, U., Müller, H., Griemmert, M., Noelle, G., & Schrappe, M.
(2017). Digitale Gesundheitsanwendungen - Rahmenbedingungen zur Nutzung in
Versorgung, Strukturentwicklung und Wissenschaft - Positionspapier der AG
Digital Health des DNVF. Gesundheitswesen, 79(12). https://doi.org/10.1055/s-0043-
122233
49. VOLOSHYN, R. (2024). COMMODITY EXPERTISE IN E-COMMERCE.
“EСONOMY. FINANСES. MANAGEMENT: Topical Issues of Science and Practical
34
Activity,” 4 (70). https://doi.org/10.37128/2411-4413-2024-4-4
50. Wang, Y., Kang, M., Liu, Y., Li, J., Xue, K., Wang, X., Du, J., Tian, Y., Ni, Q., & Wang,
F. Y. (2023). Can Digital Intelligence and Cyber-Physical-Social Systems Achieve
Global Food Security and Sustainability? IEEE/CAA Journal of Automatica Sinica,
10(11). https://doi.org/10.1109/JAS.2023.123951
51. Wijanarti, H. L., & Anisyah, T. D. A. (2022). Hubungan Antara Kualitas Tidur, Beban
Kerja Fisik Terhadap Perasaan Kelelahan Kerja Pada Perawat Rawat Inap Kelas 3 Di
Rs Pku Muhammadiyah Gamping. Jurnal Lentera Kesehatan Masyarakat, 1(1).
52. Yubo, S., Ramayah, T., & Hongmei, L. (2022). Research on the Impact of Intelligent
Manufacturing Technology on Business Model Based on CNKI’s Evidence. Global
Business and Management Research, Suppl. Special Issue, 14(3s).
53. Zavalii, T., & Lehenchuk, S. (2025). DEVELOPMENT OF RESEARCH METHODS IN
MARKETING IN THE CONDITIONS OF FORMING THE DIGITAL MARKETING
PARADIGM. Economic Scope, 204. https://doi.org/10.30838/ep.204.101-109
54. Zhao, R., Yang, X., Zhang, D., Li, J., Huang, Y., Xian, G., Kou, Y., & Sun, T. (2025).
Artificial Intelligence for Agricultural Science (AI4AS): Key Elements, Challenges
and Pathways. Smart Agriculture, 7(3).
https://doi.org/10.12133/j.smartag.SA202502019
55. Zhao, X., Wang, X., Research, R. C.-A. in R., & 2025, undefined. (2025). Research on
optimizing water resource management and improving crop yield using big data
and artificial intelligence. Jstage.Jst.Go.Jp, 5(2).
56. Zharov, A. M. (2024). THE DYNAMICS OF KNOWLEDGE:AN EVOLUTIONARY
ANALYSIS OF SCIENTIFIC DEVELOPMENT THEORIES. Vestnik Tomskogo
Gosudarstvennogo Universiteta. Filosofiya, Sotsiologiya, Politologiya, 80.
https://doi.org/10.17223/1998863x/80/3
57. Zheldibayeva, R. (2025). THE IMPACT OF AI AND PEER FEEDBACK ON
RESEARCH WRITING SKILLS: A STUDY USING THE CGSCHOLAR PLATFORM
AMONG KAZAKHSTANI SCHOLARS. Scientific Journal of Astana IT University.
https://doi.org/10.37943/21tbax7694
35
58. Zozulov, O., & Boboshko, Y. (2025). Classification of product development and
market entry models. Marketing and Digital Technologies, 9(2).
https://doi.org/10.15276/mdt.9.2.2025.4
59. Zuo, H. (2025). Human-Environment Relationship Research from a
Multidisciplinary Integration Perspective in Geography. Frontiers in Sustainable
Development, 5(8). https://doi.org/10.54691/64t83m16
60. Игнатьев, М. Е., & Степанов, В. В. (2025). Методология диссертационных
исследований в юридических науках: отдельные аспекты и проблемы.
СОВРЕМЕННОЕ ПРАВО, 7. https://doi.org/10.25799/ni.2025.32.84.001
61. Ксения Юрьевна, Б., & Елена Иосифовна, Л. (2024). Identification and regulation
of the macro-regional ecosystem’ human resources potential development trends:
global and national experience. STATE AND MUNICIPAL MANAGEMENT
SCHOLAR NOTES, 1(4). https://doi.org/10.22394/2079-1690-2024-1-4-28-33
62. Muhtar, E. (2022). PENGARUH PENDEKATAN KETRAMPILAN PROSES
DALAM MENINGKATKAN HASIL BELAJAR FISIKA PADA KONSEP
GETARAN DAN GELOMBANG PADA PESERTA DIDIK KELAS VIII SMP
NEGERI 17 HALMAHERA SELATAN. Jurnal Pembelajaran Dan Sina Fisika,
7(1).
63. Permenkes. (2022). Peraturan Menteri Kesehatan Nomor 24 Tahun 2022
Tentang Rekam Medis. Jakarta: Menteri Kesehatan Republik Indonesia.
Science, 7(1).
64. Putri, S. H. (2022). Pengaruh Brand Image terhadap keputusan pembelian
produk skincare. Science, 7(1).
65. Candaika, P. (2022). Gambaran Kesiapan Rumah Sakit Umum Daerah
(RSUD) Menghadapi Kebijakan Kelas Rawat Inap Standar (KRIS) Tahun
2022. Science, 7(1).
36
Chapter 2.
The Epistemology of Algorithms: A
Comprehensive Framework for Scientific
Methodology in Artificial Intelligence and
Data Science
1. The Fourth Paradigm and the Crisis of
Reproducibility
The emergence of Artificial Intelligence (AI) and Data Science (DS) has catalyzed a
fundamental transformation in the structure of scientific inquiry, shifting the locus of
discovery from the traditional hypothetico-deductive model toward a data-intensive,
computational paradigm often referred to as the "Fourth Paradigm" of science.1 This shift is
not merely instrumental—replacing analog tools with digital ones—but epistemological,
altering the very nature of how knowledge is generated, validated, and interpreted. Where
the industrial revolution mechanized physical labor through heat engines, the information
revolution is mechanizing cognitive labor through "data engines" capable of generating
actionable knowledge from vast, unstructured datasets.1 However, this rapid
mechanization has outpaced the development of rigorous methodological standards,
leading to a landscape where innovation frequently precedes validation.
The pathway from algorithmic innovation to reliable scientific deployment is non-
linear and fraught with complexity. Unlike classical software engineering, where logic is
deterministic and explicitly coded, modern AI systems—particularly those based on deep
learning—operate as probabilistic "black boxes" where internal decision-making processes
are opaque and emergent.2 This opacity presents a profound challenge to the scientific
method, which relies on transparency, reproducibility, and the falsifiability of hypotheses.
37
The integration of AI into high-stakes domains such as healthcare, climate science, and
criminal justice demands a transition from ad-hoc experimentation to a rigorous "Diffusion
Engine" of methodologies.1 This engine must bridge the gap between abstract computer
science and domain-specific application, ensuring that AI models are not just predictive, but
robust, fair, and scientifically valid.
Current literature suggests that the "reproducibility crisis" in AI is largely a
symptom of methodological immaturity. Issues such as "p-hacking," weak baselines, and
data leakage—long recognized in statistics—have resurfaced in machine learning research,
exacerbated by the stochastic nature of training algorithms and the proprietary nature of
large datasets.4 Furthermore, the lack of standardized reporting protocols has made it
difficult to distinguish between genuine algorithmic advances and performance gains
achieved through hyperparameter overfitting or random chance.
To address these challenges, this report synthesizes a unified methodological
framework for AI and Data Science. It draws upon diverse streams of research—from
software engineering lifecycles like CRISP-DM and TDSP to statistical rigor in hypothesis
testing and the emerging discipline of Scientific Machine Learning (SciML). By integrating
principles of data provenance, rigorous experimental design (including ablation and
sensitivity analysis), and ethical stewardship, this framework aims to establish a standard
for expert-level research that is exhaustive, reproducible, and deeply integrated with the
scientific method.
2. Structural Lifecycles: From Linear Process to
Circular Inquiry
The management of scientific research in AI requires a structured lifecycle that
accommodates the unique characteristics of data-driven projects: high uncertainty, iterative
experimentation, and the need for continuous validation. While software development has
settled on Agile methodologies, data science requires a hybrid approach that fuses
38
engineering discipline with scientific rigor.
2.1 The Evolution of Process Models: CRISP-DM and Beyond
The Cross-Industry Standard Process for Data Mining (CRISP-DM) remains the
foundational reference model for the field, widely adopted for its clarity and industry-
agnostic applicability.6 It delineates the research process into six phases: Business
Understanding, Data Understanding, Data Preparation, Modeling, Evaluation, and
Deployment.
However, the modern application of CRISP-DM acknowledges its limitations. The
original model's implicit linearity is often insufficient for the complexities of modern AI,
where "Data Preparation" and "Modeling" are deeply intertwined through feature
engineering and representation learning. Consequently, the methodology has evolved into
a circular approach.7 For instance, insights gained during the "Model Evaluation" phase
often force a return to "Problem Definition," creating a feedback loop that refines the
scientific question itself.8 This circularity is essential; a linear progression assumes that the
initial hypothesis and data quality are sufficient, which is rarely the case in high-
dimensional scientific problems.
2.2 The Team Data Science Process (TDSP) and Agile Integration
To address the collaborative and operational needs of modern research teams,
Microsoft introduced the Team Data Science Process (TDSP). TDSP modernizes CRISP-DM
by integrating it with Agile practices (Scrum, Kanban) and emphasizing the lifecycle of the
team rather than just the project.9
The TDSP framework is distinct in its focus on:
1. Standardized Project Structure: Enforcing a consistent directory structure and
document template system to reduce "technical debt" and facilitate knowledge transfer
between researchers.10
39
2. Version Control and Collaboration: Unlike solitary academic research of the past,
TDSP mandates the use of Git-based version control for both code and documentation,
treating the experimental process as a collaborative software engineering effort.9
3. Iterative Sprints: Research is broken down into fixed-length iterations (Sprints),
allowing for rapid feedback and the flexibility to pivot based on early experimental
results. This "fail fast" mentality is crucial in AI, where training large models can be
resource-intensive (see Table 3).11
Table 3: Comparative Analysis of Research Lifecycles
Feature
CRISP-DM
TDSP (Team Data
Science Process)
Agile/Kanban for
Data Science
Philosophical Basis
Industrial Data
Mining
Collaborative
Software Engineering
Lean Manufacturing /
Flow
Structure
6 Phases (Cyclical)
Role-Based, Iterative
Lifecycle
Continuous flow,
Minimize WIP
Key Artifacts
Phase Reports, Model
Git Repos, Charters,
Exit Reports
Backlogs, Kanban
Boards
Strengths
Comprehensive,
standard terminology,
focus on business
goals 6
Integration with
MLOps, strong team
focus, reproducibility
support 9
High flexibility,
maximizes
throughput, adapts to
uncertainty 11
Limitations
Can become a
"waterfall" trap; lacks
specific team roles 11
Higher overhead for
small, academic
teams; documentation
heavy
Can lack strategic
coherence without a
broader framework
2.3 The Scientific AI Lifecycle
In a purely research context—distinct from industrial deployment—the lifecycle
40
focuses heavily on validity and discovery. The "Unified View" of AI research methodology
incorporates a long-term dimension based on the scientific method, identifying strategies
such as hypothetical/deductive and hermeneutical/inductive analysis.12
The critical distinction in the scientific lifecycle is the Problem Formulation phase.
Here, abstract scientific inquiries must be translated into concrete data science problems.7
This involves defining the problem in terms of symmetry groups (e.g., supervised vs.
unsupervised learning) and identifying the mathematical metrics that will serve as proxies
for scientific truth.13 For example, in clinical AI, the problem formulation must explicitly
define the "beneficiary" and the "end-user" to preemptively address ethical downstream
effects.14 This phase sets the trajectory for the entire project; a poorly framed problem will
lead to technically accurate but scientifically irrelevant models.
3. The Science of Data: Curation, Provenance, and
Ethics
In the classical scientific method, "data" was often the result of controlled
experiments designed to minimize noise. In the AI paradigm, data is often "found"
(observational), unstructured, and noisy. Therefore, the methodology of data curation has
become a scientific discipline in its own right, critical for ensuring the validity of
downstream inference.
3.1 Data Provenance and Lineage
Data provenance—the documentation of the origin, history, and transformations of
data—is the bedrock of reproducibility in AI.15 Without a verifiable lineage, an AI model is
scientifically essentially worthless, as its predictions cannot be audited for artifacts or errors
introduced during preprocessing.
Best practices for provenance in research include:
41
● Metadata Schemas: Utilizing standardized schemas to record provenance details
(creator, date, source, license) consistently across all datasets.16
● Immutable Logging: Implementing systems that create a permanent, unalterable
record of every transformation applied to the raw data. This allows researchers to
"replay" the data processing pipeline to verify results.16
● Hash Verification: Using cryptographic hashes to verify the integrity of data files,
ensuring that the dataset analyzed today is bit-for-bit identical to the one analyzed
yesterday.16
The necessity of this rigor was highlighted by a recent audit of AI training datasets,
which found a systemic failure in provenance tracking: over 70% of datasets examined
lacked license information, and 50% had miscategorized licenses.17 Such negligence not only
creates legal liability but undermines scientific trust. Tools like the Data Provenance
Explorer have emerged to help researchers trace the lineage of fine-tuning datasets,
ensuring that the "supply chain" of ideas remains transparent and robust.17
3.2 Datasheets for Datasets
To standardize the communication of data characteristics, the AI community is
increasingly adopting the "Datasheets for Datasets" framework.18 Modeled after the
datasheets used in the electronics industry (which describe the operating limits of a resistor
or capacitor), this document provides a structured summary of a dataset's motivation,
composition, collection process, and recommended uses.
The scientific value of a Datasheet lies in its ability to force the researcher to articulate
context.
● Motivation: Why was this data collected? (e.g., specific gap in literature vs. general
purpose).
● Composition: Does the dataset contain subpopulations? Are there missing data
patterns?
42
● Collection: How was the data acquired? (e.g., API, scraping, sensors).
● Uses: What are the tasks for which the dataset is valid? Crucially, what are the tasks
for which it should not be used?.18
In the Earth Sciences, this methodology has been adapted to document geospatial
biases and technical limitations, proving that the framework is transferable across scientific
domains.20 Empirical studies show that the act of completing a Datasheet increases the
"ethical sensitivity" of machine learning engineers, helping them recognize potential biases
that might otherwise go unnoticed.21
3.3 Algorithmic Fairness and Bias Metrics
Rigorous methodology demands that "performance" be defined beyond simple
accuracy. In many scientific and social applications, a model that achieves high accuracy by
exploiting biases in the training data is considered a failure. Therefore, bias assessment is a
mandatory component of the evaluation phase.
Research identifies three primary metrics for assessing fairness, each representing a
different philosophical definition of equity:
1. Demographic Parity (Statistical Parity): This metric requires that the probability of a
positive outcome be independent of the protected attribute (e.g., gender or race).22
P(\hat{Y}=1 | A=0) = P(\hat{Y}=1 | A=1). While intuitively appealing, this metric can
be problematic if the base rates of the target variable differ genuinely between groups.
2. Equality of Opportunity: This focuses on the True Positive Rate (TPR). It requires that
qualified individuals in both groups have an equal chance of being selected.22
P(\hat{Y}=1 | Y=1, A=0) = P(\hat{Y}=1 | Y=1, A=1). This is often the preferred metric in
medical diagnostics (e.g., ensuring a cancer detection model works equally well for all
races).
3. Disparate Impact: A ratio-based metric mandated by US employment law. If the ratio
of the selection rate of the protected group to the unprotected group is less than 0.8 (the
43
"four-fifths rule"), the model is considered biased.22
The "Objective Fairness Index" (OFI) has recently been proposed to provide a legally
grounded and context-aware perspective on these metrics, addressing gaps where
traditional disparate impact calculations fail.24 Integrating these metrics into the standard
evaluation loop ensures that the "scientific discovery" is not merely an artifact of systemic
inequality.
4. Rigorous Experimental Design in Machine Learning
An ML experiment is a controlled procedure designed to falsify a hypothesis
regarding the relationship between data features and target variables.25 However, the
stochastic nature of training (random initialization, data shuffling) and the complexity of
hyperparameters make isolation of variables difficult. A robust experimental design is the
only defense against "alchemy"—the trial-and-error tweaking of models until they appear
to work.
4.1 The Importance of Strong Baselines
A pervasive issue in AI literature is the use of "weak baselines" to inflate the
perceived novelty of a proposed method. A rigorous study must compare the new model
not just against other state-of-the-art complex models, but against well-tuned simple
models.26
Guidelines for Baseline Construction:
● Classical Baselines: Always test simple, interpretable models (Linear/Logistic
Regression, Random Forests) first. These establish the "floor" of performance. If a deep
neural network barely outperforms a logistic regression, the complexity is likely
unjustified.26
● Hyperparameter Tuning: Baselines must be tuned with the same rigor as the
experimental model. A comparison between a hyper-tuned neural net and a default-
44
parameter Random Forest is scientifically invalid.
● Preprocessing consistency: Both the baseline and the new model must consume the
exact same data splits and preprocessing steps to prevent data leakage.26
4.2 Cross-Validation Methodologies
Estimating the generalization error—how the model performs on unseen data—is
the central task of evaluation. The standard "train/test split" is often insufficient due to the
high variance in result estimates.
● k-Fold Cross-Validation: The dataset is partitioned into k disjoint subsets. The process
involves k iterations; in each, a different subset is held out for testing while the
remaining k-1 are used for training. The final performance metric is the average of the
k scores.27 This reduces variance and ensures every data point is used for testing exactly
once.
● Stratified k-Fold: For classification problems with imbalanced classes, standard k-fold
can result in folds with no positive examples. Stratification ensures that the class
distribution in every fold preserves the distribution of the whole dataset.28
● Time Series Split (Rolling Window): In temporal data (e.g., stock prices, climate data),
standard cross-validation introduces "look-ahead bias" (training on future data to
predict the past). The correct methodology is a rolling window where the training set
consists of indices [0, t] and the test set consists of [t+1, t+k].27
4.3 Hyperparameter Optimization and Ablation
The search for optimal hyperparameters (learning rate, layer depth) is part of the
experimental design. However, distinguishing between performance gains from
architecture vs. hyperparameters is critical.
Ablation Studies:
An ablation study is the systematic removal of components of a machine learning
system to measure their marginal contribution to performance.29 It is the AI equivalent of
45
a "gene knockout" experiment.
● Methodology: Deconstruct the model into its additive components (e.g., "Architecture
A + Feature B + Regularization C"). Train and evaluate variants where one component
is removed or replaced with a counterfactual (zeroed out or randomized).30
● Interpretation: If removing a complex attention mechanism result in a <1% drop in
accuracy, the mechanism is likely redundant or the model is learning via a
"Compensatory Masquerade" (where other parts of the network compensate for the
missing signal).31
● Efficiency: While full grid search for ablations is ideal, it is computationally expensive.
Researchers effectively use "Leave-One-Component-Out" strategies or automated tools
like MAGGY to parallelize ablation trials.29
5. Statistical Significance and Model Comparison
In many published papers, a model is declared "superior" if its accuracy is 0.1%
higher than the baseline. Without statistical testing, such claims are scientifically vacuous.
The field has moved toward specific non-parametric tests to validate these comparisons.
5.1 The Failure of the Paired t-test
Traditionally, the paired Student’s t-test was used to compare the means of two
models' performance. However, this test assumes that the differences in performance are
normally distributed and, crucially, that the samples are independent. In k-fold cross-
validation, the training sets overlap significantly (sharing k-2 folds worth of data), violating
the independence assumption. This leads to an elevated Type I error rate (false positives).5
5.2 Recommended Statistical Tests
To address the shortcomings of the t-test, the AI community has converged on more
robust alternatives:
1. McNemar’s Test:
46
This is the standard for comparing two classifiers on a single dataset. It operates
on the contingency table of predictions:
○ N_{01}: Number of examples misclassified by Model A but correctly classified by
Model B.
○ N_{10}: Number of examples correctly classified by Model A but misclassified by
Model B.
The test statistic is approximated by \chi^2 = \frac{(|N_{01} - N_{10}| -
1)^2}{N_{01} + N_{10}}. It tests if the models make errors on the same examples. It
is computationally efficient as it requires running the models only once.5
2. 5x2 Cross-Validation Paired t-test:
Proposed by Dietterich, this method performs 5 iterations of 2-fold cross-
validation. It is designed to balance the trade-off between power and Type I error,
specifically accounting for the variation arising from the choice of training sets. It
is considered slightly more powerful than McNemar’s test but requires 10 training
runs.33
3. Wilcoxon Signed-Rank Test:
This is the recommended non-parametric test for comparing two classifiers across
multiple datasets (e.g., a benchmark study on 20 different domains). It ranks the
differences in performance and checks if the distribution of differences is
symmetric around zero. It is robust to outliers and does not assume normality (see
Table 4).27
Table 4: Selection Matrix for Statistical Tests in AI
47
6. Sensitivity Analysis and Interpretability
As models grow in complexity, "interpretability" becomes a requirement for
scientific validity. Sensitivity Analysis (SA) provides the methodological toolkit to peer
inside the black box by quantifying how changes in inputs affect outputs.
6.1 Input Perturbation and Feature Importance
The simplest form of SA involves perturbing input variables (e.g., adding Gaussian
noise, masking pixels, or shifting values) and observing the degradation in model output.
● Local Analysis: Explains a specific prediction. For example, in an image classification
model, masking a specific region of the image to see if the classification changes
identifies that region as the "cause" of the prediction.35
● Global Analysis: Averages the sensitivity across the entire dataset to rank features by
overall importance. This helps in feature selection and model simplification.36
6.2 Advanced Jacobian-Based Analysis
For deep neural networks, sensitivity is mathematically formalized using the
Jacobian matrix of the function. The norm of the input-output Jacobian ||\nabla_x f(x)||_F
measures the local sensitivity of the network.
● Generalization Insight: Research has shown a strong correlation between the Jacobian
norm and generalization error. Models that are robust to small input perturbations
(low sensitivity) tend to generalize better to unseen data.2
● Robustness Training: This insight has led to training methodologies where the
48
Jacobian norm is penalized (regularized) during training, forcing the model to learn
smoother decision boundaries that are less brittle.2
7. Scientific Machine Learning (SciML): Bridging Data
and Physics
Scientific Machine Learning (SciML) represents the frontier where data-driven
methodology meets mechanistic modeling. Unlike pure AI, which learns patterns solely
from data, SciML incorporates domain knowledge (physics, biology, chemistry) directly
into the learning process. This creates a "gray box" model that combines the flexibility of
neural networks with the interpretability of differential equations.37
7.1 Physics-Informed Neural Networks (PINNs)
The defining methodology of SciML is the Physics-Informed Neural Network
(PINN). In a standard neural network, the optimization objective is to minimize the
difference between predictions and data labels (Loss_{data}). In a PINN, the loss function is
augmented with a "residual" term derived from the governing physical laws (e.g., Navier-
Stokes equations for fluid dynamics).
Loss_{Total} = Loss_{data} + \lambda \cdot Loss_{Physics}
By minimizing this composite loss, the network is constrained to find solutions that
not only fit the observed data points but also satisfy the underlying differential equations in
the spaces between the data points.37 This regularization allows PINNs to generalize well
even in "small data" regimes where pure deep learning would fail due to overfitting.38
7.2 Operator Learning and Model Discovery
SciML also expands the scope of learning from functions to operators.
● Neural Operators (e.g., DeepONet, FNO): These architectures learn the mapping
between infinite-dimensional function spaces. For example, instead of solving a heat
49
equation for one specific initial condition, a Neural Operator learns the solution operator
that maps any initial temperature profile to its future state. This allows for real-time
simulation of complex physical systems at a fraction of the cost of traditional numerical
solvers.37
● Automated Model Discovery (SINDy): This methodology uses sparse regression to
discover the governing equations from data. It takes time-series data and finds the
sparsest combination of mathematical terms (e.g., x, x^2, \sin(x)) that describe the
dynamics. The output is not a black-box model, but a symbolic equation (e.g.,
\frac{dx}{dt} = \sigma(y-x)) that scientists can interpret and verify.39
8. Reproducibility, MLOps, and the Crisis of Trust
The ultimate test of a scientific methodology is reproducibility. In AI, this is
challenging due to the complex stack of software and hardware dependencies. The field has
moved toward "MLOps" (Machine Learning Operations) to provide the infrastructure for
reproducible science.
8.1 The Hierarchy of Reproducibility
Methodologists distinguish between three levels of validity 40:
1. Repeatability: The same team, using the same code and hardware, obtains the same
result. This is a check of the stability of the code.
2. Reproducibility (Dependent): A different team, using the original artifacts (code, data),
obtains the same result. This verifies that the result is not an artifact of a specific local
environment.
3. Replicability (Independent): A different team reimplements the algorithm from the
paper's description (without seeing the original code) and obtains a consistent result.
This is the highest standard, verifying the scientific truth of the method rather than just
the code correctness.
50
8.2 The Technological Stack for Reproducibility
Achieving reproducibility requires specific tools to control entropy in the
computational environment:
● Containerization (Docker): Docker packages the operating system, libraries, and
dependencies into an immutable image. This ensures that a model trained on a Linux
server in 2024 can be re-run on a Windows laptop in 2026 with the exact same
environment.42
● Data Versioning (DVC): Standard Git is poor at handling large binary files. Tools like
DVC (Data Version Control) allow researchers to version control datasets alongside
code. A specific Git commit can be linked to a specific "snapshot" of the data, ensuring
that if the data changes, the provenance of the model is not lost.44
● Determinism and Seeds: Deep learning libraries (PyTorch, TensorFlow) often default
to non-deterministic algorithms for speed (especially on GPUs). Rigorous methodology
requires setting global random seeds and forcing deterministic algorithms, even at the
cost of performance, to ensure that results are bit-wise identical across runs.42
8.3 The NeurIPS Checklist as a Standard
To enforce these standards, the NeurIPS conference (a premier venue for AI
research) has introduced a mandatory reproducibility checklist. Authors must explicitly
declare whether they have provided code, data, error bars, and details on computing
infrastructure.47
Recent experiments using Large Language Models (LLMs) to audit these checklists
have shown that AI tools themselves can help enforce methodological rigor. An "Author
Checklist Assistant" powered by GPT-4 was used to vet papers, providing feedback to
authors on whether their claims of reproducibility were substantiated by the provided
artifacts.49
51
9. Future Horizons: AI-Driven Hypothesis Generation
As the methodology matures, AI is transitioning from a tool for testing hypotheses
to a system for generating them. This closes the loop of the scientific method, automating the
discovery process itself.
9.1 The HypoGeniC Framework
The "HypoGeniC" framework demonstrates this new capability. It uses an iterative
"agentic" workflow:
1. Initialization: An LLM reviews vast literature and initial data to propose a set of
scientific hypotheses.
2. Refinement: A separate AI agent tests these hypotheses against challenging data
examples ("counter-examples").
3. Update: Hypotheses that fail are discarded; those that survive are refined and made
more specific. This process mimics the peer-review and revision cycle of human science
but operates at a speed and scale impossible for human researchers.51
9.2 Case Study: AlphaFold and the Protein Universe
DeepMind's AlphaFold serves as the exemplar of this new scientific methodology.
Its architecture is built on an iterative "recycling" mechanism where the model generates a
structural hypothesis, assesses it, and feeds it back as input for refinement.52 This loop
effectively performs in silico experiments.
The impact is a paradigm shift in biology: researchers can now bypass the years-long
process of X-ray crystallography for determining protein structures. Instead, they use the
AI prediction as a high-confidence hypothesis to guide downstream experiments in drug
discovery and molecular dynamics.53 In this regime, the AI methodology becomes the
scientific method.
The application of scientific methodology to AI and data science is no longer
optional; it is the prerequisite for progress. The field has graduated from the "wild west" of
52
ad-hoc scripts and leaderboard-chasing to a mature discipline governed by rigorous
frameworks.
The Unified Framework proposed in this report relies on three pillars:
1. Structural Rigor: Adopting circular lifecycles (TDSP, Scientific Lifecycle) that
emphasize feedback loops and team collaboration.
2. Experimental Rigor: Utilizing solid baselines, ablation studies, and appropriate
statistical tests (McNemar/Wilcoxon) to validate claims.
3. Ethical and Data Rigor: Treating data curation as a first-class scientific activity,
documenting provenance, and actively measuring bias.
For the modern researcher, adhering to these methodologies is not merely about
getting a paper accepted; it is about ensuring that the "Intelligence" we are building is
robust, transparent, and ultimately beneficial to the human enterprise. As we stand on the
brink of automated discovery, the rigor of our methods will determine the reliability of our
future knowledge.
53
Chapter bibliography
1. T. P., & . V. G. (2025). INVESTIGATION OF DEEP LEARNING FRAMEWORK TO
ACCELERATE AI FOR IMAGE PROCESSING APPLICATIONS. International
Journal For Multidisciplinary Research, 7(3).
https://doi.org/10.36948/ijfmr.2025.v07i03.45548
2. Abdul-Rahman, G., Zwitter, A., & Haleem, N. (2025). A systematic literature review
on the role of artificial intelligence in citizen science. Discover Artificial Intelligence,
5(1). https://doi.org/10.1007/s44163-025-00437-z
3. Acomi, N., Surugiu, G., Raicu, G., & Stanca, C. (2025). A Thematic Review of Port
Services and Emission Reduction Strategies Using ATLAS.ti. TransNav, 19(2).
https://doi.org/10.12716/1001.19.02.14
4. Aggarwal, K., Mijwil, M. M., Sonia, Al-Mistarehi, A. H., Alomari, S., Gök, M., Zein
Alaabdin, A. M., & Abdulrhman, S. H. (2022). Has the Future Started? The Current
Growth of Artificial Intelligence, Machine Learning, and Deep Learning. Iraqi Journal
for Computer Science and Mathematics, 3(1).
https://doi.org/10.52866/ijcsm.2022.01.01.013
5. Ahmadpour, M. (2024). Emerging Approaches in Enhancing Mental and Physical
Health of the Elderly: Integration of Strength Training, Mindfulness, and Artificial
Intelligence. Journal of Psychological Dynamics in Mood Disorders, 3(5).
https://doi.org/10.61838/kman.pdmd.3.5.19
6. Alshadidi, A. A. F., Alshahrani, A. A., Aldosari, L. I. N., Chaturvedi, S., Saini, R. S.,
Hassan, S. A. bin, Cicciù, M., & Minervini, G. (2023). Investigation on the Application
of Artificial Intelligence in Prosthodontics. In Applied Sciences (Switzerland) (Vol. 13,
Issue 8). https://doi.org/10.3390/app13085004
7. Asaduzzaman, M., & Flowra, M. T. (2018). Resurgence of infectious diseases due to
forced migration: is planetary health and One Health action synergistic? The Lancet
Planetary Health, 2(10).
54
8. Bock, T., & von der Oelsnitz, D. (2025). Leadership-competences in the era of
artificial intelligence – a structured review. In Strategy and Leadership (Vol. 53, Issue
3). https://doi.org/10.1108/SL-09-2024-0100
9. Boukrouh, I., & Azmani, A. (2024). ARTIFICIAL INTELLIGENCE APPLICATIONS
IN E-COMMERCE: A BIBLIOMETRIC STUDY FROM 1995 TO 2023 USING
MERGED DATA SOURCES. International Journal of Professional Business Review, 9(4).
https://doi.org/10.26668/businessreview/2024.v9i4.4537
10. Buchkremer, R., Demund, A., Ebener, S., Gampfer, F., Jagering, D., Jurgens, A.,
Klenke, S., Krimpmann, D., Schmank, J., Spiekermann, M., Wahlers, M., & Wiepke,
M. (2019). The Application of Artificial Intelligence Technologies as a Substitute for
Reading and to Support and Enhance the Authoring of Scientific Review Articles. In
IEEE Access (Vol. 7). https://doi.org/10.1109/ACCESS.2019.2917719
11. Cafuta, D., Dodig, I., Cesar, I., & Kramberger, T. (2021). Developing a modern
greenhouse scientific research facility — A case study. Sensors, 21(8).
https://doi.org/10.3390/s21082575
12. Chango, X., Flor-Unda, O., Gil-Jiménez, P., & Gómez-Moreno, H. (2024). Technology
in Forensic Sciences: Innovation and Precision. In Technologies (Vol. 12, Issue 8).
https://doi.org/10.3390/technologies12080120
13. Chen, S., & Garcia, A. (2017). Embodiment of Cognition: Exploring the Interplay
between Body and Mind. Studies in Science of Science | ISSN: 1003-2053, 35(9).
14. Espinel, R., Herrera-Franco, G., Rivadeneira García, J. L., & Escandón-Panchana, P.
(2024). Artificial Intelligence in Agricultural Mapping: A Review. In Agriculture
(Switzerland) (Vol. 14, Issue 7). https://doi.org/10.3390/agriculture14071071
15. Fil, N., Slisarenko, R., Deineko, Z., & Morozova, L. (2025). Trends in Artificial
Intelligence Research on Education: Topic Modeling Using Latent Dirichlet
Allocation. Bulletin of Kharkov National Automobile and Highway University, 108.
https://doi.org/10.30977/bul.2219-5548.2025.108.0.17
16. Georgiev, G. (2021). THE BREAKDOWN OF THE PUBLIC-PRIVATE DIVIDE IN
SECURITIES LAW: CAUSES, CONSEQUENCES, AND REFORMS GEORGE.
55
Frontiers in Neuroscience, 14(1).
17. Guizzardi, G., & Guarino, N. (2024). Explanation, semantics, and ontology. Data and
Knowledge Engineering, 153. https://doi.org/10.1016/j.datak.2024.102325
18. Gupta, S., Mahmood, A., Shetty, P., Adeboye, A., & Ramprasad, R. (2024). Data
extraction from polymer literature using large language models. Communications
Materials, 5(1). https://doi.org/10.1038/s43246-024-00708-9
19. Hou, J., Zheng, B., Li, H., & Li, W. (2025). Evolution and impact of the science of
science: from theoretical analysis to digital-AI driven research. In Humanities and
Social Sciences Communications (Vol. 12, Issue 1). https://doi.org/10.1057/s41599-025-
04617-1
20. Jungmeier, G. (2017). The Biorefinery Fact Sheet. The International Journal of Life Cycle
Assessment, 23(1).
21. Júnior, G. H. K., & Gottschalg-Duque, C. (2023). Information architecture applied on
natural language processing: a proposal Information Science contributions on data
preprocessing for training and learning of artificial neural networks. Revista Digital
de Biblioteconomia e Ciencia Da Informacao, 21.
https://doi.org/10.20396/rdbci.v21i00.8671396/30919
22. Kharimah, W. (2021). DAMPAK PENGUNDURAN DIRI AMERIKA SERIKAT
DARI PERJANJIAN JCPOA (JOINT COMPREHENSIVE PLAN OF ACTION).
Frontiers in Neuroscience, 14(1).
23. Kirkpatrick, A. W., McKee, J. L., McBeth, P. B., Ball, C. G., Laporta, A. T., Broderick,
T., Leslie, T., King, D., Wright Beatty, H. E., Keillor, J., Tien, H., de’Angelis, N., Khan,
J., Marchegiani, F., Bianchi, G., Aisoni, F., Alberti, D., Ansaloni, L., Biffl, W. L., …
MOORI, S. H. (2020). Translating Scientific Evidence into Practice using Digital
Medicine and Electronic Patient Reported Outcomes. Telemedicine and E-Health,
17(1).
24. Koçak, M., & Akçalı, Z. (2024). Development trends and knowledge framework of
artificial intelligence (AI) applications in oncology by years: a bibliometric analysis
from 1992 to 2022. Discover Oncology, 15(1). https://doi.org/10.1007/s12672-024-01415-
56
0
25. Leong, W. D., Teng, S. Y., How, B. S., Ngan, S. L., Rahman, A. A., Tan, C. P.,
Ponnambalam, S. G., Lam, H. L., Yadav, G., Luthra, S., Jakhar, S. K., Mangla, S. K.,
Rai, D. P., Tortorella, G. L., Narayanamurthy, G., Thurer, M., Dombrowski, U.,
Wullbrandt, J., Fochler, S., … Gierej, S. (2019). Elimination of waste: creation of a
successful Lean colonoscopy program at an academic medical center. Surgical
Endoscopy.
26. Luo, X., Tham, Y. C., Daher, M., Bian, Z., Chen, Y., Estill, J., & group, the G. working.
(2025). Protocol for developing the reporting guideline for the use of chatbots and
other Generative Artificial intelligence tools in MEdical Research (GAMER).
medRxiv.
27. Masso, A., Gerassimenko, J., Kasapoglu, T., & Beilmann, M. (2025). Research ethics
committees as knowledge gatekeepers: The impact of emerging technologies on
social science research. Journal of Responsible Technology, 21.
https://doi.org/10.1016/j.jrt.2025.100112
28. Morante-Carballo, F., Arcentales-Rosado, M., Caicedo-Potosí, J., & Carrión-Mero, P.
(2025). Artificial intelligence applications in hydrological studies and ecological
restoration of watersheds: A systematic review. In Watershed Ecology and the
Environment (Vol. 7). https://doi.org/10.1016/j.wsee.2025.05.004
29. Mortaheb, R., & Jankowski, P. (2023). Smart city re-imagined: City planning and
GeoAI in the age of big data. Journal of Urban Management, 12(1).
https://doi.org/10.1016/j.jum.2022.08.001
30. Munen, J. (2021). SURVEY OF CONTEMPORARY CHALLENGES FACED IN
MARKETING OF LIFE INSURANCE POLICIES IN EMBU COUNTY. Frontiers in
Neuroscience, 14(1).
31. Murikah, W., Nthenge, J. K., & Musyoka, F. M. (2024). Bias and ethics of AI systems
applied in auditing - A systematic review. Scientific African, 25.
https://doi.org/10.1016/j.sciaf.2024.e02281
32. Palazzo, F., Zambetta, G., & Palazzo, S. (2025). Artificial Intelligence at the
57
Crossroads of Engineering and Innovation. Computing&amp;AI Connect, 2(1).
https://doi.org/10.69709/caic.2025.141254
33. Palazzo, S., Palazzo, F., & Zambetta, G. (2025). The Role of Artificial Intelligence in
Advanced Engineering: Current Trends and Future Prospects. Journal of Intelligent
Communication, 4(1). https://doi.org/10.54963/jic.v4i1.959
34. Papaneophytou, C., & Nicolaou, S. A. (2025). Promoting Critical Thinking in
Biological Sciences in the Era of Artificial Intelligence: The Role of Higher Education.
In Trends in Higher Education (Vol. 4, Issue 2).
https://doi.org/10.3390/higheredu4020024
35. Pepic, I., Feldt, R., Ljungström, L., Torkar, R., Dalevi, D., Maurin Söderholm, H.,
Andersson, L. M., Axelson-Fisk, M., Bohm, K., Sjöqvist, B. A., & Candefjord, S.
(2021). Early detection of sepsis using artificial intelligence: a scoping review
protocol. Systematic Reviews, 10(1). https://doi.org/10.1186/s13643-020-01561-w
36. Perdana, R., Yulianti, D., & Bertiliya, W. A. (2025). Enhancing scientific literacy
through AI-integrated Inquiry Social Complexity module in primary education.
Journal of Educational Management and Instruction (JEMIN), 5(2).
https://doi.org/10.22515/jemin.v5i2.10891
37. Posunko, A., Chavarha, M., Bashkirova, L., Vasylyuk-Zaitseva, S., & Abramchuk, O.
(2025). Machine Learning in Medicine: A New Paradigm for Diagnosis and
Treatment. Futurity Medicine, 4(3). https://doi.org/10.57125/fem.2025.09.30.01
38. Rattan, P., Penrice, D. D., & Simonetto, D. A. (2022). Artificial Intelligence and
Machine Learning: What You Always Wanted to Know but Were Afraid to Ask. In
Gastro Hep Advances (Vol. 1, Issue 1). https://doi.org/10.1016/j.gastha.2021.11.001
39. Rincon, I. B., Soledispa-Cañarte, B. J., Sumba-Bustamante, R. V., del Carmen
Burbano-Gómez, Z., & Jiménez-Granizo, F. P. (2023). Methodological Approaches
and Paradigms in Scientific and Technological Research: a Bibliometric Review |
Abordajes Metodológicos Y Paradigmas En La Investigación Científica Y
Tecnológica: Una Revisión Bibliométrica. Bibliotecas, Anales de Investigacion, 19(1).
40. Rincon Soto, I. B., Soledispa-Cañarte, B. J., Sumba-Bustamante, R. V., del Carmen
58
Burbano-Gómez, Z., & Jiménez-Granizo, F. P. (2023). METHODOLOGICAL
APPROACHES AND PARADIGMS IN SCIENTIFIC AND TECHNOLOGICAL
RESEARCH: A BIBLIOMETRIC REVIEW. Bibliotecas, Anales de Investigacion, 19(1).
41. Rodríguez Flores, E. A., Garcés Giraldo, L. F., Valencia, J., & Valencia-Arias, A.
(2025). Research trends in the use of artificial intelligence techniques in scientific
research. Revista Venezolana de Gerencia, 30(109).
https://doi.org/10.52080/rvgluz.30.109.23
42. Saurbayev, R. Zh., Zhetpisbay, A. K., & Yerekhanova, F. T. (2024). Philological
Studies: Computer Aspects of Stylometry Automation. Iasaýı Ýnıversıtetіnіń
Habarshysy, 133(3). https://doi.org/10.47526/2024-3/2664-0686.72
43. Schneider, R., Angwin, C., Jenkinson, C., Jones, A. S. K., Jennison, C., Henley, W.,
Farmer, A., Sattar, N., Holman, R. R., Pearson, E., Shields, B., al., et, Blackwell, K.,
Donskih, R., Jones, C. M., Nixon, A., Vidal, M. J., Nakov, R., Singh, P., … Ewel, C.
(2021). Effectiveness of olive oil for the prevention of pressure ulcers caused in
immobilized patients within the scope of primary health care: study protocol for a
randomized controlled trial. Trials, 21(1).
44. Spivakovsky, O., Omelchuk, S., Kobets, V., Poltoratskyi, M., Malchykova, D., &
Lemeshchuk, O. (2025). INSTITUTIONAL REGULATION OF USING AI-BASED
SERVICES THROUGH THE KANO MODEL: THE CASE OF KHERSON STATE
UNIVERSITY. Information Technologies and Learning Tools, 108(4).
https://doi.org/10.33407/itlt.v108i4.6269
45. Taha Kandil, T. (2025). Artificial intelligence (AI) and alleviating supply chain
bullwhip effects: social network analysis-based review. Journal of Global Operations
and Strategic Sourcing, 18(1). https://doi.org/10.1108/JGOSS-04-2023-0038
46. Trejos-Gil, C. A., & Gómez-Monsalve, W. D. (2024). Artificial Intelligence in Media
and Journalism. Systematic Review on Spain and Latin America in Scopus and Web
of Science Databases (2018-2022). In Palabra Clave (Vol. 27, Issue 4).
https://doi.org/10.5294/pacla.2024.27.4.1
47. Wang, C., Nan, J., Kang, Y., Xu, X., Li, T., & Zhu, Z. (2025). Design of a
59
comprehensive teaching experiment for mechanoluminescent human biomechanical
detection. Experimental Technology and Management, 42(9).
https://doi.org/10.16791/j.cnki.sjg.2025.09.028
48. Williamson, B. (2020). New digital laboratories of experimental knowledge
production: Artificial intelligence and education research. London Review of
Education, 18(2). https://doi.org/10.14324/LRE.18.2.05
49. Zhang, B., Liu, Q., Li, X., Liu, L., Yang, B., Letu, H., Gao, L., Zhang, W., Zhang, H.,
Bian, Z., Qi, M., Chen, C., & Shang, H. (2025). The core concepts and fundamental
issues of remote sensing science. National Remote Sensing Bulletin, 29(1).
https://doi.org/10.11834/jrs.20244503
50. Zhao, T., Zhou, Y., Shi, R., Cao, Z., & Ren, Z. (2024). GWAI: Artificial intelligence
platform for enhanced gravitational wave data analysis. SoftwareX, 28.
https://doi.org/10.1016/j.softx.2024.101930
51. Zheng, Z., He, Z., Khattab, O., Rampal, N., Zaharia, M. A., Borgs, C., Chayes, J. T., &
Yaghi, O. M. (2023). Image and Data Mining in Reticular Chemistry Using GPT-4V.
ArXiv.
52. Casanova Pistón, A., & Martínez Domínguez, M. (2024). Producción científica sobre
Inteligencia Artificial y educación: un análisis cienciométrico. Hachetetepé. Revista
Científica de Educación y Comunicación, 28.
https://doi.org/10.25267/hachetetepe.2024.i28.1102
53. Guliev, I. A., Mammadov, A., & Ibrahimli, K. (2024). The Use of Artificial Intelligence
Technologies in Energy and Climate Security. Review of Business and Economics
Studies, 12(4). https://doi.org/10.26794/2308-944X-2024-12-4-58-71
60
Chapter 3.
Beyond the Binary: The Convergence of
Qualitative Inquiry and Artificial
Intelligence
The landscape of social scientific inquiry is currently undergoing a seismic
transformation, characterized by the collapsing of the traditional dichotomy between
qualitative and quantitative research. Historically, these two domains have been viewed as
distinct, often opposing, epistemological camps: qualitative research focused on "thick
description," interpretivism, and the subjective nuance of the human experience, while
quantitative research prioritized generalizability, statistical significance, and the objective
measurement of variables. However, the rapid ascent of Artificial Intelligence (AI)—
specifically Large Language Models (LLMs), Natural Language Processing (NLP), and
machine learning (ML)—has birthed a hybrid discipline: Qualitative Data Science.1
This emerging field does not merely represent the application of computational tools
to textual data; rather, it fundamentally reconfigures the epistemology of social inquiry. It
promises to resolve the "scale vs. depth" trade-off that has long plagued researchers, offering
the analytical breadth of "big data" while attempting to preserve the interpretive depth of
hermeneutics.3 As unstructured data—from social media feeds to digitized historical
archives—proliferates, the ability to analyze millions of documents with qualitative
sensitivity is no longer a luxury but a necessity for addressing what scholars term the
"futures knowledge deficit".4
However, this convergence is fraught with profound methodological and ethical
tensions. The integration of algorithmic systems into qualitative workflows raises critical
questions about "positivist drift"—the risk that the statistical logic of AI will flatten the rich,
subjective variance of human meaning into standardized, computable categories.5
61
Furthermore, the deployment of AI in social analysis introduces the threat of
"epistemological violence," where Western-centric algorithms silence or misinterpret
Indigenous and marginalized knowledge systems, perpetuating a form of "algorithmic
colonialism".7
This report provides an exhaustive, expert-level analysis of this convergence. It
dissects the theoretical frameworks enabling this shift, such as Computational Grounded
Theory and Blended Reading; it scrutinizes the specific technological architectures
facilitating it, from BERT embeddings to GPT-4's chain-of-thought reasoning; and it
evaluates the practical applications of these tools in diverse sectors, from clinical health
research to corporate ethnography. Ultimately, it argues that the future of qualitative
inquiry lies not in the automation of the researcher, but in the cultivation of "Safe Qualitative
AI"—systems designed to function as dialogic partners that enhance, rather than replace,
human meaning-making.3
1. The Epistemological Crisis and the Rise of Hybrid
Methodologies
The integration of artificial intelligence into the qualitative domain is not simply a
technical upgrade; it is an epistemological event. It forces a confrontation between two
distinct theories of knowledge: the Positivist assumption that truth is objective, observable,
and scalable, and the Interpretivist/Constructivist assumption that truth is subjective,
socially constructed, and context-dependent.
1.1 The Schism: "Thick Description" vs. "Big Data"
For decades, qualitative research has prided itself on its "human-centered" approach,
relying on the researcher's subjectivity as a primary instrument of analysis. Methods like
ethnography and phenomenology require deep immersion in the field, producing "thick
descriptions" that capture the cultural webs of meaning in which social actions are
62
suspended. In contrast, the "Big Data" revolution of the early 21st century was largely a
quantitative phenomenon, driven by the belief that with enough data, theory becomes
unnecessary—a concept famously critiqued as the "end of theory."
The rise of Qualitative Data Science challenges this schism. It posits that
computational methods can be used to perform "distant reading" of massive corpora to
identify structural patterns, which can then be interrogated through "close reading" to
derive meaning.10 This suggests a move towards Methodological Hybridity, where the
boundaries between the "qualitative" and "quantitative" are porous. As noted in recent
scholarship, AI-driven methodologies are showing improvements in consistency and
reproducibility compared to standard qualitative methods, yet they are met with resistance
from scholars who fear the loss of the "human element" essential for true interpretivism.12
1.2 Theoretical Frameworks for the Digital Age
To navigate this new landscape, scholars have developed rigorous frameworks that
justify the use of computation within interpretivist paradigms. These frameworks move
beyond viewing software as a passive container for codes (as in early Computer-Assisted
Qualitative Data Analysis Software, or CAQDAS) to viewing algorithms as active,
generative agents in the analytical process.
1.2.1 Computational Grounded Theory (CGT)
One of the most robust and systematically developed frameworks is Computational
Grounded Theory (CGT), proposed by Laura Nelson and others to bridge the gap between
expert human knowledge and the pattern recognition capabilities of computers.13
Traditional Grounded Theory, developed by Glaser and Strauss, relies on inductive
reasoning—building theory from the data up—but has historically been limited by the
human capacity to process large volumes of text (the "small N" problem).
CGT addresses this scalability issue without sacrificing the recursive, iterative
nature of the method. The framework operates through a rigorous three-step process,
63
transforming the "black box" of machine learning into a transparent partner in theory
generation 14:
● Step 1: Pattern Detection (Computational/Inductive)
This initial phase involves the inductive computational exploration of text. The
researcher employs unsupervised machine learning techniques, such as topic
modeling, clustering, or word scores, to "read" the entire corpus. This step is purely
inductive; the algorithm identifies latent structures, lexical co-occurrences, and novel
patterns that might escape human notice due to cognitive bias or sheer data volume. It
serves as a "lens" or a "map" to view the data's topography, highlighting outliers and
clusters that warrant attention.13
● Step 2: Pattern Refinement (Qualitative/Interpretive)
This is the critical "human-in-the-loop" phase. The researcher returns to the data with
a hermeneutic approach, engaging in "deep reading" of the computationally identified
clusters. Here, the raw outputs of the machine—which are merely statistical
associations—are interrogated, contextualized, and refined into meaningful
sociological or psychological concepts. The machine suggests a pattern (e.g., words
"home" and "trap" appearing together); the human interprets the meaning (e.g.,
"domestic confinement during pandemic lockdowns").14 This step honors the
interpretivist commitment to context and meaning.
● Step 3: Pattern Confirmation (Computational/Deductive)
In the final phase, the researcher uses further computational techniques (such as
supervised learning, specific NLP queries, or network analysis) to test the validity of
the refined patterns across the entire corpus. This step ensures that the insights derived
from deep reading are not idiosyncratic to a few documents but are representative of
the broader dataset. It provides a measure of rigor and reproducibility often lacking in
purely manual qualitative analysis.14
This framework allows for "resampling"—an iterative process of accessing the field,
creating a model, and computationally searching for relevant cases—thereby integrating the
64
principle of constant comparison at a scale previously impossible.17
1.2.2 Blended Reading and the Digital Humanities
Parallel to developments in sociology, the Digital Humanities have pioneered the
concept of Blended Reading. This methodology reconciles "close reading" (the careful,
nuanced analysis of individual texts) with "distant reading" (the quantitative analysis of
massive corpora, as popularized by Franco Moretti).10
Blended Reading acknowledges that in the digital age, the "classic duality of
interpreter and text has changed" due to the immense volume of digitally available data.10
It proposes a modular analysis process where computational tools are used not to replace
the scholar, but to augment their capacity to navigate "archives of abundance (see Table 5)."
Table 5: Distant, close and blended reading
Component
Methodological
Focus
Role of
AI/Computation
Analytical Outcome
Distant Reading
Macro-analysis,
structural trends,
metadata
Identifies outliers,
aberrations,
inconsistencies, and
large-scale temporal
shifts across millions
of words.11
A "topological map" of
the discourse;
identification of "hot
spots" for further
analysis.
Close Reading
Micro-analysis,
nuance, sentiment,
cultural context
None; relies on human
hermeneutics.
Provides the "ground
truth" and interpretive
depth.18
"Thick description" of
specific texts;
validation of
computational
findings.
Blended Reading
Integration of macro
and micro
Creates a feedback
loop: distant reading
guides text selection
for close reading; close
reading refines the
algorithms for distant
A multi-scalar
understanding that
links individual
narrative to structural
phenomenon.
65
reading.19
This approach is particularly vital for mitigating the "black box" problem. By forcing
a dialogue between the algorithmic output and the raw text, Blended Reading ensures that
the researcher remains anchored in the data's reality while leveraging the computer's ability
to see the "longue durée" or structural topology of the discourse.11
1.2.3 Abductive Analysis and Machine Learning
While induction builds theory from data and deduction tests theory against data,
Abduction is the logic of discovery—the leap to the best explanation for a surprising
observation. Machine learning is increasingly viewed by methodologists as an engine for
abductive analysis.20
In this context, algorithms act as generators of "surprising observations." When an
unsupervised model clusters data in a way that defies existing theoretical expectations—for
example, clustering "economic anxiety" with "health optimization" in a dataset of political
discourse—it creates a "breakdown" in understanding. This breakdown invites abductive
reasoning, forcing the researcher to generate a new hypothesis to explain this
juxtaposition.21
Scholars argue that ML systems are inherently abductive because they rely on the
contingent biases of their training data to make predictions, effectively "guessing" the nature
of new data based on a learned model of the world.22 This makes them powerful tools for
"conjectural narrativization," helping qualitative researchers identify non-obvious
relationships that require theoretical elaboration.22
1.3 The "Quantitized Qualitative" Paradigm
A growing trend within this convergence is the "Quantitized Qualitative" paradigm,
which involves the systematic transformation of qualitative observations into quantitative
66
data without losing the descriptive validity of the original observation.23
● Methodology: Qualitative themes derived from interviews or open-ended survey
responses are converted into binary (0/1) or ordinal variables. For instance, a theme of
"distrust in medical authority" identified in a patient interview is coded as a variable.
● Application: These variables are then used in advanced statistical models, such as
regression models or Structural Equation Modeling (SEM), to test causal relationships
between qualitative themes and quantitative outcomes (e.g., health status).25
● Collinearity Risks: A major critique of this approach is the risk of "collinearity," where
response categories are linked because of the coding strategy rather than reality. Mixed
methods researchers must rigorously validate that their "quantitized" data retains its
semantic integrity and is not an artifact of the coding frame.23
Structural Topic Modeling (STM) represents the state-of-the-art in this paradigm.
Unlike standard Latent Dirichlet Allocation (LDA), which treats documents as bags of
words, STM allows researchers to incorporate metadata (covariates) into the model. This
means the model can estimate how the prevalence of a qualitative topic varies according to
attributes like the author's gender, date of publication, or political affiliation.26 STM
essentially automates the "quantitizing" process, preserving the semantic richness of the text
while allowing for rigorous statistical testing of how themes relate to external variables.28
2. The Technics of Qualitative AI: Tools,
Architectures, and Performance
The theoretical frameworks described above are implemented through a rapidly
evolving stack of technologies. The shift from simple keyword searching to "semantic
understanding" has been driven by the evolution of Natural Language Processing (NLP)
architectures, specifically Transformers.
2.1 Large Language Models (LLMs) in Coding and Analysis
67
The release of generative models like ChatGPT (OpenAI), Claude (Anthropic), and
Gemini (Google) has "upended scientific and educational paradigms," specifically in the
domain of qualitative coding.29 These models possess an unprecedented ability to parse
syntax, semantics, and pragmatics, allowing them to perform tasks that previously required
human intuition.
2.1.1 Performance and Reliability: Human vs. Machine
Research comparing human coders to LLMs reveals a complex landscape where
performance is highly dependent on the model's size and the sophistication of the
prompting strategy.
● Coding Fidelity: Studies utilizing GPT-4 have demonstrated "human-equivalent
interpretations" for certain socio-historical codes, achieving high inter-coder reliability
(Cohen's Kappa \ge 0.79).30 In direct comparisons, GPT-4 significantly outperformed
earlier models like GPT-3.5, which struggled with nuance and achieved much lower
reliability scores (Mean Kappa = 0.34).30 This highlights that the capacity for nuance is a
function of model scale and architectural sophistication.
● Chain-of-Thought (CoT) Prompting: The "push-button" efficacy of these tools is
largely illusory without sophisticated interaction. Research indicates that Chain-of-
Thought (CoT) prompting—where the model is instructed to explain its reasoning
before assigning a code—considerably improves coding fidelity.30 This mirrors the
human qualitative process of writing memos or justifications for coding decisions,
suggesting that AI performs best when forced to simulate the reflexive steps of a
human researcher.
● Cost and Efficiency: Automated analysis using LLMs is significantly more cost-
effective than human coding, often reducing the time and financial investment by
orders of magnitude. However, this comes at the cost of specificity and depth.31
2.1.2 The Problem of Consistency and Hallucination
68
Despite their promise, LLMs struggle with consistency, a trait often referred to as
being a "stochastic parrot."
● Inconsistent Output: "LLMq" (Large Language Model qualitative) values can stabilize
over iterations, but the actual analytical output may remain inconsistent across runs.32
A code identified in one pass may be missed in the next, or a quote attributed to a
specific theme may be hallucinated. This unpredictability conflicts with the rigorous
audit trails required in qualitative inquiry.5
● Prompt Sensitivity: LLMs can be "distracted" by the vagaries of natural language
interfaces. Slight variations in prompts—even those that appear semantically identical
to a human—can lead to divergent analytical outcomes.33 This fragility necessitates a
new form of methodological rigor: "Prompt Engineering as Research Method".34
2.2 BERT and Embedding Models: Mapping the Semantic Space
While LLMs generate text, BERT (Bidirectional Encoder Representations from
Transformers) and similar embedding models are used to map the semantic space of a
dataset. These models convert text into high-dimensional vectors, allowing researchers to
measure the mathematical distance between concepts.
● Thematic Clustering: In thematic analysis, BERT has been employed to cluster
interview transcripts and open-ended survey responses with high precision.35 The
BERT+UMAP+HDBSCAN pipeline—which first generates embeddings (BERT),
reduces their dimensionality (UMAP), and then clusters them (HDBSCAN)—has been
identified as particularly effective for semi-structured interviews. This approach yields
topics that are both diverse and interpretable, often outperforming traditional LDA
models in coherence.36
● Predictive Validity: Studies show that BERT-based topic modeling (like BERTopic) can
outperform traditional human coding in predicting specific variables. For instance, in
an analysis of 552 psychotherapy transcripts, BERT-derived topics related to "negative
experiences" successfully predicted symptom severity.37
69
● The "Black Box" Limitation: However, the use of these models introduces a persistent
"black box" element. While a human coder can explain why they grouped two
statements (e.g., "both reflect underlying anxiety"), an embedding model groups them
based on vector proximity. This vector proximity is a mathematical abstraction of
semantic usage, which may or may not correspond to a conceptual or theoretical link
meaningful to a sociologist.38
2.3 The Evolution of CAQDAS (Computer-Assisted Qualitative Data
Analysis Software)
The market for Qualitative Data Analysis software is shifting rapidly as traditional
players integrate AI and new "AI-native" tools emerge (see Table 6).
Table 6: Tool category of key platforms
Tool Category
Key Platforms
AI Integration
Features
Strengths &
Weaknesses
Traditional
CAQDAS
ATLAS.ti, NVivo,
MAXQDA
ATLAS.ti: "AI
Coding" (powered by
OpenAI) automates
initial coding passes,
reducing time from
days to hours.39
NVivo: AI for
sentiment analysis
and autocoding by
theme; features
framed as assistive.40
Strength: deeply
integrated into
established
workflows; robust
data management.
Weakness: AI features
can feel "bolted on";
legacy interface
complexity.
AI-Native Tools
Delve, AILYZE,
Looppanel, Dovetail
Delve: "AI-assisted"
workflows; positions
AI as a "peer
debriefer" or "junior
researcher" to suggest
themes.41
Looppanel:
Strength: intuitive,
built for speed and
collaboration.
Weakness: can over-
simplify analysis; risk
of generating surface-
level "topic
70
Specialized for
UX/Focus groups;
auto-tags sentiment
and behavior
patterns.42
summaries" rather
than deep themes.44
Specialized Research
AI
Infranodus
Uses network analysis
and BERT to visualize
discourse structure
and identify
"structural gaps" in
narratives.35
Strength: visualizes
the topology of ideas.
Weakness: steeper
learning curve for
non-technical
researchers.
Critically, while tools like ATLAS.ti promise to "remove the headache" of coding,
methodological experts warn that they can only generate descriptive themes that "barely
scratch the surface" of the data's true depth.44 The consensus is that these tools are best for
deductive coding (finding known patterns) rather than inductive theory building
(discovering new meaning).45
3. Ethics, Bias, and the Politics of Algorithms:
"Epistemological Violence"
The use of AI in qualitative research introduces profound ethical questions that go
far beyond standard concerns of privacy and data security. There are deeper issues
regarding the politics of knowledge produced by algorithmic systems—issues that threaten to
undermine the very purpose of qualitative inquiry.
3.1 Epistemological Violence and Algorithmic Colonialism
A critical area of concern, drawing on postcolonial theory, is Epistemological
Violence. This concept, articulated by scholars like Gayatri Chakravorty Spivak and
Miranda Fricker (via "testimonial injustice"), refers to the harm inflicted when a dominant
knowledge system silences, invalidates, or marginalizes the knowledge systems of others.7
71
In the context of AI, this violence is structural. AI models are trained on vast corpora
of text scraped from the internet (e.g., Common Crawl), which is disproportionately English,
Western, and hegemonic. When these models are used to analyze data from non-Western
cultures or marginalized communities, they inevitably impose Western frameworks of
understanding on that data.47 This is not merely "bias" in the statistical sense; it is a form of
Algorithmic Colonialism that erases local nuance and enforces a "monoculture of the
mind".7
● Invalidation of Knowledge: AI systems may flag Indigenous concepts, non-linear
narrative structures, or dialect-specific epistemologies as "incoherent," "irrelevant," or
"errors" because they do not fit the statistical patterns of the dominant training data.49
● The "Othering" of Data: Research on migration in the Balkans has shown how
"epistemic violence" is central to the EU's border regime. Similarly, AI systems used in
asylum processing can perpetuate "testimonial injustice" by failing to recognize the
credibility of asylum seekers' narratives due to linguistic or cultural mismatches.46
● Indigenous Data Sovereignty: In response, groups like the Indigenous Protocol and
Artificial Intelligence Working Group argue for "decolonial AI." This involves developing
systems trained on and governed by Indigenous data sovereignty principles, ensuring
that the technology respects distinct epistemologies rather than flattening them.51
3.2 Automated Bias: Intrinsic and Extrinsic
Qualitative researchers have long acknowledged their own subjectivity (reflexivity).
However, AI systems are often falsely perceived by stakeholders as neutral or objective. In
reality, AI harbors deep-seated biases:
● Intrinsic Bias: This stems from the training data itself. Generative AI tends to
reproduce dominant discourses and stereotypes found in its corpus. In qualitative
coding, an AI might consistently code descriptions of poverty as "personal failure"
rather than "structural inequality" if its training data reflects neoliberal ideologies that
prioritize individual responsibility.5
72
● Extrinsic Bias: This emerges from the deployment context. A model that works "fairly"
in a Western clinical setting may fail catastrophically when applied to a Global South
context due to different cultural expressions of symptoms or distress.47
● The "Formalism Trap": Organizations often fall into the trap of thinking that using AI
makes a process "fair" or "objective" purely because it is mathematical. This ignores the
fact that fairness is a constructed, contextual social concept, not a mathematical
constant. Qualitative researchers have empirically illustrated how organizations fall
into this trap, failing to account for the "full meaning of social concepts" when they
delegate decision-making to algorithms.52
3.3 Qualitative Auditing of Algorithms
Interestingly, qualitative methods are becoming the primary tool for fixing these
quantitative problems. Qualitative Auditing involves using ethnography, interviews, and
document analysis to inspect how algorithms are actually functioning in society.53
Case Study: Hiring Algorithms – An Ethnography of Fairness
In a landmark ethnographic study of a multinational company ("MultiCo")
implementing AI for hiring, researchers van den Broek, Sergeeva, and Huysman revealed
the "formalism trap" in practice.
● Pre-AI: The HR team viewed themselves as the "guardians of fairness," defined as
suppressing bias through human judgment.
● Post-AI: The implementation of AI shifted the definition of fairness. Fairness became
synonymous with "consistency" (everyone gets the same algorithm) and "accuracy"
(predictive validity).
● The Shift: The researchers observed that the HR team began to "enroll" other
stakeholders by emphasizing these new, narrower definitions of fairness. The
qualitative inquiry uniquely uncovered that the use of AI did not just automate the
process; it fundamentally altered the values of the organization, shifting focus from
"candidate experience" to "operational efficiency".52 This demonstrates the power of
73
qualitative auditing to reveal the socio-technical reality of AI systems.
4. Applied Case Studies and Sector Analysis
The theoretical and ethical debates are grounded in a growing body of applied
research. Across health, education, and corporate sectors, the "Quantitized Qualitative"
approach is yielding mixed but potent results.
4.1 Health and Clinical Research: HIV and Psychotherapy
In health research, qualitative data (patient narratives) is critical for understanding
the "lived experience" of illness, but the volume of data often limits sample sizes. AI is being
used to scale this analysis.
● HIV Research Ethics: A commentary from Johns Hopkins researchers emphasizes that
while AI can efficiently code HIV-related qualitative data (e.g., identifying risk
behaviors in text), it must be aligned with the "underlying epistemology" of the study.
Using AI to categorize "risk behaviors" is methodologically feasible but using it to
interpret the "lived experience of stigma" risks pathologizing patients if the AI lacks the
necessary cultural nuance.29 The researchers propose a framework where the goal of the
research determines the appropriateness of AI: strictly descriptive tasks are AI-suitable,
while phenomenological interpretation remains human-bound.
● Psychotherapy and Topic Modeling: A mixed-methods study used BERTopic to
analyze 552 psychotherapy transcripts. The goal was to predict symptom severity and
"therapeutic alliance" (the bond between therapist and patient).
○ Findings: The model successfully identified topics (e.g., "negative experiences,"
"health") that strongly correlated with symptom severity (r=0.45). This
demonstrates that "quantitized" qualitative themes can be robust predictors in
clinical models.
○ Nuance: However, the "therapeutic alliance" was better predicted by the therapist's
speech patterns than the patient's, a subtle dynamic that required qualitative
74
domain expertise to interpret. The study concludes that AI allows for "treatment-
relevant metrics" to be predicted with reasonable accuracy, but only when
"explainable AI" (XAI) techniques are used to validate the topics.37
4.2 Corporate Ethnography: The "Future of Work" at Anthropic
One of the most revealing studies of AI's impact comes from within the industry
itself. Anthropic conducted an internal qualitative study, involving 53 in-depth interviews
with their own engineers, to understand how using their model (Claude) was changing their
work practices.57
● Loss of Craft: Engineers reported a complex emotional response. While productivity
soared, many expressed a "sense of loss" regarding the act of coding itself. One engineer
noted, "I thought that I really enjoyed writing code, and I think instead I actually just
enjoy what I get out of writing code."
● Shift in Social Dynamics: The study found that AI was altering workplace
socialization. Claude became the "first stop" for questions that previously would have
been directed to senior colleagues. This led to a decrease in mentorship opportunities
and human collaboration ("I like working with people and it's sad that I ‘need’ them
less now").
● Existential Uncertainty: Employees expressed "genuine uncertainty" about the future
of their profession, with some fearing they were "automating themselves out of a job."
● Methodological Significance: This study underscores that even in a hyper-
quantitative environment like an AI lab, qualitative methods (interviews,
phenomenology) were deemed essential to understand the human impact of the
technology. Quantitative metrics could measure code output, but only qualitative
inquiry could reveal the shifting professional identity of the engineers.
4.3 Education and Focus Groups
Focus groups present a unique challenge for AI due to the multi-speaker dynamics
75
and overlapping speech.
● Focus Group Analysis: Tools like Looppanel and Dovetail are increasingly used to
transcribe and analyze focus group data. AI excels at tracking "who said what" and
identifying high-level consensus. However, it often fails to capture the "group effect"—
the specific dynamic of how participants influence each other's opinions, which is the
methodological core of focus group research.58
● Student Evaluations: Structural Topic Modeling (STM) has been applied to analyze
nearly 300,000 open-ended student evaluations of teaching. The study found that topic
correlations were consistent across instructor genders, challenging the persistent
narrative that male and female instructors are evaluated on fundamentally different
criteria.28 This finding, which has significant policy implications for higher education,
was only possible due to the scale afforded by AI analysis of qualitative comments.
5. The "Safe Qualitative AI" Manifesto
In response to the dangers of "positivist drift" and "epistemic violence," a movement
for "Safe Qualitative AI" has emerged. This framework argues that researchers should not
reject AI, but rather build "dedicated qualitative AI systems" from the ground up, designed
specifically for interpretive goals rather than borrowing general-purpose tools optimized
for commercial tasks.9
5.1 Principles of Safe Qualitative AI
The "Safe Qualitative AI" framework outlines several core design principles
intended to preserve the integrity of qualitative inquiry:
1. Explanatory over Action-Oriented: AI systems should be designed to build
explanatory models (why did this happen? what does it mean?) rather than just
predictive models (what will happen next?).60 This aligns with the "Scientist AI"
concept, where the goal is understanding rather than optimization.
2. Explicit Uncertainty Quantification: The AI should articulate the limits of its
76
interpretation. Instead of presenting a code as an authoritative fact, the system should
express the certainty of its coding and provide alternative interpretations.3
3. Transparency and Reproducibility: The "black box" must be opened. Researchers need
to see how the AI arrived at a theme. This necessitates tools that provide the "chain of
thought" or the specific vector path used for classification, allowing for a rigorous audit
of the machine's "reasoning".9
4. Privacy-First Architecture: Given the sensitivity of qualitative data (often involving
vulnerable populations), Safe Qualitative AI must operate on local or private
architectures. Data should never be fed back into public base models for training,
ensuring strict adherence to confidentiality protocols.9
5.2 Human-in-the-Loop (HITL) Workflows
The consensus across the literature is that AI cannot replace the qualitative
researcher; it must be a Human-in-the-Loop (HITL) system.
● Augmentation, not Automation: AI is best used for "low-stakes," "boring," or
"verifiable" tasks (e.g., initial open coding, transcription cleaning), allowing the
researcher to focus on high-level synthesis, theory building, and "taste-based"
judgments.57
● The Dialogic Partner: The AI should be treated as a "dialogic partner"—an entity to
argue with, to test hypotheses against, and to use for "peer debriefing." The goal is to
use the AI to challenge the researcher's bias, not just to confirm it.40
● Final Interpretive Authority: The human must always retain the final say on the
validity of a code or theme. AI suggestions are treated as heuristic starting points, not
endpoints.61
6. Future Horizons: The Quantitized Qualitative
Researcher
77
As we look forward, the boundaries between the qualitative and quantitative are
likely to blur further, giving rise to a new professional identity: the Quantitized Qualitative
Researcher.
6.1 The Rise of "Scientist AI"
The future trajectory involves AI moving from a passive tool to an active
methodologist. The concept of Scientist AI envisions systems that can:
● Propose Hypotheses: Scan vast literatures to identify theoretical gaps and propose new
hypotheses.
● Design Sampling Strategies: Analyze current data saturation in real-time and suggest
which stakeholders to interview next to achieve maximum variation.62
● Synthetic Piloting: Conduct "synthetic interviews" with persona-based AI agents to
pilot interview guides before entering the field, allowing researchers to refine their
questions.62
6.2 Professional Implications
This shift will require a re-skilling of the social science workforce. Researchers will
need to be fluent in "Prompt Engineering as Methodology" and understand the basics of
Vector Semantics to effectively audit their tools. The danger is a bifurcation of the discipline
into "purists" who reject AI and "computationalists" who embrace it. However, the most
impactful research will likely come from those who can inhabit the middle ground—using
"Safe Qualitative AI" to scale their inquiry while maintaining the "thick description" that
makes qualitative research irreplaceable.
7. Conclusions and Strategic Outlook
The integration of AI into qualitative research represents a second Copernican
Revolution in the social sciences—a shift from the human as the sole center of meaning-
making to a system where meaning is co-constructed by humans and algorithms.63
78
Key Takeaways and Recommendations:
1. Hybridity is Inevitable: The volume of digital data makes purely manual qualitative
analysis increasingly untenable for large-scale societal questions. Methodologies like
Computational Grounded Theory and Blended Reading are the necessary
adaptations to this new reality. They offer a rigorous path to integrating "big data" with
"thick description."
2. Epistemology First, Technology Second: The successful use of AI depends not on the
power of the model, but on the clarity of the epistemological framework. Researchers
must explicitly define why they are using AI (e.g., for pattern detection, not truth
verification) to avoid "positivist drift."
3. Ethics as Method: Auditing for "epistemological violence" is no longer an optional
ethical step; it is a core methodological requirement. Researchers must actively
interrogate their tools for Western-centric bias and algorithmic colonialism,
prioritizing tools that allow for local / Indigenous data sovereignty.
4. The Human Remains Essential: Far from becoming obsolete, the qualitative
researcher's role is elevating. The value shifts from the labor of coding to the insight of
synthesis. The researcher becomes the guarantor of validity, the auditor of the
algorithm, and the bridge between mathematical pattern and human meaning.
In conclusion, AI offers qualitative research a telescope to see the universe of
data, but the researcher must remain the astronomer who interprets the stars. The goal
is not to automate the understanding of the human condition, but to deepen it through
a reflexive, critical, and "safe" collaboration with the machine (see Table 7).
Table 7: Key Methodological Comparisons
Feature
Traditional
Qualitative
Computational
Grounded Theory
Fully Automated (AI)
Analysis
Primary Processor
Human Brain
Human + Algorithm
Algorithm
79
(Iterative)
(LLM/BERT)
Scale
Small (N < 100)
Large (N > 1,000s)
Massive (N >
1,000,000s)
Logic of Inquiry
Inductive / Abductive
Hybrid (Inductive
detection, Qualitative
refinement)
Predominantly
Deductive / Pattern
Matching
Role of Context
Deep, "Thick
Description"
Contextualized
Patterns
Often
Decontextualized /
"Flattened"
Key Risk
Researcher Bias /
Burnout
"Black Box" opacity
Hallucination /
Epistemological
Violence
Outcome
Rich Narrative Theory
Reproducible,
Scalable Theory
Surface-level Topics /
Quantitized Data
80
Chapter bibliographies
1. Abdul Rahman, H., Masri, N. A., Husaini, A., Shaharuddin, M. Y., & Zaim, S. N. N.
(2025). Integration of Machine Learning and Artificial Intelligence Techniques for
Qualitative Research: The Rise of New Research Paradigms. International Journal of
Qualitative Methods , 24. https://doi.org/10.1177/16094069251347545
2. Access, O., Adams, W. C., Alfonita, F., Alfred, M., Briggs, S., Aloisi, A., Amorim, H.,
Moda, F., Anderson, D. N., Anwar, M. A., Graham, M., Apouey, B., Roulet, A., Solal,
I., Stabile, M., Background, P., Bajwa, U., Gastaldo, D., di Ruggiero, E., …
ZACCARIA, M. L. (2019). No 主観的健康感を中心とした在宅高齢者における 健康
関連指標に関する共分散構造分析Title. Economic and Labour Relations Review, 14(1).
3. Anis, S., & French, J. A. (2023). Efficient, Explicatory, and Equitable: Why Qualitative
Researchers Should Embrace AI, but Cautiously. In Business and Society (Vol. 62,
Issue 6). https://doi.org/10.1177/00076503231163286
4. Asiva Noor Rachmayani, Inaam Akhtar, M., Njie, B., Asimiran, S., Alase, A., Etikan,
I., Laughlin, R., Francis, J. J., Johnston, M., Robertson, C., Glidewell, L., Entwistle, V.,
Eccles, M. P., Grimshaw, J. M., Alhojailan, M. I., Ibrahim, M., Braun, V., Clarke, V.,
Taherdoost, H., … Zairul, M. (2019). No 主観的健康感を中心とした在宅高齢者にお
ける 健康関連指標に関する共分散構造分析Title. ISRA International Journal of Islamic
Finance, 8(1).
5. Avula, S., Chadwick, G., Arguello, J., Capra, R., Lippold, D., Bana e Costa, C. A., de
Corte, J.-M., Vansnick, J.-C., Linstone, H. A., Carver, C. S., Scheier, M. F., Segerstrom,
S. C., Berger, G., Armand, L., Bresard, S., Grimanelli, P., Landucci, A., Schwartz, B.,
Bolger, F., … Birbilis, G. (2015). Editorial Board. Technological Forecasting and Social
Change, 77(1).
6. Bhatia, A., Chandani, A., Atiq, R., Mehta, M., & Divekar, R. (2021). Artificial
intelligence in financial services: a qualitative research to discover robo-advisory
services. Qualitative Research in Financial Markets, 13(5).
81
https://doi.org/10.1108/QRFM-10-2020-0199
7. Brailas, A. (2025). Artificial Intelligence in Qualitative Research: Beyond
Outsourcing Data Analysis to the Machine. Psychology International, 7(3).
https://doi.org/10.3390/psycholint7030078
8. Brondani, M., Alves, C., Ribeiro, C., Braga, M. M., Garcia, R. C. M., Ardenghi, T., &
Pattanaporn, K. (2024). Artificial intelligence, ChatGPT, and dental education:
Implications for reflective assignments and qualitative research. Journal of Dental
Education, 88(12). https://doi.org/10.1002/jdd.13663
9. Cheekaty, S., & Muneeswari, G. (2024). Exploring Sparse Gaussian Processes for
Bayesian Optimization in Convolutional Neural Networks for Autism Classification.
IEEE Access, 12. https://doi.org/10.1109/ACCESS.2024.3351168
10. Christou, P. (2023). How to Use Artificial Intelligence (AI) as a Resource, ow
Methodological and Analysis Tool in Qualitative Research? Qualitative Report, 28(7).
https://doi.org/10.46743/2160-3715/2023.6406
11. Christou, P. A. (2023a). A Critical Perspective Over Whether and How to
Acknowledge the Use of Artificial Intelligence (AI) in Qualitative Studies. Qualitative
Report, 28(7). https://doi.org/10.46743/2160-3715/2023.6407
12. Christou, P. A. (2023b). The Use of Artificial Intelligence (AI) in Qualitative Research
for Theory Development. Qualitative Report, 28(9). https://doi.org/10.46743/2160-
3715/2023.6536
13. de Seta, G., Pohjonen, M., & Knuutila, A. (2024). Synthetic ethnography: Field
devices for the qualitative study of generative models. Big Data and Society, 11(4).
https://doi.org/10.1177/20539517241303126
14. Durrani, B. A., Lecturer, S., Ivanauskien, N., Volung, J., Rahmani-Nejad, L.,
Firoozbakht, Z., Taghipoor, A., Arokiasamy, A. R. A., Pourdehghan, A., Tjan, S.,
Sukamto, R., Lumintan, D. B., Dhurup, M., Mafini, C., Dumasi, T., Kheng, L. L.,
Mahamad, O., Ramayah, T., Mosahab, R., … Wirawan, V. (2014). Implementation of
E-Government in Welcoming the Contemporary Industrial Revolution 4.0 Era in
Indonesia. International Journal of Marketing Studies, 1(1).
82
15. Fahyuni, E. F., Wasis, Bandono, A., & Arifin, M. B. U. B. (2020). Integrating islamic
values and science for millennial students’ learning on using seamless mobile media.
Jurnal Pendidikan IPA Indonesia, 9(2). https://doi.org/10.15294/jpii.v9i2.23209
16. Fieldhouse, J. K., Ge, J., Randhawa, N., Wolking, D., Genovese, B. N., Mazet, J. A. K.,
& Desai, A. (2025). The intersection of artificial intelligence with qualitative or mixed
methods for communicable disease research: a scoping review. In Public Health (Vol.
248). https://doi.org/10.1016/j.puhe.2025.105961
17. Flores, L. G. R., & Mejía Rivera, K. A. (2024). ARTIFICIAL INTELLIGENCE IN
QUALITATIVE RESEARCH: BIBLIOMETRIC ANALYSIS OF SCIENTIFIC
PRODUCTION INDEXED IN SCOPUS. New Trends in Qualitative Research, 20(4).
https://doi.org/10.36367/ntqr.20.4.2024.e1116
18. Friedman, C., Owen, A., & VanPuymbrouck, L. (2025). Should ChatGPT help with
my research? A caution against artificial intelligence in qualitative analysis.
Qualitative Research, 25(5). https://doi.org/10.1177/14687941241297375
19. Göçer, Ö., Hua, Y., Göçer, K., Gunhan, S., Moaveni, S., Chou, K. C., Charlson, J.,
Chinyio, E., Zhu, Y. Q., Zuo, J. J., Korpela, J., Kerosuo, H., Abuelmaatti, A., Ahmed,
V., Jing, X., Wang, Q. K. Q., Li, Y. Y. Y., Zhao, D., McCoy, A. P., … Wu, W. W. Y.
(2014). A Ship in a Box. Applied Mechanics and Materials, 29(2).
20. Goyanes, M., Lopezosa, C., & Jordá, B. (2025). Thematic analysis of interview data
with ChatGPT: designing and testing a reliable research protocol for qualitative
research. Quality and Quantity, 59(6). https://doi.org/10.1007/s11135-025-02199-3
21. Gunewardena, D., Seck, A., Warnecke, T., Green, E. P., Blattman, C., Jamison, J.,
Annan, J., Dvouletý, O., Pepelasis, I. S., Protogerou, A., Krasniqi, B. A., Williams, N.,
Osirim, M. J., Tobias, J. M., Boudreaux, K. C., Botterill, L. C., Free, M., Demirgüc-
Kunt, A., Klapper, L. F., … Naszályi, P. (2018). Abstracts. Journal of Planning
Literature, 29(2).
22. Güzer, B., Caner, H., Fathoni, A. F., van der Mescht, H., Whittaker, J. K., Staker, H.,
Horn, M. B., Crawford, R., Richey, R. C., Christensen, C. M., Horn, M. B., Staker, H.,
Thoms, C. L. v, Chen, J. J., Petty, J., Albiladi, W. S., Alshareef, K. K., Zhang, W., Zhu,
83
C., … Province, Q. (2019). No 主観的健康感を中心とした在宅高齢者における 健康
関連指標に関する共分散構造分析Title. Computers and Education, 7(1).
23. Hogg, H. D. J., Al-Zubaidy, M., Keane, P. A., Hughes, G., Beyer, F. R., &
Maniatopoulos, G. (2023). Evaluating the translation of implementation science to
clinical artificial intelligence: a bibliometric study of qualitative research. In Frontiers
in Health Services (Vol. 3). https://doi.org/10.3389/frhs.2023.1161822
24. Hogg, H. D. J., Sendak, M. P., Denniston, A. K., Keane, P. A., & Maniatopoulos, G.
(2023). Unlocking the potential of qualitative research for the implementation of
artificial intelligence-enabled healthcare. In Journal of Medical Artificial Intelligence
(Vol. 6). https://doi.org/10.21037/jmai-23-28
25. Inaam Akhtar, M., Njie, B., Asimiran, S., Alase, A., Etikan, I., Laughlin, R., Francis, J.
J., Johnston, M., Robertson, C., Glidewell, L., Entwistle, V., Eccles, M. P., Grimshaw,
J. M., Alhojailan, M. I., Ibrahim, M., Braun, V., Clarke, V., Taherdoost, H.,
SWELHEIM, T., … Zairul, M. (2019). No 主観的健康感を中心とした在宅高齢者にお
ける 健康関連指標に関する共分散構造分析Title. ISRA International Journal of Islamic
Finance, 8(1).
26. Jain, H., & KUMAR, Prof. D. (2025). Artificial Intelligence as a Catalyst for Supply
Chain Evolution: A Multi-Case Thematic Study. INTERNATIONAL JOURNAL OF
SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT, 09(09).
https://doi.org/10.55041/ijsrem52737
27. Jiang, Y., Ko-Wong, L., & Valdovinos Gutierrez, I. (2025). The Feasibility and
Comparability of Using Artificial Intelligence for Qualitative Data Analysis in
Equity-Focused Research. In Educational Researcher (Vol. 54, Issue 3).
https://doi.org/10.3102/0013189X251314821
28. John, A. O., & Chukwumerije, T. (2014). Perceptions of Accountants on Factors
Affecting Auditor’s Independence in Nigeria. International Journal of Research in
Commerce, IT & Management, 2(11).
29. Kang, R., Xuan, Z., Tong, L., Wang, Y., Jin, S., & Xiao, Q. (2025). Nurse Researchers’
84
Experiences and Perceptions of Generative AI: Qualitative Semistructured Interview
Study. Journal of Medical Internet Research, 27. https://doi.org/10.2196/65523
30. Kozinets, R. v., & Gretzel, U. (2024). Netnography evolved: New contexts, scope,
procedures and sensibilities. Annals of Tourism Research, 104.
https://doi.org/10.1016/j.annals.2023.103693
31. Lei, P., Duo, E. N., Gao, Y., & Zhu, X. (2025). Research on the experience of utilizing
artificial intelligence in nursing learning from the perspective of nursing students –
a qualitative study. BMC Nursing, 24(1). https://doi.org/10.1186/s12912-025-03704-7
32. Li, Z. Q., Shroff, P. K., Venkataraman, R., Zhang, I. X., Gu, F., Lev, B., He, G., Kang,
J. K. H. K., Shivdasani, A., Yamada, T., Bhattacharyya, S., Nain, A. S., Lee, K. B. H.,
Homburg, C., Bucerius, M., Erel, I., Liao, R. C., WEISBACH, M. S., Andrade, G., …
Lys, T. Z. (2019). CORPORATE TAKEOVERS AND PRODUCTIVITY -
LICHTENBERG,FR. JOURNAL OF BANKING & FINANCE, 37(2).
33. Lim, W. M. (2025). What Is Qualitative Research? An Overview and Guidelines.
Australasian Marketing Journal, 33(2). https://doi.org/10.1177/14413582241264619
34. Longo, L. (2020). Empowering Qualitative Research Methods in Education with
Artificial Intelligence. Advances in Intelligent Systems and Computing, 1068.
https://doi.org/10.1007/978-3-030-31787-4_1
35. Luongo, M., Ponticorvo, M., Ligorio, M. B., Crescenzo, P., & Ritella, G. (2024).
Artificial Intelligence to Enhance Qualitative Research: Methodological Reflections
on a Pilot Study. Journal of Educational, Cultural and Psychological Studies, 2024(30).
https://doi.org/10.7358/ecps-2024-030-luon
36. Marshall, D. T., & Naff, D. B. (2024). The Ethics of Using Artificial Intelligence in
Qualitative Research. Journal of Empirical Research on Human Research Ethics, 19(3).
https://doi.org/10.1177/15562646241262659
37. Nashwan, A. J., & Abukhadijah, H. (2023). Harnessing Artificial Intelligence for
Qualitative and Mixed Methods in Nursing Research. Cureus.
https://doi.org/10.7759/cureus.48570
38. Nicmanis, M., & Spurrier, H. (2025). Getting Started with Artificial Intelligence
85
Assisted Qualitative Analysis: An Introductory Guide to Qualitative Research
Approaches with Exploratory Examples from Reflexive Content Analysis.
International Journal of Qualitative Methods , 24.
https://doi.org/10.1177/16094069251354863
39. Olawade, D. B., Omeni, D., Gore, M. N., & Hadi, M. (2025). Enhancing qualitative
research through virtual focus groups and artificial intelligence: A review.
International Journal of Medical Informatics, 203.
https://doi.org/10.1016/j.ijmedinf.2025.106004
40. Owoahene Acheampong, I., & Nyaaba, M. (2024). Review of Qualitative Research in
the Era of Generative Artificial Intelligence. SSRN Electronic Journal.
https://doi.org/10.2139/ssrn.4686920
41. Pratt, M. G. (2025). On the Evolution of Qualitative Methods in Organizational
Research. In Annual Review of Organizational Psychology and Organizational Behavior
(Vol. 12). https://doi.org/10.1146/annurev-orgpsych-111722-032953
42. Roberts, J., Baker, M., & Andrew, J. (2024). Artificial intelligence and qualitative
research: The promise and perils of large language model (LLM) ‘assistance.’ Critical
Perspectives on Accounting, 99. https://doi.org/10.1016/j.cpa.2024.102722
43. Rodríguez Flores, E. A., & Sánchez Trujillo, M. de los Á. (2025). Scientific research
and artificial intelligence in graduate students. A qualitative analysis. European
Public and Social Innovation Review, 10. https://doi.org/10.31637/epsir-2025-1049
44. Rosenberg, J. (2024). Automated Transcription of Interviews in Qualitative Research
Using Artificial Intelligence A Simple Guide. Journal of Surgery Research and Practice.
https://doi.org/10.46889/jsrp.2024/5204
45. Saheb, T., Sidaoui, M., & Schmarzo, B. (2024). Convergence of artificial intelligence
with social media: A bibliometric & qualitative analysis. In Telematics and Informatics
Reports (Vol. 14). https://doi.org/10.1016/j.teler.2024.100146
46. Sampaio, R. C., Chagas, V., Sanchez, C. S., Gonçalves, J., Borges, T., Alison, M. B.,
Tigrinho, C. S., de Souza, J. R., & Paz, F. S. (2024). AN ARTIFICIAL INTELLIGENCE
(AI)-ASSISTED SCOPING REVIEW OF EMERGING USES OF AI IN QUALITATIVE
86
RESEARCH AND ITS ETHICAL CONSIDERATIONS. Revista Pesquisa Qualitativa,
12(30). https://doi.org/10.33361/RPQ.2024.v.12.n.30.729
47. Schmidt, D. A., Alboloushi, B., Thomas, A., & Magalhaes, R. (2025). Integrating
artificial intelligence in higher education: perceptions, challenges, and strategies for
academic innovation. Computers and Education Open, 9.
https://doi.org/10.1016/j.caeo.2025.100274
48. Sposato, M. (2025). Introducing AI to qualitative research: Theoretical and practical
implications. In Exploring Qualitative Research in Business: Approaches, Strategies, And
Methods. https://doi.org/10.4018/979-8-3693-7698-0.ch010
49. Tierney, A., Peasey, P., & Gould, J. (2025). Student perceptions on the impact of AI
on their teaching and learning experiences in higher education. Research and Practice
in Technology Enhanced Learning, 20. https://doi.org/10.58459/rptel.2025.20005
50. Tong, L., Niu, Y., Zhou, L., Jin, S., Wang, Y., & Xiao, Q. (2025). Perceptions and
experiences of generative artificial intelligence training to support research for
Chinese nurses: A qualitative focus group study. International Journal of Nursing
Sciences, 12(3). https://doi.org/10.1016/j.ijnss.2025.04.010
51. Yang, Y., & Ma, L. (2025). Artificial intelligence in qualitative analysis: a practical
guide and reflections based on results from using GPT to analyze interview data in
a substance use program. Quality and Quantity, 59(3). https://doi.org/10.1007/s11135-
025-02066-1
52. Yang, Y., & Siau, K. (2018). A Qualitative Research on Marketing and Sales in the
Artificial Intelligence Age. Midwest Association for Information Systems
ConferenceAssociation for Information Systems Conference, 5(17-18 May 2018).
53. Yıldırım, H. M., Güvenç, A., & Can, B. (2025). Reporting Quality in Qualitative
Research Using Artificial Intelligence and the COREQ Checklist: A Systematic
Review of Tourism Literature. International Journal of Qualitative Methods , 24.
https://doi.org/10.1177/16094069251351487
54. Choudrie, J., Manandhar, N., Castro, C., & Obuekwe, C. (2023). Hey Siri, Google!
Can you help me? A qualitative case study of smartphones AI functions in SMEs.
87
Technological Forecasting and Social Change, 189.
https://doi.org/10.1016/j.techfore.2023.122375
55. Dai, Y., Lai, S., Lim, C. P., & Liu, A. (2023). ChatGPT and its impact on research
supervision: Insights from Australian postgraduate research students. Australasian
Journal of Educational Technology, 39(4). https://doi.org/10.14742/ajet.8843
56. Djakfar Musthafa, F. A. (2024). Penggunaan Artificial Intelligence (AI) dalam
Pembelajaran: Fenomena Transformasi Otoritas Pengetahuan di Kalangan
Mahasiswa. Journal of Contemporary Islamic Education, 4(1).
https://doi.org/10.25217/jcie.v4i1.4386
57. Hogg, H. D. J., Al-Zubaidy, M., Talks, J., Denniston, A. K., Kelly, C. J., Malawana, J.,
Papoutsi, C., Teare, M. D., Keane, P. A., Beyer, F. R., & Maniatopoulos, G. (2023).
Stakeholder Perspectives of Clinical Artificial Intelligence Implementation:
Systematic Review of Qualitative Evidence. In Journal of Medical Internet Research
(Vol. 25). https://doi.org/10.2196/39742
58. Kantor, J. (2024). Best practices for implementing ChatGPT, large language models,
and artificial intelligence in qualitative and survey-based research. In JAAD
International (Vol. 14). https://doi.org/10.1016/j.jdin.2023.10.001
59. Naeem, M., Smith, T., & Thomas, L. (2025). Thematic Analysis and Artificial
Intelligence: A Step-by-Step Process for Using ChatGPT in Thematic Analysis.
International Journal of Qualitative Methods , 24.
https://doi.org/10.1177/16094069251333886
60. Orea-Giner, A., Fusté-Forné, F., & Soliman, M. (2025). How do tourists perceive
green customer-love service in restaurants? A qualitative exploration of ai and
human collaboration. International Journal of Hospitality Management, 131.
https://doi.org/10.1016/j.ijhm.2025.104300
61. Tschisgale, P., Wulff, P., & Kubsch, M. (2023). Integrating artificial intelligence-based
methods into qualitative research in physics education research: A case for
computational grounded theory. Physical Review Physics Education Research, 19(2).
https://doi.org/10.1103/PhysRevPhysEducRes.19.020123
88
62. Venter, I. M., Blignaut, R. J., Cranfield, D. J., Tick, A., & Achi, S. el. (2025). AI versus
tradition: shaping the future of higher education. Journal of Applied Research in Higher
Education, 17(7). https://doi.org/10.1108/JARHE-12-2024-0702
63. Wang, S., & Kim, R. Y. (2025). Exploring the role of generative AI in enhancing
critical thinking in multicultural classrooms: a case study from korea. Multicultural
Education Review, 17(1). https://doi.org/10.1080/2005615X.2025.2467787
89
Chapter IV.
Quantitative Research Methodology in
Artificial Intelligence and Data Science: A
Comprehensive Framework for Empirical
Analysis
1. Epistemological Foundations of Quantitative
Analysis in Computational Intelligence
The evolution of Artificial Intelligence (AI) and Data Science has fundamentally
shifted the discipline of computer science from a deterministic, logic-based field into an
empirical science heavily reliant on quantitative research methodologies. In the early eras
of symbolic AI, systems were constructed upon rigid, rule-based logic where validation was
a binary state: a theorem was either proven or it was not; a logic gate returned true or false.
However, the modern paradigm—dominated by stochastic machine learning (ML), deep
learning (DL), and probabilistic graphical models—requires a fundamental epistemological
shift. We no longer validate systems based on logical absolutes but rather evaluate them
within a probabilistic state space.1
In this contemporary context, quantitative research is defined as the systematic
investigation of phenomena through the collection of quantifiable data and the application
of statistical, mathematical, and computational techniques.2 The objective is not merely to
build a system that functions but to construct a rigorous evidentiary basis that quantifies
how well it functions, under what conditions it fails, and how confident we can be in its
predictions. This transition necessitates a rigorous framework for empirical evaluation that
mirrors the experimental rigor found in physics or medicine. A model achieving 95%
accuracy is not "correct" in the absolute sense; it is merely statistically likely to be correct
90
under a specific, observed distribution of data.3 Therefore, the methodologies used to
evaluate these systems must be rooted in statistical theory, experimental design, and
quantitative metrics that can capture the nuance of probabilistic performance.4
The scope of quantitative methodology in AI extends far beyond simple
performance metrics like accuracy or precision. It encompasses the rigorous design of
experiments to control for confounding variables, the statistical comparison of algorithms
to ensure reproducibility, the quantification of bias and fairness to align with ethical
standards, and the measurement of operational efficiency metrics such as latency and
carbon footprint.5 As AI systems become integral to critical decision-making processes in
healthcare, finance, and autonomous systems, the demand for "scientific rigor"—defined by
reproducibility, statistical significance, and transparent methodology—has become
paramount.5 The "reproducibility crisis" currently facing the field serves as a stark reminder
that without robust quantitative standards, the rapid pace of innovation risks producing
fragile, unreliable, or biased systems.
1.1 The Shift from Qualitative to Quantitative Paradigms
Historically, computer science education and research utilized a mix of
methodologies, often leaning heavily on theoretical proofs or qualitative demonstrations of
capability. However, recent bibliometric analyses indicate a dominant, accelerating trend
toward quantitative methods in AI research.1 This shift is driven by the necessity of
empirical validation for high-dimensional, non-linear models where theoretical bounds—
such as the Vapnik–Chervonenkis (VC) dimension—are often too loose to be practically
useful for predicting performance on real-world data. Instead, researchers rely on large-
scale empirical testing on benchmark datasets to quantify progress.
This quantitative dominance brings its own set of challenges. The reliance on
empirical metrics has led to a "leaderboard culture," where marginal improvements in
quantitative scores are prioritized over methodological soundness or interpretability.
91
Furthermore, the "reproducibility crisis" in AI suggests that while quantitative metrics are
widely used, the methodologies surrounding their reporting often lack the necessary rigor.5
Issues such as data leakage, improper splitting strategies, and the lack of statistical
significance testing contribute to inflated claims of state-of-the-art (SOTA) performance.
Consequently, a nuanced understanding of quantitative methodology is not merely an
academic exercise but a critical requirement for valid engineering practice in the age of AI.
The following report details a comprehensive framework for this quantitative practice,
spanning experimental design, metric selection, statistical validation, fairness auditing, and
operational assessment.
2. Experimental Design and Sampling Strategies
The foundation of any rigorous quantitative study is the experimental design. In the
context of AI and Data Science, this primarily concerns how data is collected, partitioned,
and utilized to train and evaluate models. The objective of experimental design is to estimate
the generalization error—the expected performance of the model on unseen data drawn
from the same underlying distribution—as accurately and unbiasedly as possible.8 Without
a robust design, even the most sophisticated model architectures and metrics yield
meaningless results.
2.1 The Bias-Variance Decomposition in Data Partitioning
The central challenge in estimating model performance is the bias-variance trade-off
inherent in resampling methods. If we use all available data for training, we maximize the
information available to the model (reducing bias), but we have no independent data left
for evaluation (preventing variance estimation). Conversely, if we reserve a large portion of
data for testing, the model trained on the smaller remainder may not represent the true
potential of the algorithm (increasing bias), and the small test set may yield unstable
performance estimates (increasing variance).
2.1.1 The Holdout Method and Its Limitations
92
The simplest quantitative method is the holdout method, where the dataset D is
partitioned into two disjoint sets: a training set D_{train} and a test set D_{test}. A common
split ratio is 70:30 or 80:20. While computationally efficient, the holdout method is highly
sensitive to the specific partition of data. A "lucky" split might place all easy-to-classify
instances in the test set, resulting in an optimistic bias, while an "unlucky" split does the
reverse.8 This variance is particularly problematic in small to medium-sized datasets,
making the holdout method less robust for rigorous scientific comparison. In modern "Deep
Research," reliance on a single holdout split is often considered insufficient evidence of
superiority unless the dataset size is massive (e.g., millions of examples) such that the law
of large numbers stabilizes the error estimates.
2.1.2 K-Fold Cross-Validation: The Gold Standard
To mitigate the high variance of the holdout method, K-Fold Cross-Validation (CV)
has emerged as the standard practice in quantitative evaluation. The dataset is randomly
partitioned into k equal-sized subsamples (folds). The process is an iterative rotation: of the
k subsamples, a single subsample is retained as the validation data for testing the model,
and the remaining k-1 subsamples are used as training data.8 This process is repeated k
times, with each of the k subsamples used exactly once as the validation data.
The k results are then averaged to produce a single estimation. The advantage of this
method is that all observations are used for both training and validation, and each
observation is used for validation exactly once. The formula for the cross-validation estimate
is:
CV_{(k)} = \frac{1}{k} \sum_{i=1}^{k} L_i
where L_i is the loss or error metric on the i-th fold.
Stratified K-Fold CV: In classification tasks, particularly with imbalanced datasets
(e.g., fraud detection where positive cases are <1%), random splitting can result in folds that
93
have no positive examples, rendering training or testing impossible. Stratification addresses
this by ensuring that the proportion of samples for each class in every fold matches the
proportion in the complete dataset. This reduces the variance of the performance estimate
and ensures that the model is evaluated on a representative distribution of the minority
class.9
2.1.3 Repeated Cross-Validation and the 5x2cv Protocol
While 10-fold CV is standard, it still possesses variance due to the randomness of the
initial partitioning. Repeated K-Fold CV involves repeating the entire K-Fold process
multiple times with different random seeds for the splits. This provides a better Monte-
Carlo estimate of the complete cross-validation performance.9
A specific and statistically critical variant is 5x2 Cross-Validation, proposed by
Dietterich (1998) specifically for the statistical comparison of machine learning algorithms.
It involves performing 2-fold cross-validation five times. This protocol was developed to
address the high Type I error rates (false positives) often observed when comparing
algorithms using standard 10-fold CV. In standard 10-fold CV, the training sets for each fold
overlap by 90%, violating the independence assumption required by most statistical tests.
The 5x2cv method creates training sets that are more disjoint within each replication,
providing a more robust basis for hypothesis testing.10
2.2 Resampling Techniques: The Bootstrap
Bootstrapping is a powerful statistical method for estimating the sampling
distribution of an estimator by resampling with replacement from the original data.12 In AI
evaluation, bootstrapping serves a distinct purpose from cross-validation and is particularly
valuable for quantifying uncertainty.
In the bootstrap method, a dataset of size n is resampled with replacement n times
to create a "bootstrap sample." Probability theory dictates that the probability of any specific
instance not being chosen in a sample of size n is (1 - 1/n)^n. As n \to \infty, this converges
94
to 1/e \approx 0.368. Thus, on average, a bootstrap sample contains approximately 63.2\%
of the original unique instances, leaving about 36.8\% of the data as "out-of-bag" (OOB)
samples.13
The.632+ Bootstrap Method:
The OOB samples can serve as a test set. However, since the training set (the
bootstrap sample) only contains \approx 63\% of unique data, models might underperform
compared to those trained on the full n samples (pessimistic bias). The.632 bootstrap
estimator attempts to correct this bias by taking a weighted average of the training error
(which is usually optimistically low, often zero for overfitted models) and the OOB error:
Err_{.632} = 0.368 \cdot Err_{train} + 0.632 \cdot Err_{OOB}
While useful for small datasets, bootstrapping is computationally expensive. Its
modern application is increasingly focused not just on point estimates of accuracy, but for
generating Confidence Intervals (CIs) around performance metrics. By calculating the
metric on thousands of bootstrap replicates, researchers can report a 95% confidence
interval (e.g., "Accuracy: 85% \pm 2%"), which is a critical aspect of rigorous quantitative
reporting.14
2.3 Dealing with Data Leakage and Temporal Dependencies
A rigorous experimental design must explicitly prevent data leakage, a phenomenon
where information from the test set improperly influences the training process, leading to
overly optimistic performance estimates.16 This is one of the most common causes of
reproducibility failure in AI (see Table 8).
Table 8: Common causes of reproducibility failure in AI
95
For time-series data or spatial data, random splitting (like standard K-Fold)
constitutes severe data leakage. If a model is predicting stock prices, training on data from
Tuesday and Thursday to predict Wednesday is invalid because it violates the causality of
the information flow. In such cases, Rolling Window Validation or Walk-Forward
Validation must be used, where the training set consists only of data historically preceding
the validation set.16
2.4 Reproducibility and Rigor in Experimental Design
The "reproducibility crisis" in AI has catalyzed a movement toward stricter
quantitative standards. Evaluating an AI model is not merely about reporting a final
number but documenting the entire experimental process. The NeurIPS Reproducibility
Checklist has become a de facto standard for quantitative rigor.17 Adherence to this checklist
ensures that the experimental design is transparent and replicable.
Key elements of reproducible quantitative design include:
● Data Splitting Transparency: Explicitly stating how train/validation/test splits were
96
generated, including the exact random seed used. This allows other researchers to
reconstruct the exact data subsets.17
● Hyperparameter Search Space: Reporting the range of hyperparameters considered
and the method of selection (e.g., grid search, random search, Bayesian optimization).
Reporting only the best result from a massive search without adjusting for the "multiple
comparisons problem" leads to overfitting the validation set, a practice sometimes
called "p-hacking" in statistics or "gradient descent on the test set" in ML.5
● Infrastructure Specifications: Documenting the hardware (GPU/TPU types) and
software environment (library versions), as floating-point arithmetic can vary slightly
across platforms, impacting the exact reproducibility of training trajectories.17
3. Performance Evaluation Metrics: Quantifying
Success
In quantitative research, the choice of metric defines the optimization landscape. A
mismatch between the quantitative metric and the actual business or scientific objective can
lead to "successful" models that fail in practice. This phenomenon, often summarized by
Goodhart's Law ("When a measure becomes a target, it ceases to be a good measure"),
necessitates a careful selection of diverse metrics to capture the full behavior of the system.
3.1 Metrics for Classification: Beyond Accuracy
For categorical prediction tasks, Accuracy is the most intuitive metric but is often
misleading, particularly in imbalanced datasets—a situation known as the "accuracy
paradox." In a fraud detection dataset where only 0.1% of transactions are fraudulent, a
naive classifier that predicts "valid" for every single transaction achieves 99.9% accuracy but
has zero utility. Thus, rigorous quantitative analysis requires a decomposition of errors.3
3.1.1 The Confusion Matrix and Derived Metrics
The foundational tool for this decomposition is the Confusion Matrix, an N \times
97
N table that cross-tabulates predicted classes against actual classes. For a binary problem,
this breaks down predictions into True Positives (TP), True Negatives (TN), False Positives
(FP), and False Negatives (FN).20 From these primitives, we derive metrics that isolate
specific types of performance:
● Precision (Positive Predictive Value): \frac{TP}{TP+FP}. This measures the
trustworthiness of a positive prediction. In a spam filter, high precision is critical to
avoid flagging legitimate emails as spam.
● Recall (Sensitivity/True Positive Rate): \frac{TP}{TP+FN}. This measures the ability of
the model to capture all positive instances. In medical diagnosis for a lethal but curable
disease, high recall is paramount, even at the cost of precision.3
● F1 Score: The harmonic mean of Precision and Recall. The harmonic mean is used
because it punishes extreme values more than the arithmetic mean; if either Precision
or Recall is low, the F1 score will be low.
F1 = 2 \cdot \frac{Precision \cdot Recall}{Precision + Recall}
3.1.2 Probabilistic Threshold-Invariant Metrics
Classifiers typically output a probability score (e.g., 0.85 chance of churn) rather than
a hard label. Converting this to a label requires a threshold (e.g., >0.5). Metrics like Accuracy
and F1 are sensitive to this threshold choice. To evaluate the discriminative power of the
model independent of the threshold, we use the ROC (Receiver Operating Characteristic)
curve and the AUC (Area Under the Curve).
● ROC Curve: Plots the True Positive Rate (Recall) against the False Positive Rate (1 -
Specificity) at every possible classification threshold.
● AUC-ROC: Represents the probability that a randomly chosen positive instance is
ranked higher (has a higher predicted probability) than a randomly chosen negative
instance. An AUC of 0.5 implies random guessing, while 1.0 implies perfect
separability. The AUC is robust to class imbalance and provides a holistic view of the
98
classifier's ability to distinguish between classes.20
3.2 Metrics for Regression: Quantifying Error Magnitude
For tasks predicting continuous output variables (e.g., price, temperature), metrics
must quantify the magnitude of the distance between predicted values (\hat{y}) and actual
values (y).23
● RMSE (Root Mean Squared Error): \sqrt{\frac{1}{n}\sum_{i=1}^{n} (y_i -
\hat{y}_i)^2}. By squaring the errors before averaging, RMSE penalizes large errors
disproportionately. This is desirable when large errors are particularly costly (e.g.,
predicting the trajectory of a self-driving car).
● MAE (Mean Absolute Error): \frac{1}{n}\sum_{i=1}^{n} |y_i - \hat{y}_i|. MAE
provides a linear score, meaning all errors are weighted equally. It is often more
interpretable as the "average mistake."
● R^2 (Coefficient of Determination): Represents the proportion of the variance in the
dependent variable that is predictable from the independent variables. While useful for
explanation, R^2 can be misleading in non-linear models and does not indicate whether
the predictions are biased.
3.3 Advanced Metrics for Natural Language Processing (NLP)
The quantitative evaluation of Generative AI and NLP is significantly more complex
because "correctness" is subjective. Unlike a classification label, there is rarely a single
correct sequence of words for a translation or summary. Evaluation thus relies on measuring
similarity to human-generated reference texts.
3.3.1 N-gram Based Metrics: BLEU and ROUGE
BLEU (Bilingual Evaluation Understudy): Originally designed for machine
translation, BLEU measures the precision of n-grams (sequences of n words) in the
candidate text compared to the reference text(s). It is a Precision-oriented metric.25 The BLEU
99
score formulation includes a Brevity Penalty (BP) to prevent the model from gaming the
metric by outputting very short, high-precision sentences (e.g., outputting just "The" when
the reference is "The cat sat on the mat" would yield 100% unigram precision without the
penalty).
BLEU = BP \cdot \exp\left(\sum_{n=1}^{N} w_n \log p_n\right)
where p_n is the modified n-gram precision and w_n are weights (typically uniform). The
use of the geometric mean ensures that if precision for any n-gram order is zero, the entire
score is zero (unless smoothed), enforcing quality across different granularities (1-gram for
adequacy, 4-gram for fluency).26
ROUGE (Recall-Oriented Understudy for Gisting Evaluation): Commonly used for
text summarization, ROUGE focuses on Recall—how much of the reference summary was
captured by the generated text?.28
● ROUGE-N: Measures the overlap of N-grams.
● ROUGE-L: Based on the Longest Common Subsequence (LCS). ROUGE-L captures
sentence structure by identifying the longest co-occurring sequence of words in
sequence, even if they are not consecutive (allowing for interruptions). This makes it
more flexible than strict n-gram matching.28
METEOR (Metric for Evaluation of Translation with Explicit ORdering):
METEOR addresses weaknesses in BLEU by calculating similarity based on unigram
matching but extends this to include stemmed words (e.g., "running" matches "run") and
synonyms (using WordNet). It calculates the harmonic mean of precision and recall
(weighted towards recall) and includes a penalty for poor ordering, providing a better
correlation with human judgment than BLEU in many contexts.29
3.3.2 Embedding-Based Metrics: The Semantic Shift
Traditional n-gram metrics (BLEU, ROUGE) fail to capture semantic similarity. If a
100
model generates "automobile" and the reference is "car," n-gram metrics score 0, despite the
perfect semantic match. BERTScore represents a paradigm shift toward semantic
evaluation. It uses pre-trained contextual embeddings (like BERT) to represent tokens and
calculates the cosine similarity between the embeddings of the candidate and reference
tokens.30
● Mechanism: BERTScore computes a pairwise cosine similarity matrix between all
tokens in the candidate and reference. It then performs a "greedy matching" to find the
most similar reference token for each candidate token.
● Advantage: It captures paraphrasing and semantic equivalence that surface-level
metrics miss.
● Robustness: BERTScore has been shown to correlate better with human judgments of
quality, particularly for complex generation tasks where diverse vocabulary is used.30
4. Statistical Significance and Hypothesis Testing
In rigorous quantitative AI research, reporting a higher average metric is insufficient
to claim superiority. Due to the stochastic nature of initialization, data shuffling, and non-
convex optimization landscapes, performance differences may be due to random chance.
Statistical hypothesis testing provides the framework to distinguish signal from noise.
4.1 The Problem with Simple t-tests in Cross-Validation
A common methodological error in AI research is using a standard Student's t-test
on the k scores resulting from a single run of k-fold cross-validation. This approach violates
the fundamental assumption of the t-test: the independence of samples. In 10-fold CV, the
training sets for any two folds share approximately 90% of the same data. Consequently, the
performance estimates are highly correlated. This correlation leads to a gross
underestimation of the variance and an inflated Type I error rate (detecting a significant
difference where none exists).10
101
4.2 Recommended Statistical Tests for Algorithm Comparison
To address the violation of independence, specialized statistical tests have been
developed for the specific structure of machine learning experiments.
4.2.1 McNemar’s Test
For comparing two classifiers on a single test set, McNemar’s Test is the
recommended non-parametric standard. It operates not on the accuracy scores themselves,
but on the contingency table of the two algorithms' predictions on individual test instances:
● n_{00}: Number of examples where both Model A and Model B were wrong.
● n_{11}: Number of examples where both were correct.
● n_{01}: Number of examples where Model A was correct, but Model B was wrong.
● n_{10}: Number of examples where Model A was wrong, but Model B was correct.
The test focuses entirely on the discordant pairs (n_{01} and n_{10}). Under the null
hypothesis that the classifiers have equal error rates, n_{01} and n_{10} should be roughly
equal. The test statistic is calculated as:
\chi^2 = \frac{(|n_{01} - n_{10}| - 1)^2}{n_{01} + n_{10}}
This statistic follows a Chi-Squared distribution with 1 degree of freedom. If the p-value is
below the significance threshold (usually 0.05), we reject the null hypothesis and conclude
that the models have different performance profiles.10
4.2.2 The 5x2cv Paired t-Test
For comparing algorithms using cross-validation (when a single large test set is
unavailable), Dietterich (1998) analyzed various testing schemes and recommended the
5x2cv paired t-test. This protocol involves performing five replications of 2-fold cross-
validation.
● Why 2-fold? In 2-fold CV, the training sets for the two folds are completely disjoint
102
(they are the inverse of each other). This maximizes the independence between the two
estimates in a single run.
● Why 5 repetitions? To gather enough samples for a t-test without re-introducing
excessive correlation. The resulting test statistic provides a much better balance
between Type I errors (false positives) and Type II errors (low power) compared to the
naive k-fold t-test.10
4.3 Bootstrap Confidence Intervals and Effect Sizes
Beyond binary hypothesis testing (significant vs. not significant), modern
quantitative practice emphasizes Effect Sizes and Confidence Intervals (CIs). A result can be
statistically significant but practically meaningless (e.g., an accuracy improvement of
0.001% with a massive sample size).
Effect Size (Cohen’s d): This metric quantifies the magnitude of the difference
between two groups in terms of standard deviations.
d = \frac{\bar{x}_1 - \bar{x}_2}{s_{pooled}}
A Cohen’s d of 0.2 is considered a small effect, 0.5 a medium effect, and 0.8 a large
effect. Reporting effect size helps practitioners understand whether an "improvement"
justifies the computational or financial cost of deploying the new model.33
Bootstrap Confidence Intervals: Bootstrapping the test set predictions allows for the
construction of non-parametric CIs around metrics like Accuracy, F1, or AUC.
1. Resample the test set predictions B times (e.g., 1000).
2. Recalculate the metric for each resample.
3. The distribution of these B metrics approximates the sampling distribution.
4. The 2.5th and 97.5th percentiles define the 95% CI.
If the 95% CI of the difference between two models (Metric_A - Metric_B) does not
contain zero, the result is statistically significant. This method is robust to non-normal
103
distributions of errors.15
5. Algorithmic Fairness: Quantitative Ethics
As AI systems increasingly impact human lives (hiring, lending, criminal justice),
"Fairness" has moved from an abstract ethical concept to a quantifiable engineering
constraint. Quantitative research in fairness involves defining mathematical criteria for
unbiased decision-making and measuring deviations from these criteria.
5.1 The Taxonomy of Fairness Metrics
There are three primary families of fairness metrics used in quantitative auditing.
Crucially, these metrics are often mathematically incompatible with one another, forcing
researchers to make explicit trade-offs.
5.1.1 Disparate Impact (Demographic Parity)
This metric represents the legal concept of "disparate impact." It requires that the
probability of a positive outcome (e.g., getting a loan) be equal across groups (e.g., gender,
race), regardless of the ground truth or other variables.
P(\hat{Y}=1 | A=a) = P(\hat{Y}=1 | A=b)The **Disparate Impact Ratio** is often calculated
as:\frac{P(\hat{Y}=1 | A=minority)}{P(\hat{Y}=1 | A=majority)}
A common heuristic is the "four-fifths rule" (80%), which suggests this ratio should
be at least 0.80 to avoid prima facie evidence of discrimination.35
5.1.2 Equalized Odds (Separation)
Equalized Odds requires that the model performs equally well for both groups,
specifically mandating equality of True Positive Rates (TPR) and False Positive Rates (FPR)
104
across groups.
P(\hat{Y}=1 | Y=y, A=a) = P(\hat{Y}=1 | Y=y, A=b), \quad y \in \{0, 1\}
This ensures that qualified individuals in both groups have the same chance of
selection, and unqualified individuals have the same chance of rejection. This metric allows
for different selection rates if the base rates of qualification differ between groups.36
5.1.3 Predictive Parity (Sufficiency)
This requires that the Positive Predictive Value (PPV)—the probability that a
positive prediction is actually positive—be equal across groups.
P(Y=1 | \hat{Y}=1, A=a) = P(Y=1 | \hat{Y}=1, A=b)
This is often used in risk assessment tools (like recidivism prediction) to ensure that
a "High Risk" score means the same probability of re-offense regardless of the demographic
of the defendant.37
5.2 The Impossibility Theorems
A critical insight from quantitative fairness research is the proof that, in the presence
of unequal base rates (different prevalence of the target variable Y between groups), it is
mathematically impossible to satisfy disparate impact, equalized odds, and predictive
parity simultaneously.
For example, if Group A has a higher rate of loan repayment than Group B (unequal
base rates), and we enforce Predictive Parity (equal PPV), mathematics dictates that we must
have unequal False Positive Rates or False Negative Rates, thereby violating Equalized
Odds.37 This "Impossibility Theorem" (proven by Chouldechova, Kleinberg, and others)
forces researchers to choose which definition of fairness is appropriate for their specific
context. It shifts the burden from "optimizing for fairness" to "choosing the fairness metric
105
that aligns with the specific ethical and legal goals of the application."
6. Operational Efficiency and Green AI Metrics
The quantitative evaluation of AI is incomplete without considering the
computational cost. As models scale exponentially (e.g., Large Language Models),
operational metrics become as critical as accuracy. The "Red AI" trend (buying performance
with massive compute) is giving way to "Green AI," which prioritizes efficiency.
6.1 Computational Complexity and FLOPS
Counting parameters is a poor proxy for computational cost. A sparse model with
many parameters might be faster than a dense model with fewer. The standard unit for
training cost is FLOPS (Floating Point Operations).
For Transformer-based Large Language Models (LLMs), the training cost is often
approximated using the Kaplan/OpenAI scaling laws:
C \approx 6 N D
where C is total FLOPs, N is the number of parameters, and D is the dataset size (in tokens).
This factor of 6 arises from the mechanics of the backpropagation algorithm: the forward
pass requires approximately 2 FLOPs per parameter (multiply-add), and the backward pass
requires approximately 4 FLOPs per parameter.38
Model FLOPs Utilization (MFU): This metric describes how efficiently the
hardware is being used. It is calculated as the ratio of observed FLOPs per second to the
theoretical peak FLOPs of the hardware. High MFU indicates that the training run is not
bottlenecked by memory bandwidth or communication latency.39
6.2 Latency and Throughput
In production environments, the average latency is often less important than the tail
distribution.
106
● P99 and P95 Latency: The time below which 99% or 95% of requests fall. This captures
the "tail latency" which affects the worst-case user experience. A system might have a
fast average response but occasional massive hangs; P99 exposes this.
● Throughput: The number of inferences processed per second.
● Cold vs. Warm Start: Quantitative analysis must distinguish between the latency of a
system spinning up from zero (cold start) versus one already resident in memory
(warm start). Cold starts are a critical metric for serverless AI deployments.40
6.3 Carbon Footprint and Energy Consumption
"Green AI" introduces metrics to quantify environmental impact. Tools like
CodeCarbon and CarbonTracker have been developed to measure the energy consumption
(E) of hardware (GPU/CPU/RAM) in real-time and multiply it by the Carbon Intensity (CI)
of the local power grid.42
Carbon (gCO_2) = E_{kWh} \times CI_{gCO_2/kWh}
This formula highlights a critical operational insight: the carbon footprint depends
heavily on where the model is trained. Training a Transformer in a region powered by coal
(high carbon intensity) emits vastly more CO_2 than training the exact same model in a
region powered by hydroelectricity or wind.43 Consequently, reporting the "training carbon
cost" has become a standard requirement in high-quality research papers, promoting the
migration of heavy workloads to greener data centers.6 Furthermore, researchers
distinguish between Training Carbon (a one-time fixed cost) and Inference Carbon (a
variable cost that scales with usage). For widely used models like ChatGPT, inference
carbon quickly eclipses training carbon, making model compression and distillation critical
quantitative goals.45
7. Production-Grade Experimentation: A/B Testing
and Beyond
107
While offline metrics (Accuracy, F1, AUC) are useful for model development, they
are merely proxies for business value. A model with higher AUC might fail to drive user
engagement due to latency or unexpected behavioral dynamics. The final phase of
quantitative validation occurs in production through online experimentation.
7.1 A/B Testing for ML
A/B testing (Split testing) is the gold standard for causal inference in production. It
compares a Control (current model) against a Treatment (new model) on live traffic.
Unlike standard UI A/B testing, ML A/B tests face specific challenges:
● Effect Stability: ML models may degrade over time (concept drift) or adapt to user
behavior. Short tests may miss long-term degradation.
● OEC (Overall Evaluation Criterion): The success metric must be a business KPI (e.g.,
conversion rate, watch time), not a model metric (e.g., RMSE). There is often a
disconnect between the offline loss function (minimizing error) and the online OEC
(maximizing revenue).46
● Power Analysis: Before starting the test, researchers must perform a power analysis to
determine the minimum sample size needed to detect a statistically significant
difference (Minimum Detectable Effect). This prevents "peeking" at results and
stopping early, which inflates false positives.46
7.2 Multivariate Testing and Bandits
● Multivariate Testing (MVT): Allows testing multiple variables simultaneously (e.g.,
Model Architecture + UI Layout + Copy). While it allows for the detection of interaction
effects (e.g., the model works better with Layout A than Layout B), it requires
significantly larger sample sizes than A/B testing.47
● Multi-Armed Bandits (MAB): A dynamic form of A/B testing where traffic is strictly
not split 50/50 for the duration. Instead, algorithms (like Thompson Sampling or Upper
Confidence Bound) progressively route more traffic to the winning variation during the
test. This balances Exploration (gathering data to find the winner) and Exploitation
108
(serving the best model to maximize reward). MABs are preferred when the cost of
routing users to an inferior model is high (high "regret"), such as in news
recommendation or ad targeting.46
8. Conclusion: The Holistic Quantitative Framework
Quantitative research methodology in AI is a multi-dimensional discipline that has
evolved far beyond simple accuracy reporting. It is a hybrid discipline integrating statistical
theory (for hypothesis testing and error estimation), software engineering (for
reproducibility and testing), ethics (for fairness quantification), and environmental science
(for carbon accounting).
A rigorous quantitative report in modern AI must encompass a holistic view:
1. Robust Experimental Design: Utilizing stratified cross-validation, preventing data
leakage, and documenting hyperparameter search spaces.
2. Comprehensive Metrics: Reporting precision, recall, AUC, and domain-specific
metrics (BLEU/BERTScore) rather than a single summary number.
3. Statistical Validity: Applying appropriate hypothesis tests (McNemar’s, 5x2cv t-test)
and reporting confidence intervals and effect sizes to distinguish signal from noise.
4. Operational Viability: Quantifying latency distributions (P99), FLOPs, and carbon
footprint to ensure sustainability.
5. Fairness Auditing: Measuring disparate impact and equalized odds to ensure ethical
deployment, while acknowledging the mathematical trade-offs involved.
6. Production Validation: Moving beyond offline metrics to rigorous online A/B testing
and bandit strategies.
As the field matures, the "State of the Art" is defined not just by the novelty of the
neural architecture, but by the rigor of the quantitative evidence supporting it. This
empirical framework transforms AI from a speculative art form into a measurable,
predictable, and reliable engineering science. The transition from "It works on my machine"
109
to "It works statistically significantly on the population" is the hallmark of the modern
quantitative AI researcher (see Table 9).
Table 9: Summary of Quantitative Metrics and Methods by Domain
Domain
Method/Metric
Application
Key Characteristic
Exp. Design
Stratified K-Fold CV
General ML
Preserves class
distribution; reduces
variance.
5x2cv
Algo Comparison
Low Type I error;
ideal for statistical
testing.
Bootstrapping
Uncertainty
Generates Confidence
Intervals; robust to
non-normality.
Performance
F1 Score
Classification
Harmonic mean of
Precision/Recall; good
for imbalance.
AUC-ROC
Classification
Threshold-
independent; holistic
view of
discrimination.
RMSE
Regression
Penalizes large errors
heavily; sensitive to
outliers.
NLP
BLEU
Translation
Precision-based; n-
gram overlap; brevity
penalty.
ROUGE-L
Summarization
Recall-based; Longest
Common
110
Subsequence.
BERTScore
GenAI / Semantic
Embedding cosine
similarity; captures
meaning.
Statistics
McNemar's Test
Binary Classifiers
Tests disagreement
between models; non-
parametric.
Cohen's d
Effect Size
Standardized
difference; measures
magnitude of effect.
Fairness
Disparate Impact
Bias Auditing
Ratio of positive
outcomes between
groups.
Equalized Odds
Bias Auditing
Equality of TPR and
FPR across groups.
Efficiency
FLOPS
Compute Cost
Floating Point
Operations; scaling
law estimation.
P99 Latency
Production
99th percentile
response time;
measures tail lag.
Carbon Intensity
Green AI
CO_2 emitted per
kWh; depends on
energy mix.
111
Chapter bibliographies
1. Abadio, A. K. R., Kioshima, E. S., Martins, N. F., Maigret, B., Felipe, M. S. S., Freitas,
S. M., Maigret, B., Felipe, M. S. S., Abagyan, R. A., Kufareva, I., Totrov, M., Abbasi,
M., Sadeghi-Aliabadi, H., Hassanzadeh, F., Amanlou, M., Abdelmonsef, A. H.,
Dulapalli, R., Dasari, T., Padmarao, L. S., … Olson, A. J. A. J. (2015). A
pharmacophore-based virtual screening approach for the discovery of Trypanosoma
cruzi GAPDH inhibitors (vol 5, pg 2019, 2013). J Chem Inf Model.
https://doi.org/Book_Doi 10.1002/9783527633326
2. Ackoff, R. L., Aguilera-Caracuel, J., Hurtado-Torres, N. E., Aragón-Correa, J. A., Al-
Mawali, H., Al-Shbiel, S. O., Alavi, M., Kayworth, T. R., Leidner, D. E., Albino, V.,
Garavelli, A. C., Gorgoglione, M., Alfaadhel, S., Allame, S. M., Nouri, B. A., Tavakoli,
S. Y., Shokrani, S. A. R., Allame, S. M., Nouri, B. A., … van Tartwijk, J. (2010).
Business research methods. Journal of Knowledge Management, 14(1).
3. Al, P., Tentang, G., Arab, P. B., Relevansinya Di Era, D., Lailatul, G., & Maftuhah, H.
(2025). Perspekif Al Ghazali Tentang Pendidikan Bahasa Arab Dan Relevansinya di
Era Globalisasi. Taqdir, 11(1).
4. Bris, A., Wang, T. Y. H., Zatzick, C. D., Miller, D. J. P., Fern, M. J., Cardinal, L. B.,
Gregoire, D. A., Shepherd, D. A., Westphal, J. D., Shani, G., Troster, C., van
Quaquebeke, N., Lanaj, K., Hollenbeck, J. R., Ilgen, D. R., Barnes, C. M., Harmon, S.
J., Feldman, E. R., DesJardine, M. R., … Sangiorgi, F. (2021). KNIGHTS, RAIDERS,
AND TARGETS - THE IMPACT OF THE HOSTILE TAKEOVER - COFFEE,JC,
LOWENSTEIN,L, ROSEACKERMAN,S. JOURNAL OF BANKING & FINANCE,
37(1).
5. Cantarelli, C. C., Flybjerg, B., Molin, E. J. E., & Wee, B. van. (2010). Cost Overruns in
Large-Scale Transport Infrastructure Projects. European Journal of Transport and
Infrastructure Research, 10(1).
6. Chodvadiya, J., Shukla, G., & Sharma, G. (2025). 16S-23S-5S rRNA Database: a
comprehensive integrated database of archaeal and bacterial rRNA sequences,
112
alignments, intragenomic heterogeneity, and secondary structures. Sustainability
(Switzerland), 11(1).
7. Committee, S. (1998). IEEE Standard for Software Verification and Validation IEEE
Standard for Software Verification and Validation. In IEEEInstitute of Electrical and
Electronics Engineers (Vol. 1998, Issue July).
8. Darrick Jerome Willis. (2025). Spiritual Disciplines and Christian Formative
Education: Tools to Jump-Start Spiritual Formation and Eliminate Stagnation.
Sustainability (Switzerland), 11(1).
9. Goddard, A., King, V., Garcia-Perez, A., Graham, R., Jones, C. C., Tickle, A., Wilson,
L., Press, F., Sumsion, J., Preston-Shoot, M., Elmes, J., Burke, P. P.-J., Ratke, D. C.,
Tite, P. L., Heille, G., Elias, J. W., Jagessar, M. N., Hepokoski, C., Esterline, D., …
Koswatte, I. (2005). Editorial. Times Higher Education Supplement, 20(1).
10. Gottlieb, M. M. M., Hetzroni, O., Agada, H., Leikin, M., Tan, P., Alant, E., Hughes,
E. M. E. M. E. M. E. M. M., Yakubova, G., Weng, P.-L. L., Bouck, E. C., Baron-Cohen,
S., Wheelwright, S., Skinner, R., Martin, J. J.-C., Clubley, E., Bae, Y. S., Chiang, H.-M.
M., Hickson, L., Titeca, D., … Kageyama, R. (2021). Preface. Journal of Autism and
Developmental Disorders, 11(1).
11. Hopkins, Al. (2019). Wirtschafts- und Insolvenzrecht. SSRN Electronic Journal, 1(1).
12. Hopkins, K., Mutasam, U., Ibrohim, I., Susilo, H., Puttick, S., Cullinane, A., Voss, S.,
Kruse, J. W., Kent-Schneider, I., Wei, B., Jiang, Z., Gai, L., Torres, J., Vasconcelos, C.,
Cavallo, A. A., Burgers, H., Ghezzi, A., van de Vrande, V., Erduran, S., … Schmidt
Kjærgaard, R. (2020). Exploring the Use of Lesson Study to Develop Elementary
Preservice Teachers’ Pedagogical Content Knowledge for Teaching Nature of
Science. Science and Education, 16(2).
13. Jung, E. E. J., Jung, E. E. J., Ye, J., Kurashima, Y., Kobayashi, T., Tsuda, H., Takahara,
T., Sakurai, W., Lima, L. M. M. de, Sá, L. R. de, Macambira, A. F. U. dos S., Nogueira,
J. de A., Vianna, R. P. de T., Moraes, R. M. de, Barzegar, M., Sadeghi-Niaraki, A.,
Shakeri, M., Choi, S.-M., Oduntan, G., … Buchanan, R. (2020). Abstracts. Journal of
Planning Literature, 10(1).
113
14. Khan, M., Hussain, M., Gunasekaran, A., Ajmal, M. M., Helo, P. T., Unger, S., Landis,
A., Katircioglu, K., Gooby, R., Helander, M., Drissi, Y., Chowdhary, P., Johnson, M.,
Yonezawa, T., Logrieco, A. F., Miller, J. D., Eskola, M., Krska, R., Ayalew, A., …
Singh, A. (2019). Impact of Occupational Exposure to Chemicals in Life Cycle
Assessment: A Novel Characterization Model Based on Measured Concentrations
and Labor Hours. Journal of Cleaner Production, 11(1).
15. Kolzow, D. R., Smith, C. C. C., Serrat, O., Dilie, H. M., Zeeshan, S., Ng, S. I., Ho, J. A.,
Jantan, A. H., Massey, J., Sulak, T., Sriram, R., Dennis, R. S., Bocarnea, M., Hai, T. N.,
Van, Q. N., Herbert, S. L., So-Jung Kim, Kyoung-Seok Kim, Y.-G. C., Guillaume, Dr.
O., Honeycutt, Dr. A., … Ingram, O. C. Jr. (2021). Unit 5 Theories of Leadership.
International Journal of Organizational Leadership, 1(1).
16. Ministry of Education. (2018). Education strategic plan 2018 - 2030. In Accra, Republic
of Ghana.
17. Muhammad, N. I., Judijanto, L., Sulistiani, I., Syarifuddin, & Syahrin, M. (2023).
STRATEGIC CONTENT MARKETING: A MULTI-CHANNEL ANALYSIS OF THE
IMPACT ON BRAND AWARENESS, CONSUMER ENGAGEMENT, AND
CONVERSION RATES IN THE CONTEXT OF EVOLVING DIGITAL
LANDSCAPES. International Journal of Society Reviews (INJOSER).
18. Rahmawati. (2025). EFEKTIVITAS PELAYANAN DUKCAPIL ONLINE PADA
DINAS KEPENDUDUKAN DAN CATATAN SIPIL DI KOTA MAKASSAR.
SKRIPSI UNIVERSITAS NEGERI MAKASSAR, 11(1).
19. Ronzon, T., Gurria, P., Carus, M., Cingiz, K., El-Meligi, A., Hark, N., Iost, S., M’barek,
R., Philippidis, G., van Leeuwen, M., Wesseler, J., Medina-Lozano, I., Grimplet, J.,
Díaz, A., Tejedor-Calvo, E., Marco, P., Fischer, M., Creydt, M., Sánchez-Hernández,
E., … Miras Ávalos, J. M. (2025). Understanding Material Recovery Facility Policies
in the Philippines: A Scoping Review. Sustainability (Switzerland), 2(1).
20. Sani, A., Maharani, A. R., & Nazimah. (2025). Peran Komunitas Dosen Dalam
Peningkatan Kualitas Pembelajaran Di Perguruan Tinggi. Journal Genta Mulia, 15(1).
21. Shimray, S. R., & Subaveerapandiyan, A. (2025). AI and workforce dynamics: a
114
bibliometric analysis of job creation, displacement and reskilling. Global Knowledge,
Memory and Communication. https://doi.org/10.1108/GKMC-12-2024-0824
22. Shivarama Krishna, K., & Sathish Kumar, K. (2015). A review on hybrid renewable.
Renewable and Sustainable Energy Reviews, 52.
23. Skyrme, D., & Amidon, D. (2010). Business research methods. Journal of Knowledge
Management, 14(1).
24. Sovia Lenny. (2006). RINGKASAN Puding merah (Gruptophyllum pictum L Griff).
Jurnal Farmasi Galenika (Galenika Journal of Pharmacy) (e-Journal), 1(1).
25. Tàpias, M. (2019). Clothing as Solar Radiation Protection. Autex Research Journal, 2(1).
26. Tursunbayeva, A., Chalutz-Ben Gal, H., Santos Fonseca, S. S., Benito, P. A., Piña
Ramírez, C., Rubeis, G., Filz, M.-A., Bosse, J. P., Herrmann, C., Cavalieri, A., Reis, J.,
Amorim, M., Santarsiero, F., Carlucci, D., Schiuma, G., Sukums, F., Wamala-Larsson,
C. W., Kisenge, R., Ravi, S. J., … Kazancoglu, Y. (2024). A Spotlight on Environmental
Sustainability in View of the European Green Deal. Sustainability (Switzerland), 16(1).
27. UU Republik Indonesia, Munawir, M., Nasional, B. S., Susanta, G., Jatimnet.com,
MajaMojokerto.net, Radarmojokerto.jawapos.com, Wijayakusuma, D. M. S.,
Nahman, A., Godfrey, L., Jacobsen, R., Buysse, J., Gellynck, X., Bayu, D., Depkes, R.,
Soekidjo Notoatmodjo, Boulanger, L., Ismoyo, I. H., Tarigan, R., … Sanyal, S. (2022).
PENENTUAN ALTERNATIF LOKASI TEMPAT PEMBUANGAN AKHIR (TPA)
SAMPAH DI KABUPATEN SIDOARJO. Energies, 15(1).
28. Verifier, S. D., & Drive, A. H. (2015). Simulink ® Verification and Validation TM
Reference. ReVision.
29. Wang, Z., Wang, D., Xu, Y., Zhou, L., & Zhou, Y. (2025). Intelligent Computing Social
Modeling and Methodological Innovations in Political Science in the Era of Large
Language Models. Journal of Chinese Political Science. https://doi.org/10.1007/s11366-
025-09917-6
30. Warhurst, A. (2002). Sustainability indicators and sustainability performance
management. Sustainability Indicators and Sustainability Performance Management,
43(43).
115
31. Waryono, T. (2019). RINGKASAN Puding merah (Gruptophyllum pictum L Griff).
Jurnal Farmasi Galenika (Galenika Journal of Pharmacy) (e-Journal), 1(1).
32. Abdellatief, M., Sultan, A. B. M., Jabar, M. A., Abdullah, R., Achsan, H. T. Y.,
Wibowo, W. C., Agarwal, S., Tomar, D., Ahmadpour, A., Mirdamadi, M., Hosseini,
J. F., Chizari, M., Al-Dosari, H., Al-Khanjari, Z. A., Kutti, N. S., Dorvlo, A. S. S.,
Ramadhan, H. A., AlQahtani, S. A., Al-Saadoon, G. M. W., … Srinath. (2014). What
are Learning Analytics? Journal of Computer Science, 3(1).
33. Dini, M. (2022). Pengaruh academic self efficacy, komitmen beragama dan
prokrastinasi terhadap ketidak jujuran akademik dimoderasi oleh social desirability
pada mahasiswa diperguruan tinggi keagamaan islam swasta. Science, 7(1).
34. Annuš, N., & Kmeť, T. (2024). Learn with M.E.—Let Us Boost Personalized Learning
in K-12 Math Education! Education Sciences, 14(7).
https://doi.org/10.3390/educsci14070773
35. Fardian, D., Suryadi, D., Prabawanto, S., & Jupri, A. (2025). Integrating Chat-GPT in
the Classroom: A Study on Linear Algebra Learning in Higher Education.
International Journal of Information and Education Technology, 15(4).
https://doi.org/10.18178/ijiet.2025.15.4.2279
36. Foy, P. (2023). Large Language Models (LLMs) in Venture Capital - This Week in VC. MLQ
AI.
37. Kulkarni, R. v., Choudhari, G., Bhavik, M., Bura, R., & Bhosale, V. (2025).
GAMIFYING EARLY MATH EDUCATION: AN INNOVATIVE AI GAME FOR
CHILDREN. Journal of Information Technology Education: Innovations in Practice, 24.
https://doi.org/10.28945/5595
38. Kumar Kanvaria, V., & Tarance Suraj, M. (2024). The Role Of AI In Mathematics
Education: Assessing The Effects Of An “Auto Draw” Webtool On Middle-Level
Achievement. The Online Journal of Distance Education and E-Learning, 12(1).
39. Maguey-González, J. A., Nava-Ramírez, M. de J., Gómez-Rosales, S., Ángeles, M. de
L., Solís-Cruz, B., Hernández-Patlán, D., Merino-Guzmán, R., Hernández-Velasco,
X., Figueroa-Cárdenas, J. de D., Vázquez-Durán, A., Hargis, B. M., Téllez-Isaías, G.,
116
Méndez-Albores, A., Angelico, R., Colombo, C., di Iorio, E., Brtnický, M., Fojt, J.,
Conte, P., … Chorover, J. (2018). Modeling of Soil Cation Exchange Capacity Based
on. Geoderma, 3(4).
40. Muñoz-Olano, J. F., Hurtado-Parrado, C., Troll, E. S., Friese, M., Loschelder, D. D.,
Alblwi, A., McAlaney, J., al Thani, D. A. S., Phalp, K., Ali, R., Elhai, J. D., Sapci, O.,
Yang, H., Amialchuk, A., Rozgonjuk, D., Montag, C., Bäulke, L., Daumiller, M.,
Dresel, M., … Cevallos Pozo, G. (2020). A review of web-based support systems for
students in higher education. International Journal of Environmental Research and Public
Health, 10(1).
41. Pérez-Santín, E., De-La-fuente-valentín, L., García, M. G., Bravo, K. A. S., Hernández,
F. C. L., & Sánchez, J. I. L. (2023). Applicability domains of neural networks for
toxicity prediction. In AIMS Mathematics (Vol. 8, Issue 11).
https://doi.org/10.3934/math.20231426
42. Saragih, B., & , Ika Karyati, D. S. (2010). Pengaruh Pewarna Ekstrak Cair Alami
Bawang Tiwai (Eleutherine americana Merr) Terhadap Mutu Selai Kulit Pisang
Kepok (Musa paradisiaca Linn). Jurnal Tenknologi Pertanian, 6(2).
43. Sart, G. (2017). HOW DOES THE MAKER MOVEMENT AFFECT THE SCHOOL
ADMINISTRATIONS’ ROLES ON IMPROVING THE STEM EDUCATION?
INTED2017 Proceedings, 1. https://doi.org/10.21125/inted.2017.2452
44. Sentioningtias, I. (2025). Pemanfaatan Aplikasi Math Kids Untuk Mengembangkan
Berhitung Anak Usia Dini Di TK Islam Nrhasanah Bandar Lampung. In
Sustainability (Switzerland) (Vol. 11, Issue 1).
45. Starrett, A., Yow, J., Lotter, C., Irvin, M. J., Adams, P., Guillemot, F., Lacroix, F.,
Nocus, I., Assen, J. H. E., Otting, H., Jin, X., Tigelaar, D., van der Want, A., Admiraal,
W., Peschl, H., Deng, C., Larson, N., Kuno, C. B., Hein, S., … León, J. (2022). Editorial
Board. Teaching and Teacher Education, 90(3).
46. van Merrienboer, J. J. G., Paas, F., Traxler, A., Brewe, E., Hazari, Z., Cass, C., Beattie,
C., Kelly, A. M., Wilcox, B. R., Lewandowski, H. J., Bao, L., Xiao, Y., Koenig, K., Han,
J., Eisinga, R., Grotenhuis, M., Pelzer, B., Cronbach, L. J., Butts, M. M., … VanLehn,
117
K. A. (2016). Evaluation of Colorado Learning Attitudes about Science Survey.
Physical Review Physics Education Research, 12(2).
47. Vicente, N. R. (2010). Sistemas agroflorestais sucessionais como estratégia de uso e
conservação de recursos florestais em zonas ripárias da Microbacia Arroio Primeiro
de Janeiro, Anchieta-SC. Journal of Mental Science, 7(37).
118
Conclusion
To do research in Articial Intelligence (AI) and Data Science, we must
rst understand the nature of the knowledge that these disciplines produce.
Unlike traditional programming, where knowledge is explicit and encoded by
humans, modern AI infers knowledge from observation. This chapter explores
that paradigm shift and its implications for scientic validity.
Historically, classical computing operated under a deductive logic. The
programmer acted as a legislator, writing explicit rules (if-then) that the system
blindly obeyed. If we wanted to detect spam, we wrote: "If the email contains the
word 'free' and 'urgent', then it is spam."
Articial Intelligence, specically Machine Learning, reverses this process
towards inductive logic.
- Traditional approach: data + rules = answers.
- Machine Learning Approach: Data + Answers = Rules.
In research, this implies that we no longer design the solution directly; we
design the mechanism that will nd the solution. Epistemologically, this is risky:
the model can nd paerns that are statistically correct in the training data, but
conceptually false in the real world (overing). AI research means auditing these
learned "rules" to ensure that they represent valid knowledge and not mere
numerical matches.
A common mistake in early childhood researchers is to confuse the dataset
with reality. In the methodology of science, we must remember that data is
always a reduction of reality.
119
- Representation Bias: If we train a facial recognition model only with images
of light-skinned people, the model has not learned to "see humans", it has
learned to "see clear pixel paerns".
- The fallacy of objectivity: It is often believed that "data does not lie".
However, data collection, cleaning, and labeling are subjective processes
fraught with human decisions.
Deep learning models often act as "black boxes". We can observe the input
and output, but the internal transformations are so complex (millions of
parameters) that they are unintelligible to human cognition. This raises a
fundamental question for further research on the subject: Is it science if I can't
explain how it works?
With Big Data, it is easy to nd spurious correlations. If we analyze enough
variables, we will nd by pure chance that "cheese consumption per capita"
correlates perfectly with "the number of people who died entangled in their
sheets" (a classic example of spurious correlation).
Algorithms are machines for looking for correlations, not for
understanding causes. The researcher must provide the theoretical framework.
Scientic methodology demands that we move from the question "What will
happen?" (prediction) to "What if...?" (causal inference). Without this step, our
models are fragile in the face of any change in the environment.
Research in AI and Data Science is not just software engineering; it is a
form of radical empiricism. As researchers, our responsibility does not end with
obtaining high accuracy. Our job is to validate that this accuracy comes from real,
causal and generalizable paerns, and not from hidden biases in the data or
statistical coincidences. In the following chapters, we will see how to translate
these philosophical principles into a concrete experimental design.
120
Bibliography
Abadio, A. K. R., Kioshima, E. S., Martins, N. F., Maigret, B., Felipe, M. S. S., Freitas, S.
M., Maigret, B., Felipe, M. S. S., Abagyan, R. A., Kufareva, I., Totrov, M., Abbasi, M.,
Sadeghi-Aliabadi, H., Hassanzadeh, F., Amanlou, M., Abdelmonsef, A. H., Dulapalli, R.,
Dasari, T., Padmarao, L. S., … Olson, A. J. A. J. (2015). A pharmacophore-based virtual
screening approach for the discovery of Trypanosoma cruzi GAPDH inhibitors (vol 5,
pg 2019, 2013). J Chem Inf Model. hps://doi.org/Book_Doi 10.1002/9783527633326
Abdellatief, M., Sultan, A. B. M., Jabar, M. A., Abdullah, R., Achsan, H. T. Y., Wibowo,
W. C., Agarwal, S., Tomar, D., Ahmadpour, A., Mirdamadi, M., Hosseini, J. F., Chizari,
M., Al-Dosari, H., Al-Khanjari, Z. A., Kui, N. S., Dorvlo, A. S. S., Ramadhan, H. A.,
AlQahtani, S. A., Al-Saadoon, G. M. W., … Srinath. (2014). What are Learning Analytics?
Journal of Computer Science, 3(1).
Abdul Rahman, H., Masri, N. A., Husaini, A., Shaharuddin, M. Y., & Zaim, S. N. N. (2025).
Integration of Machine Learning and Articial Intelligence Techniques for Qualitative
Research: The Rise of New Research Paradigms. International Journal of Qualitative
Methods , 24. hps://doi.org/10.1177/16094069251347545
Abdul-Rahman, G., Zwier, A., & Haleem, N. (2025). A systematic literature
review on the role of articial intelligence in citizen science. Discover Articial
Intelligence, 5(1). hps://doi.org/10.1007/s44163-025-00437-z
Abisoye, A. (2024). A Conceptual Framework for Integrating Articial Intelligence into
STEM Research Methodologies for Enhanced Innovation. International Journal of Future
Engineering Innovations, 1(1). hps://doi.org/10.54660/ijfei.2024.1.1.56-66
Abu Sharshouh, A. (2025). The Use of Articial İntelligence in Accounting and Auditing
TT - Muhasebe ve Denetim Alanında Yapay Zekâ Kullanımı. Karadeniz Ekonomi
Araştırmaları Dergisi, 6(1).
Access, O., Adams, W. C., Alfonita, F., Alfred, M., Briggs, S., Aloisi, A., Amorim, H.,
Moda, F., Anderson, D. N., Anwar, M. A., Graham, M., Apouey, B., Roulet, A., Solal, I.,
Stabile, M., Background, P., Bajwa, U., Gastaldo, D., di Ruggiero, E., … ZACCARIA, M.
L. (2019). No 主観的健康感を中心とした在宅高齢者における 健康関連指標に関する共
121
分散構造分析 Title. Economic and Labour Relations Review, 14(1).
Acko, R. L., Aguilera-Caracuel, J., Hurtado-Torres, N. E., Aragón-Correa, J. A., Al-
Mawali, H., Al-Shbiel, S. O., Alavi, M., Kayworth, T. R., Leidner, D. E., Albino, V.,
Garavelli, A. C., Gorgoglione, M., Alfaadhel, S., Allame, S. M., Nouri, B. A., Tavakoli, S.
Y., Shokrani, S. A. R., Allame, S. M., Nouri, B. A., … van Tartwijk, J. (2010). Business
research methods. Journal of Knowledge Management, 14(1).
Acomi, N., Surugiu, G., Raicu, G., & Stanca, C. (2025). A Thematic Review of Port
Services and Emission Reduction Strategies Using ATLAS.ti. TransNav, 19(2).
hps://doi.org/10.12716/1001.19.02.14
Aggarwal, K., Mijwil, M. M., Sonia, Al-Mistarehi, A. H., Alomari, S., Gök, M.,
Zein Alaabdin, A. M., & Abdulrhman, S. H. (2022). Has the Future Started? The
Current Growth of Articial Intelligence, Machine Learning, and Deep Learning.
Iraqi Journal for Computer Science and Mathematics, 3(1).
hps://doi.org/10.52866/ijcsm.2022.01.01.013
Ahmadpour, M. (2024). Emerging Approaches in Enhancing Mental and Physical
Health of the Elderly: Integration of Strength Training, Mindfulness, and
Articial Intelligence. Journal of Psychological Dynamics in Mood Disorders, 3(5).
hps://doi.org/10.61838/kman.pdmd.3.5.19
Al, P., Tentang, G., Arab, P. B., Relevansinya Di Era, D., Lailatul, G., & Maftuhah, H.
(2025). Perspekif Al Ghazali Tentang Pendidikan Bahasa Arab Dan Relevansinya di Era
Globalisasi. Taqdir, 11(1).
Alshadidi, A. A. F., Alshahrani, A. A., Aldosari, L. I. N., Chaturvedi, S., Saini, R.
S., Hassan, S. A. bin, Cicciù, M., & Minervini, G. (2023). Investigation on the
Application of Articial Intelligence in Prosthodontics. In Applied Sciences
(Swierland) (Vol. 13, Issue 8). hps://doi.org/10.3390/app13085004
Anis, S., & French, J. A. (2023). Ecient, Explicatory, and Equitable: Why Qualitative
Researchers Should Embrace AI, but Cautiously. In Business and Society (Vol. 62, Issue 6).
hps://doi.org/10.1177/00076503231163286
Annus, N., & Kmet, T. (2024). Learn with M.E.-Let Us Boost Personalized Learning in K-
12 Math Education! EDUCATION SCIENCES, 14(7).
122
Arslantaş, S. (2024). Articial intelligence and big data from digital health applications:
publication trends and analysis. Journal of Health Organization and Management.
hps://doi.org/10.1108/JHOM-06-2024-0241
Asaduzzaman, M., & Flowra, M. T. (2018). Resurgence of infectious diseases due
to forced migration: is planetary health and One Health action synergistic? The
Lancet Planetary Health, 2(10).
Asiva Noor Rachmayani, Inaam Akhtar, M., Njie, B., Asimiran, S., Alase, A., Etikan, I.,
Laughlin, R., Francis, J. J., Johnston, M., Robertson, C., Glidewell, L., Entwistle, V., Eccles,
M. P., Grimshaw, J. M., Alhojailan, M. I., Ibrahim, M., Braun, V., Clarke, V., Taherdoost,
H., … Zairul, M. (2019). No 主観的健康感を中心とした在宅高齢者における 健康関連指
標に関する共分散構造分析 Title. ISRA International Journal of Islamic Finance, 8(1).
Avula, S., Chadwick, G., Arguello, J., Capra, R., Lippold, D., Bana e Costa, C. A., de Corte,
J.-M., Vansnick, J.-C., Linstone, H. A., Carver, C. S., Scheier, M. F., Segerstrom, S. C.,
Berger, G., Armand, L., Bresard, S., Grimanelli, P., Landucci, A., Schwar, B., Bolger,
F., … Birbilis, G. (2015). Editorial Board. Technological Forecasting and Social Change, 77(1).
Axiotis, E., Kontogiannis, A., Kalpouakis, E., & Giannakopoulos, G. (2021). A
personalized machine-learning-enabled method for ecient research in
ethnopharmacology. The case of the southern balkans and the coastal zone of asia minor.
Applied Sciences (Swierland), 11(13). hps://doi.org/10.3390/app11135826
Benedeti Rosa, M., & Kubota, L. C. (2025). Articial intelligence: where does Brazil stand
in global scientic production and what are the main technical determinants of adoption
by Brazilian companies. EconomiA. hps://doi.org/10.1108/ECON-01-2025-0010
Bhardwaj, M., Kumar, P., & Singh, A. (2025). Bibliometric review of digital
transformation in agriculture: innovations, trends and sustainable futures. Journal of
Agribusiness in Developing and Emerging Economies. hps://doi.org/10.1108/JADEE-02-
2025-0083
Bhatia, A., Chandani, A., Atiq, R., Mehta, M., & Divekar, R. (2021). Articial intelligence
in nancial services: a qualitative research to discover robo-advisory services. Qualitative
Research in Financial Markets, 13(5). hps://doi.org/10.1108/QRFM-10-2020-0199
123
Bock, T., & von der Oelsni, D. (2025). Leadership-competences in the era of
articial intelligence – a structured review. In Strategy and Leadership (Vol. 53,
Issue 3). hps://doi.org/10.1108/SL-09-2024-0100
Boukrouh, I., & Azmani, A. (2024). ARTIFICIAL INTELLIGENCE
APPLICATIONS IN E-COMMERCE: A BIBLIOMETRIC STUDY FROM 1995 TO
2023 USING MERGED DATA SOURCES. International Journal of Professional
Business Review, 9(4). hps://doi.org/10.26668/businessreview/2024.v9i4.4537
Brailas, A. (2025). Articial Intelligence in Qualitative Research: Beyond Outsourcing
Data Analysis to the Machine. Psychology International, 7(3).
hps://doi.org/10.3390/psycholint7030078
Bris, A., Wang, T. Y. H., Zaick, C. D., Miller, D. J. P., Fern, M. J., Cardinal, L. B., Gregoire,
D. A., Shepherd, D. A., Westphal, J. D., Shani, G., Troster, C., van Quaquebeke, N., Lanaj,
K., Hollenbeck, J. R., Ilgen, D. R., Barnes, C. M., Harmon, S. J., Feldman, E. R., DesJardine,
M. R., … Sangiorgi, F. (2021). KNIGHTS, RAIDERS, AND TARGETS - THE IMPACT OF
THE HOSTILE TAKEOVER - COFFEE,JC, LOWENSTEIN,L, ROSEACKERMAN,S.
JOURNAL OF BANKING & FINANCE, 37(1).
Brondani, M., Alves, C., Ribeiro, C., Braga, M. M., Garcia, R. C. M., Ardenghi, T., &
Paanaporn, K. (2024). Articial intelligence, ChatGPT, and dental education:
Implications for reective assignments and qualitative research. Journal of Dental
Education, 88(12). hps://doi.org/10.1002/jdd.13663
Buchkremer, R., Demund, A., Ebener, S., Gampfer, F., Jagering, D., Jurgens, A.,
Klenke, S., Krimpmann, D., Schmank, J., Spiekermann, M., Wahlers, M., &
Wiepke, M. (2019). The Application of Articial Intelligence Technologies as a
Substitute for Reading and to Support and Enhance the Authoring of Scientic
Review Articles. In IEEE Access (Vol. 7).
hps://doi.org/10.1109/ACCESS.2019.2917719
Budiman, A. S., Putri, D. N. N., Candra, H., Djuana, E., Sari, T. K., Aji, D. P. B., Putri, L.
R., Sitepu, E., Speaks, D., & Pasang, T. (2024). Crack Catcher AI – Enabling smart fracture
mechanics approaches for damage control of thin silicon cells or wafers. Solar Energy
Materials and Solar Cells, 273. hps://doi.org/10.1016/j.solmat.2024.112927
124
Cafuta, D., Dodig, I., Cesar, I., & Kramberger, T. (2021). Developing a modern
greenhouse scientic research facility — A case study. Sensors, 21(8).
hps://doi.org/10.3390/s21082575
Candaika, P. (2022). Gambaran Kesiapan Rumah Sakit Umum Daerah (RSUD)
Menghadapi Kebijakan Kelas Rawat Inap Standar (KRIS) Tahun 2022. Science, 7(1).
Cantarelli, C. C., Flybjerg, B., Molin, E. J. E., & Wee, B. van. (2010). Cost Overruns in
Large-Scale Transport Infrastructure Projects. European Journal of Transport and
Infrastructure Research, 10(1).
Casanova Pistón, A., & Martínez Domínguez, M. (2024). Producción cientíca sobre
Inteligencia Articial y educación: un análisis cienciométrico. Hachetetepé. Revista
Cientíca de Educación y Comunicación, 28.
hps://doi.org/10.25267/hachetetepe.2024.i28.1102
Chango, X., Flor-Unda, O., Gil-Jiménez, P., & Gómez-Moreno, H. (2024).
Technology in Forensic Sciences: Innovation and Precision. In Technologies (Vol.
12, Issue 8). hps://doi.org/10.3390/technologies12080120
Cheekaty, S., & Muneeswari, G. (2024). Exploring Sparse Gaussian Processes for
Bayesian Optimization in Convolutional Neural Networks for Autism Classication.
IEEE Access, 12. hps://doi.org/10.1109/ACCESS.2024.3351168
Chen, S., & Garcia, A. (2017). Embodiment of Cognition: Exploring the Interplay
between Body and Mind. Studies in Science of Science | ISSN: 1003-2053, 35(9).
Cheng, J. C. P., Chen, W., Chen, K., Wang, Q., Sadri, H., Karki, B. R., Basnet, S., Xiang, J.,
Montoya, J., Porras, J., Ucar, A., Karakose, M., Kırımça, N., Abdelalim, A. M., Essawy,
A., Alnaser, A. A., Shibeika, A., Sherif, A., van Roosmale, S., … Liu, G. (2024). applied
sciences Articial Intelligence for Predictive Maintenance Applications: Journal of
Building Engineering, 12(1).
Chodvadiya, J., Shukla, G., & Sharma, G. (2025). 16S-23S-5S rRNA Database: a
comprehensive integrated database of archaeal and bacterial rRNA sequences,
alignments, intragenomic heterogeneity, and secondary structures. Sustainability
(Swierland), 11(1).
125
Choudrie, J., Manandhar, N., Castro, C., & Obuekwe, C. (2023). Hey Siri, Google! Can
you help me? A qualitative case study of smartphones AI functions in SMEs.
Technological Forecasting and Social Change, 189.
hps://doi.org/10.1016/j.techfore.2023.122375
Christou, P. (2023). How to Use Articial Intelligence (AI) as a Resource, ow
Methodological and Analysis Tool in Qualitative Research? Qualitative Report, 28(7).
hps://doi.org/10.46743/2160-3715/2023.6406
Christou, P. A. (2023a). A Critical Perspective Over Whether and How to Acknowledge
the Use of Articial Intelligence (AI) in Qualitative Studies. Qualitative Report, 28(7).
hps://doi.org/10.46743/2160-3715/2023.6407
Christou, P. A. (2023b). The Use of Articial Intelligence (AI) in Qualitative Research for
Theory Development. Qualitative Report, 28(9). hps://doi.org/10.46743/2160-
3715/2023.6536
Colina Vargas, A. M., Espinoza Mina, M. A., Catalán, L. L., & Catalán, B. L. (2024).
Computational intelligence, educational robotics, and articial intelligence in the
educational eld. A bibliometric study and thematic modelling. International Journal of
Educational Research and Innovation, 2024(22). hps://doi.org/10.46661/ijeri.10369
Commiee, S. (1998). IEEE Standard for Software Verication and Validation IEEE
Standard for Software Verication and Validation. In IEEEInstitute of Electrical and
Electronics Engineers (Vol. 1998, Issue July).
Cricchio, J. (2025). Balancing openness and ownership: open innovation strategies for AI
development. European Journal of Innovation Management. hps://doi.org/10.1108/EJIM-
04-2024-0470
Dai, Y., Lai, S., Lim, C. P., & Liu, A. (2023). ChatGPT and its impact on research
supervision: Insights from Australian postgraduate research students. Australasian
Journal of Educational Technology, 39(4). hps://doi.org/10.14742/ajet.8843
Darrick Jerome Willis. (2025). Spiritual Disciplines and Christian Formative Education:
Tools to Jump-Start Spiritual Formation and Eliminate Stagnation. Sustainability
(Swierland), 11(1).
de Seta, G., Pohjonen, M., & Knuutila, A. (2024). Synthetic ethnography: Field devices
for the qualitative study of generative models. Big Data and Society, 11(4).
126
hps://doi.org/10.1177/20539517241303126
Demidenko, S. Y., & Romanovskiy, N. v. (2023). THEORETICAL UNDERSTANDING
OF SOCIAL REALITY: THE 25th ANNIVERSARY OF KHARCHEV READINGS.
Sotsiologicheskie Issledovaniya, 2023(9). hps://doi.org/10.31857/S013216250027774-4
Dini, M. (2022). Pengaruh academic self ecacy, komitmen beragama dan prokrastinasi
terhadap ketidak jujuran akademik dimoderasi oleh social desirability pada mahasiswa
diperguruan tinggi keagamaan islam swasta. Science, 7(1).
Djakfar Musthafa, F. A. (2024). Penggunaan Articial Intelligence (AI) dalam
Pembelajaran: Fenomena Transformasi Otoritas Pengetahuan di Kalangan Mahasiswa.
Journal of Contemporary Islamic Education, 4(1). hps://doi.org/10.25217/jcie.v4i1.4386
Durrani, B. A., Lecturer, S., Ivanauskien, N., Volung, J., Rahmani-Nejad, L., Firoozbakht,
Z., Taghipoor, A., Arokiasamy, A. R. A., Pourdehghan, A., Tjan, S., Sukamto, R.,
Lumintan, D. B., Dhurup, M., Mani, C., Dumasi, T., Kheng, L. L., Mahamad, O.,
Ramayah, T., Mosahab, R., … Wirawan, V. (2014). Implementation of E-Government in
Welcoming the Contemporary Industrial Revolution 4.0 Era in Indonesia. International
Journal of Marketing Studies, 1(1).
Educause. (2018). Horizon report: 2018 higher education edition. Retrieved from
EDUCAUSE Learning Initiative and The New Media Consortium. Liila.
Espinel, R., Herrera-Franco, G., Rivadeneira García, J. L., & Escandón-Panchana,
P. (2024). Articial Intelligence in Agricultural Mapping: A Review. In Agriculture
(Swierland) (Vol. 14, Issue 7). hps://doi.org/10.3390/agriculture14071071
Evolution Analysis and Prediction of Scientic Topic Popularity in the Field of LIS.
(2021). Documentation, Information and Knowledge, 38(2).
hps://doi.org/10.13366/j.dik.2021.02.035
Fahyuni, E. F., Wasis, Bandono, A., & Arin, M. B. U. B. (2020). Integrating islamic values
and science for millennial students’ learning on using seamless mobile media. Jurnal
Pendidikan IPA Indonesia, 9(2). hps://doi.org/10.15294/jpii.v9i2.23209
Fardian, D., Suryadi, D., Prabawanto, S., & Jupri, A. (2025). Integrating Chat-GPT in the
Classroom: A Study on Linear Algebra Learning in Higher Education. International
Journal of Information and Education Technology, 15(4).
hps://doi.org/10.18178/ijiet.2025.15.4.2279
127
Fieldhouse, J. K., Ge, J., Randhawa, N., Wolking, D., Genovese, B. N., Mazet, J. A. K., &
Desai, A. (2025). The intersection of articial intelligence with qualitative or mixed
methods for communicable disease research: a scoping review. In Public Health (Vol. 248).
hps://doi.org/10.1016/j.puhe.2025.105961
Fil, N., Slisarenko, R., Deineko, Z., & Morozova, L. (2025). Trends in Articial
Intelligence Research on Education: Topic Modeling Using Latent Dirichlet
Allocation. Bulletin of Kharkov National Automobile and Highway University, 108.
hps://doi.org/10.30977/bul.2219-5548.2025.108.0.17
Flores, L. G. R., & Mejía Rivera, K. A. (2024). ARTIFICIAL INTELLIGENCE IN
QUALITATIVE RESEARCH: BIBLIOMETRIC ANALYSIS OF SCIENTIFIC
PRODUCTION INDEXED IN SCOPUS. New Trends in Qualitative Research, 20(4).
hps://doi.org/10.36367/ntqr.20.4.2024.e1116
Foy, P. (2023). Large Language Models (LLMs) in Venture Capital - This Week in VC. MLQ AI.
Friedman, C., Owen, A., & VanPuymbrouck, L. (2025). Should ChatGPT help with my
research? A caution against articial intelligence in qualitative analysis. Qualitative
Research, 25(5). hps://doi.org/10.1177/14687941241297375
Georgiev, G. (2021). THE BREAKDOWN OF THE PUBLIC-PRIVATE DIVIDE IN
SECURITIES LAW: CAUSES, CONSEQUENCES, AND REFORMS GEORGE.
Frontiers in Neuroscience, 14(1).
Göçer, Ö., Hua, Y., Göçer, K., Gunhan, S., Moaveni, S., Chou, K. C., Charlson, J., Chinyio,
E., Zhu, Y. Q., Zuo, J. J., Korpela, J., Kerosuo, H., Abuelmaai, A., Ahmed, V., Jing, X.,
Wang, Q. K. Q., Li, Y. Y. Y., Zhao, D., McCoy, A. P., … Wu, W. W. Y. (2014). A Ship in a
Box. Applied Mechanics and Materials, 29(2).
Goddard, A., King, V., Garcia-Perez, A., Graham, R., Jones, C. C., Tickle, A., Wilson, L.,
Press, F., Sumsion, J., Preston-Shoot, M., Elmes, J., Burke, P. P.-J., Ratke, D. C., Tite, P. L.,
Heille, G., Elias, J. W., Jagessar, M. N., Hepokoski, C., Esterline, D., … Koswae, I. (2005).
Editorial. Times Higher Education Supplement, 20(1).
Gong, J., Zhao, Z., Niu, X., Ji, Y., Sun, H., Shen, Y., Chen, B., & Wu, B. (2025). AI reshaping
life sciences: intelligent transformation, application challenges, and future convergence
in neuroscience, biology, and medicine. In Frontiers in Digital Health (Vol. 7).
hps://doi.org/10.3389/fdgth.2025.1666415
128
Golieb, M. M. M., Heroni, O., Agada, H., Leikin, M., Tan, P., Alant, E., Hughes, E. M.
E. M. E. M. E. M. M., Yakubova, G., Weng, P.-L. L., Bouck, E. C., Baron-Cohen, S.,
Wheelwright, S., Skinner, R., Martin, J. J.-C., Clubley, E., Bae, Y. S., Chiang, H.-M. M.,
Hickson, L., Titeca, D., … Kageyama, R. (2021). Preface. Journal of Autism and
Developmental Disorders, 11(1).
Govindan, K., Cheng, T. C. E., Mishra, N., Shukla, N., Pei-Chen, L., Arya, V., Sharma, P.,
Singh, A., de Silva, P. T. M., Chateld, A. T., Reddick, C. G., Karlsson, P.-M., Lal, S.,
Rexhausen, D., Ajak, A. D., Lilford, E., Topal, E., Zhang, Y. Y., Fu, Z., … Perez-Franco, R.
(2018). Industry Surveys IT Consulting & Other Services. Journal of Cleaner Production,
25(1).
Goyanes, M., Lopezosa, C., & Jordá, B. (2025). Thematic analysis of interview data with
ChatGPT: designing and testing a reliable research protocol for qualitative research.
Quality and Quantity, 59(6). hps://doi.org/10.1007/s11135-025-02199-3
Guizzardi, G., & Guarino, N. (2024). Explanation, semantics, and ontology. Data
and Knowledge Engineering, 153. hps://doi.org/10.1016/j.datak.2024.102325
Guliev, I. A., Mammadov, A., & Ibrahimli, K. (2024). The Use of Articial Intelligence
Technologies in Energy and Climate Security. Review of Business and Economics Studies,
12(4). hps://doi.org/10.26794/2308-944X-2024-12-4-58-71
Gunewardena, D., Seck, A., Warnecke, T., Green, E. P., Blaman, C., Jamison, J., Annan,
J., Dvouletý, O., Pepelasis, I. S., Protogerou, A., Krasniqi, B. A., Williams, N., Osirim, M.
J., Tobias, J. M., Boudreaux, K. C., Boerill, L. C., Free, M., Demirgüc-Kunt, A., Klapper,
L. F., … Naszályi, P. (2018). Abstracts. Journal of Planning Literature, 29(2).
Guo, Z. (2024). Innovative Functionality of the Scientic and Technical Information
System of China. Visnyk of Kharkiv State Academy of Culture, 65.
hps://doi.org/10.31516/2410-5333.065.04
Gupta, S., Mahmood, A., Shey, P., Adeboye, A., & Ramprasad, R. (2024). Data
extraction from polymer literature using large language models. Communications
Materials, 5(1). hps://doi.org/10.1038/s43246-024-00708-9
Güzer, B., Caner, H., Fathoni, A. F., van der Mescht, H., Whiaker, J. K., Staker, H., Horn,
M. B., Crawford, R., Richey, R. C., Christensen, C. M., Horn, M. B., Staker, H., Thoms, C.
L. v, Chen, J. J., Pey, J., Albiladi, W. S., Alshareef, K. K., Zhang, W., Zhu, C., … Province,
129
Q. (2019). No 主観的健康感を中心とした在宅高齢者における 健康関連指標に関する共
分散構造分析 Title. Computers and Education, 7(1).
Harahulia, S., & Symonenko, T. (2025). RESEARCH SUPPORT SERVICES IN DIGITAL
HUMANITIES:APPROACHES, TECHNOLOGIES, TOOLS. Manuscript and Book
Heritage of Ukraine, 2025(2). hps://doi.org/10.15407/rksu.37.189
Hogg, H. D. J., Al-Zubaidy, M., Keane, P. A., Hughes, G., Beyer, F. R., & Maniatopoulos,
G. (2023). Evaluating the translation of implementation science to clinical articial
intelligence: a bibliometric study of qualitative research. In Frontiers in Health Services
(Vol. 3). hps://doi.org/10.3389/frhs.2023.1161822
Hogg, H. D. J., Al-Zubaidy, M., Talks, J., Denniston, A. K., Kelly, C. J., Malawana, J.,
Papoutsi, C., Teare, M. D., Keane, P. A., Beyer, F. R., & Maniatopoulos, G. (2023).
Stakeholder Perspectives of Clinical Articial Intelligence Implementation: Systematic
Review of Qualitative Evidence. In Journal of Medical Internet Research (Vol. 25).
hps://doi.org/10.2196/39742
Hogg, H. D. J., Sendak, M. P., Denniston, A. K., Keane, P. A., & Maniatopoulos, G. (2023).
Unlocking the potential of qualitative research for the implementation of articial
intelligence-enabled healthcare. In Journal of Medical Articial Intelligence (Vol. 6).
hps://doi.org/10.21037/jmai-23-28
Hopkins, Al. (2019). Wirtschafts- und Insolvenzrecht. SSRN Electronic Journal, 1(1).
Hopkins, K., Mutasam, U., Ibrohim, I., Susilo, H., Puick, S., Cullinane, A., Voss, S.,
Kruse, J. W., Kent-Schneider, I., Wei, B., Jiang, Z., Gai, L., Torres, J., Vasconcelos, C.,
Cavallo, A. A., Burgers, H., Ghezzi, A., van de Vrande, V., Erduran, S., … Schmidt
Kjærgaard, R. (2020). Exploring the Use of Lesson Study to Develop Elementary
Preservice Teachers’ Pedagogical Content Knowledge for Teaching Nature of Science.
Science and Education, 16(2).
Hou, J., Zheng, B., Li, H., & Li, W. (2025). Evolution and impact of the science of
science: from theoretical analysis to digital-AI driven research. In Humanities and
Social Sciences Communications (Vol. 12, Issue 1). hps://doi.org/10.1057/s41599-
025-04617-1
Inaam Akhtar, M., Njie, B., Asimiran, S., Alase, A., Etikan, I., Laughlin, R., Francis, J. J.,
130
Johnston, M., Robertson, C., Glidewell, L., Entwistle, V., Eccles, M. P., Grimshaw, J. M.,
Alhojailan, M. I., Ibrahim, M., Braun, V., Clarke, V., Taherdoost, H., SWELHEIM, T., …
Zairul, M. (2019). No 主観的健康感を中心とした在宅高齢者における 健康関連指標に関
する共分散構造分析 Title. ISRA International Journal of Islamic Finance, 8(1).
Irtyshcheva, I., Kim, V., & Kletsov, Y. (2022). MANAGING BUSINESS PROCESSES IN
THE CONDITIONS OF THE DEVELOPMENT OF THE DIGITAL ECONOMY:
GLOBAL AND NATIONAL EXPERIENCES. Baltic Journal of Economic Studies, 8(5).
hps://doi.org/10.30525/2256-0742/2022-8-5-101-107
Iurasova, O., Ivashko, L., Maksymova, J., & Maksymov, O. (2025). INNOVATION
TRENDS: IMPACT ON DEVELOPMENT OF FINANCIAL SECTOR. Смарт-Економіка,
Підприємництво Та Безпека, 3(1). hps://doi.org/10.60022/sis.3.(01).1
Jain, H., & KUMAR, Prof. D. (2025). Articial Intelligence as a Catalyst for Supply Chain
Evolution: A Multi-Case Thematic Study. INTERNATIONAL JOURNAL OF SCIENTIFIC
RESEARCH IN ENGINEERING AND MANAGEMENT, 09(09).
hps://doi.org/10.55041/ijsrem52737
Jeon, Y. (2021). A Study on the Ethics of the Concept and Purpose of Culture Technology
in the Post-COVID Era. Society for International Cultural Institute, 14(1).
hps://doi.org/10.34223/jic.2021.14.1.1
Jiang, Y., Ko-Wong, L., & Valdovinos Gutierrez, I. (2025). The Feasibility and
Comparability of Using Articial Intelligence for Qualitative Data Analysis in Equity-
Focused Research. In Educational Researcher (Vol. 54, Issue 3).
hps://doi.org/10.3102/0013189X251314821
Jirandehi, A. B., & Sajadi, S. M. (2024). Co-Citation Analysis and Co-Occurrence Analysis
of Resilience Keywords in StartUps. Scientometrics Research Journal, 10(2).
hps://doi.org/10.22070/rsci.2024.17648.1667
John, A. O., & Chukwumerije, T. (2014). Perceptions of Accountants on Factors Aecting
Auditor’s Independence in Nigeria. International Journal of Research in Commerce, IT &
Management, 2(11).
Jung, E. E. J., Jung, E. E. J., Ye, J., Kurashima, Y., Kobayashi, T., Tsuda, H., Takahara, T.,
Sakurai, W., Lima, L. M. M. de, Sá, L. R. de, Macambira, A. F. U. dos S., Nogueira, J. de
A., Vianna, R. P. de T., Moraes, R. M. de, Barzegar, M., Sadeghi-Niaraki, A., Shakeri, M.,
131
Choi, S.-M., Oduntan, G., … Buchanan, R. (2020). Abstracts. Journal of Planning Literature,
10(1).
Jungmeier, G. (2017). The Biorenery Fact Sheet. The International Journal of Life
Cycle Assessment, 23(1).
Júnior, G. H. K., & Goschalg-Duque, C. (2023). Information architecture applied
on natural language processing: a proposal Information Science contributions on
data preprocessing for training and learning of articial neural networks. Revista
Digital de Biblioteconomia e Ciencia Da Informacao, 21.
hps://doi.org/10.20396/rdbci.v21i00.8671396/30919
Kang, R., Xuan, Z., Tong, L., Wang, Y., Jin, S., & Xiao, Q. (2025). Nurse Researchers’
Experiences and Perceptions of Generative AI: Qualitative Semistructured Interview
Study. Journal of Medical Internet Research, 27. hps://doi.org/10.2196/65523
Kantor, J. (2024). Best practices for implementing ChatGPT, large language models, and
articial intelligence in qualitative and survey-based research. In JAAD International (Vol.
14). hps://doi.org/10.1016/j.jdin.2023.10.001
Khabsa, M., Elmagarmid, A., Ilyas, I., Hammady, H., Ouzzani, M., Kouznetsov, A.,
Japkowicz, N., Visser, E., Felizardo, K. K. R., Souza, S. R. S., Maldonado, J. C. J. C. J. C. J.
C., Salleh, N., Martins, R. M., Mendes, E. E., MacDonell, S. G., Maldonado, J. C. J. C. J. C.
J. C., Li, D., Wang, Z. Z., Wang, L., … Demner-Fushman, D. (2020). A New Method to
Enhance. Systematic Reviews, 9(1).
Khan, M., Hussain, M., Gunasekaran, A., Ajmal, M. M., Helo, P. T., Unger, S., Landis, A.,
Katircioglu, K., Gooby, R., Helander, M., Drissi, Y., Chowdhary, P., Johnson, M.,
Yonezawa, T., Logrieco, A. F., Miller, J. D., Eskola, M., Krska, R., Ayalew, A., … Singh, A.
(2019). Impact of Occupational Exposure to Chemicals in Life Cycle Assessment: A
Novel Characterization Model Based on Measured Concentrations and Labor Hours.
Journal of Cleaner Production, 11(1).
Kharimah, W. (2021). DAMPAK PENGUNDURAN DIRI AMERIKA SERIKAT
DARI PERJANJIAN JCPOA (JOINT COMPREHENSIVE PLAN OF ACTION).
Frontiers in Neuroscience, 14(1).
Khmelkov, A. (2023). INSTITUTIONAL DESIGN OF ANALYTICAL TOOLS OF THE
132
FINANCIAL CONTROL INSTITUTION. Social Economics, 66.
hps://doi.org/10.26565/2524-2547-2023-66-08
Kim, T. Y., Gang, J. Y., Oh, H. J., Ahmad, K., JianMing, Z., Ra, M., Allen, G. G., Travis,
T. A., Ramirez, C., , . . . . . ., Kyrillidou, M., Manifesto, L. S., Johnson, L.,
Smith, R., Willis, H., Levine, A., Haywood, K., Vice-chancellor, S. D., Vice-chancellor, D.
V.-S. D., … Briggs, T. (2019). Using Data Visualization and Dashboards to Support the
Decision Making of Public Library Branch Managers. Library Hi Tech, 38(1).
Kirkpatrick, A. W., McKee, J. L., McBeth, P. B., Ball, C. G., Laporta, A. T.,
Broderick, T., Leslie, T., King, D., Wright Beay, H. E., Keillor, J., Tien, H.,
de’Angelis, N., Khan, J., Marchegiani, F., Bianchi, G., Aisoni, F., Alberti, D.,
Ansaloni, L., Bi, W. L., … MOORI, S. H. (2020). Translating Scientic Evidence
into Practice using Digital Medicine and Electronic Patient Reported Outcomes.
Telemedicine and E-Health, 17(1).
Koçak, M., & Akçalı, Z. (2024). Development trends and knowledge framework
of articial intelligence (AI) applications in oncology by years: a bibliometric
analysis from 1992 to 2022. Discover Oncology, 15(1).
hps://doi.org/10.1007/s12672-024-01415-0
Kolzow, D. R., Smith, C. C. C., Serrat, O., Dilie, H. M., Zeeshan, S., Ng, S. I., Ho, J. A.,
Jantan, A. H., Massey, J., Sulak, T., Sriram, R., Dennis, R. S., Bocarnea, M., Hai, T. N., Van,
Q. N., Herbert, S. L., So-Jung Kim, Kyoung-Seok Kim, Y.-G. C., Guillaume, Dr. O.,
Honeycu, Dr. A., … Ingram, O. C. Jr. (2021). Unit 5 Theories of Leadership. International
Journal of Organizational Leadership, 1(1).
Kozinets, R. v., & Greel, U. (2024). Netnography evolved: New contexts, scope,
procedures and sensibilities. Annals of Tourism Research, 104.
hps://doi.org/10.1016/j.annals.2023.103693
Kulkarni, R. v., Choudhari, G., Bhavik, M., Bura, R., & Bhosale, V. (2025). GAMIFYING
EARLY MATH EDUCATION: AN INNOVATIVE AI GAME FOR CHILDREN. Journal of
Information Technology Education: Innovations in Practice, 24. hps://doi.org/10.28945/5595
133
Kumar Kanvaria, V., & Tarance Suraj, M. (2024). The Role Of AI In Mathematics
Education: Assessing The Eects Of An “Auto Draw” Webtool On Middle-Level
Achievement. The Online Journal of Distance Education and E-Learning, 12(1).
Lausiri, N., & Nahda, K. (2022). Pengaruh Rasio Likuiditas, Leverage Dan Protabilitas
Terhadap Kondisi Financial Distress Dengan Kepemilikan Manajerial Sebagai Variabel
Moderasi. Selekta Manajemen: Jurnal Mahasiswa & Manajemen, 1(5).
Lei, P., Duo, E. N., Gao, Y., & Zhu, X. (2025). Research on the experience of utilizing
articial intelligence in nursing learning from the perspective of nursing students – a
qualitative study. BMC Nursing, 24(1). hps://doi.org/10.1186/s12912-025-03704-7
Leong, W. D., Teng, S. Y., How, B. S., Ngan, S. L., Rahman, A. A., Tan, C. P.,
Ponnambalam, S. G., Lam, H. L., Yadav, G., Luthra, S., Jakhar, S. K., Mangla, S.
K., Rai, D. P., Tortorella, G. L., Narayanamurthy, G., Thurer, M., Dombrowski,
U., Wullbrandt, J., Fochler, S., … Gierej, S. (2019). Elimination of waste: creation
of a successful Lean colonoscopy program at an academic medical center. Surgical
Endoscopy.
Li, Z. Q., Shro, P. K., Venkataraman, R., Zhang, I. X., Gu, F., Lev, B., He, G., Kang, J. K.
H. K., Shivdasani, A., Yamada, T., Bhaacharyya, S., Nain, A. S., Lee, K. B. H., Homburg,
C., Bucerius, M., Erel, I., Liao, R. C., WEISBACH, M. S., Andrade, G., … Lys, T. Z. (2019).
CORPORATE TAKEOVERS AND PRODUCTIVITY - LICHTENBERG,FR. JOURNAL OF
BANKING & FINANCE, 37(2).
Lim, W. M. (2025). What Is Qualitative Research? An Overview and Guidelines.
Australasian Marketing Journal, 33(2). hps://doi.org/10.1177/14413582241264619
Litynska, V. (2025). MODERN METHODS OF CONDUCTING MARKETING
RESEARCH. Social Economics, 70. hps://doi.org/10.26565/2524-2547-2025-70-06
Longo, L. (2020). Empowering Qualitative Research Methods in Education with Articial
Intelligence. Advances in Intelligent Systems and Computing, 1068.
hps://doi.org/10.1007/978-3-030-31787-4_1
Luo, J., Zhuo, W., Liu, S., & Xu, B. (2024). The Optimization of Carbon Emission
Prediction in Low Carbon Energy Economy under Big Data. IEEE Access, 12.
hps://doi.org/10.1109/ACCESS.2024.3351468
134
Luo, X., Tham, Y. C., Daher, M., Bian, Z., Chen, Y., Estill, J., & group, the G.
working. (2025). Protocol for developing the reporting guideline for the use of
chatbots and other Generative Articial intelligence tools in MEdical Research
(GAMER). medRxiv.
Luongo, M., Ponticorvo, M., Ligorio, M. B., Crescenzo, P., & Ritella, G. (2024). Articial
Intelligence to Enhance Qualitative Research: Methodological Reections on a Pilot
Study. Journal of Educational, Cultural and Psychological Studies, 2024(30).
hps://doi.org/10.7358/ecps-2024-030-luon
Maguey-González, J. A., Nava-Ramírez, M. de J., Gómez-Rosales, S., Ángeles, M. de L.,
Solís-Cruz, B., Hernández-Patlán, D., Merino-Guzmán, R., Hernández-Velasco, X.,
Figueroa-Cárdenas, J. de D., Vázquez-Durán, A., Hargis, B. M., Téllez-Isaías, G.,
Méndez-Albores, A., Angelico, R., Colombo, C., di Iorio, E., Brtnický, M., Fojt, J., Conte,
P., … Chorover, J. (2018). Modeling of Soil Cation Exchange Capacity Based on. Geoderma,
3(4).
Marshall, D. T., & Na, D. B. (2024). The Ethics of Using Articial Intelligence in
Qualitative Research. Journal of Empirical Research on Human Research Ethics, 19(3).
hps://doi.org/10.1177/15562646241262659
Masso, A., Gerassimenko, J., Kasapoglu, T., & Beilmann, M. (2025). Research
ethics commiees as knowledge gatekeepers: The impact of emerging
technologies on social science research. Journal of Responsible Technology, 21.
hps://doi.org/10.1016/j.jrt.2025.100112
Mekhovych, K. (2025). TRANSFORMATION OF THE THEORY OF INNOVATIVE
MARKETING EFFICIENCY IN THE CONDITIONS OF AN UNDETERMINED
FUTURE. Energy Saving. Power Engineering. Energy Audit., 1 (204).
hps://doi.org/10.20998/2313-8890.2025.01.07
Mi, S., Dong, C. Q., & Liu, Y. (2025). Data insights and talent portraits of science and
technology talents: Principles and applications. Studies in Science of Science, 43(6).
Ministry of Education. (2018). Education strategic plan 2018 - 2030. In Accra, Republic of
Ghana.
135
Morante-Carballo, F., Arcentales-Rosado, M., Caicedo-Potosí, J., & Carrión-Mero,
P. (2025). Articial intelligence applications in hydrological studies and
ecological restoration of watersheds: A systematic review. In Watershed Ecology
and the Environment (Vol. 7). hps://doi.org/10.1016/j.wsee.2025.05.004
Morozova, I. A., Smetanin, A. S., & Smetanina, A. I. (2024). Management of Business
Competitiveness Based on Articial Intelligence and Big Data for Its Sustainable
Development. Journal of Modern Competition, 18(1). hps://doi.org/10.37791/2687-0657-
2024-18-1-29-40
Mortaheb, R., & Jankowski, P. (2023). Smart city re-imagined: City planning and
GeoAI in the age of big data. Journal of Urban Management, 12(1).
hps://doi.org/10.1016/j.jum.2022.08.001
Mosleh, G., Hemmati, S., Mohagheghzadeh, A., & Zarshenas, M. M. (2023). Integrating
Systems Biology and Articial Intelligence in Traditional Pharmacy Research:
Advancements, Challenges, and Opportunities. Trends in Pharmaceutical Sciences, 9(3).
Mossalanejad, A. (2025). Articial Intelligence Policymaking and the Power Equation
Priorities. Geopolitics Quarterly, 21(3). hps://doi.org/10.22034/igq.2025.511096.2023
Muhammad, N. I., Judijanto, L., Sulistiani, I., Syarifuddin, & Syahrin, M. (2023).
STRATEGIC CONTENT MARKETING: A MULTI-CHANNEL ANALYSIS OF THE
IMPACT ON BRAND AWARENESS, CONSUMER ENGAGEMENT, AND
CONVERSION RATES IN THE CONTEXT OF EVOLVING DIGITAL LANDSCAPES.
International Journal of Society Reviews (INJOSER).
Muhtar, E. (2022). PENGARUH PENDEKATAN KETRAMPILAN PROSES DALAM
MENINGKATKAN HASIL BELAJAR FISIKA PADA KONSEP GETARAN DAN
GELOMBANG PADA PESERTA DIDIK KELAS VIII SMP NEGERI 17 HALMAHERA
SELATAN. Jurnal Pembelajaran Dan Sina Fisika, 7(1).
Munen, J. (2021). SURVEY OF CONTEMPORARY CHALLENGES FACED IN
MARKETING OF LIFE INSURANCE POLICIES IN EMBU COUNTY. Frontiers in
Neuroscience, 14(1).
Muñoz-Olano, J. F., Hurtado-Parrado, C., Troll, E. S., Friese, M., Loschelder, D. D.,
Alblwi, A., McAlaney, J., al Thani, D. A. S., Phalp, K., Ali, R., Elhai, J. D., Sapci, O., Yang,
136
H., Amialchuk, A., Rozgonjuk, D., Montag, C., Bäulke, L., Daumiller, M., Dresel, M., …
Cevallos Pozo, G. (2020). A review of web-based support systems for students in higher
education. International Journal of Environmental Research and Public Health, 10(1).
Murikah, W., Nthenge, J. K., & Musyoka, F. M. (2024). Bias and ethics of AI
systems applied in auditing - A systematic review. Scientic African, 25.
hps://doi.org/10.1016/j.sciaf.2024.e02281
Naeem, M., Smith, T., & Thomas, L. (2025). Thematic Analysis and Articial Intelligence:
A Step-by-Step Process for Using ChatGPT in Thematic Analysis. International Journal of
Qualitative Methods , 24. hps://doi.org/10.1177/16094069251333886
Nashwan, A. J., & Abukhadijah, H. (2023). Harnessing Articial Intelligence for
Qualitative and Mixed Methods in Nursing Research. Cureus.
hps://doi.org/10.7759/cureus.48570
Nazari, M., Ansari, M., & Pour, M. J. A. (2025). Bibliographic Analysis and Future
Research Directions in the Field of Sustainable and Innovative Business Models.
Scientometrics Research Journal, 11(2). hps://doi.org/10.22070/rsci.2024.19650.1759
Nedoshytko, I., & Patriak, O. (2022). Electronic Document Management and Its Value for
Business. Digital Platform: Information Technologies in Sociocultural Sphere, 5(2).
hps://doi.org/10.31866/2617-796x.5.2.2022.270142
Nicmanis, M., & Spurrier, H. (2025). Geing Started with Articial Intelligence Assisted
Qualitative Analysis: An Introductory Guide to Qualitative Research Approaches with
Exploratory Examples from Reexive Content Analysis. International Journal of
Qualitative Methods , 24. hps://doi.org/10.1177/16094069251354863
Nyomba, Muh. A. (2021). Faktor Yang Berhubungan Dengan Kejadian Infeksi Saluran
Pernapasan Akut (ISPA) Pada Balita di Sekitar Wilayah Tempat Pembuangan Akhir
(TPA) Sampah Antang Kota Makassar Tahun 2021. Factors Associated of The Incidence of
Acute Respiratory Infection (ARI) in. Universitas Hasanuddin., 9.
Olawade, D. B., Omeni, D., Gore, M. N., & Hadi, M. (2025). Enhancing qualitative
research through virtual focus groups and articial intelligence: A review. International
Journal of Medical Informatics, 203. hps://doi.org/10.1016/j.ijmedinf.2025.106004
Orea-Giner, A., Fusté-Forné, F., & Soliman, M. (2025). How do tourists perceive green
customer-love service in restaurants? A qualitative exploration of ai and human
137
collaboration. International Journal of Hospitality Management, 131.
hps://doi.org/10.1016/j.ijhm.2025.104300
Owoahene Acheampong, I., & Nyaaba, M. (2024). Review of Qualitative Research in the
Era of Generative Articial Intelligence. SSRN Electronic Journal.
hps://doi.org/10.2139/ssrn.4686920
Palazzo, F., Zambea, G., & Palazzo, S. (2025). Articial Intelligence at the
Crossroads of Engineering and Innovation. Computing&amp;AI Connect, 2(1).
hps://doi.org/10.69709/caic.2025.141254
Palazzo, S., Palazzo, F., & Zambea, G. (2025). The Role of Articial Intelligence
in Advanced Engineering: Current Trends and Future Prospects. Journal of
Intelligent Communication, 4(1). hps://doi.org/10.54963/jic.v4i1.959
Pan, C., Bowen, W., Peng, Z., & Shao, L. (2024). Biological network-based exploration of
the principles of traditional Chinese medicine. Kexue Tongbao/Chinese Science Bulletin,
69(1). hps://doi.org/10.1360/TB-2023-0730
Papaneophytou, C., & Nicolaou, S. A. (2025). Promoting Critical Thinking in
Biological Sciences in the Era of Articial Intelligence: The Role of Higher
Education. In Trends in Higher Education (Vol. 4, Issue 2).
hps://doi.org/10.3390/higheredu4020024
Pepic, I., Feldt, R., Ljungström, L., Torkar, R., Dalevi, D., Maurin Söderholm, H.,
Andersson, L. M., Axelson-Fisk, M., Bohm, K., Sjöqvist, B. A., & Candeord, S.
(2021). Early detection of sepsis using articial intelligence: a scoping review
protocol. Systematic Reviews, 10(1). hps://doi.org/10.1186/s13643-020-01561-w
Perano, M., Cammarano, A., Varriale, V., del Regno, C., Michelino, F., & Caputo, M.
(2023). Embracing supply chain digitalization and unphysicalization to enhance supply
chain performance: a conceptual framework. International Journal of Physical Distribution
and Logistics Management, 53(5–6). hps://doi.org/10.1108/IJPDLM-06-2022-0201
Perdana, R., Yulianti, D., & Bertiliya, W. A. (2025). Enhancing scientic literacy
through AI-integrated Inquiry Social Complexity module in primary education.
Journal of Educational Management and Instruction (JEMIN), 5(2).
hps://doi.org/10.22515/jemin.v5i2.10891
138
Pérez-Santín, E., De-La-fuente-valentín, L., García, M. G., Bravo, K. A. S., Hernández, F.
C. L., & Sánchez, J. I. L. (2023). Applicability domains of neural networks for toxicity
prediction. In AIMS Mathematics (Vol. 8, Issue 11). hps://doi.org/10.3934/math.20231426
Permenkes. (2022). Peraturan Menteri Kesehatan Nomor 24 Tahun 2022 Tentang Rekam
Medis. Jakarta: Menteri Kesehatan Republik Indonesia. Science, 7(1).
Posunko, A., Chavarha, M., Bashkirova, L., Vasylyuk-Zaitseva, S., & Abramchuk,
O. (2025). Machine Learning in Medicine: A New Paradigm for Diagnosis and
Treatment. Futurity Medicine, 4(3). hps://doi.org/10.57125/fem.2025.09.30.01
Pra, M. G. (2025). On the Evolution of Qualitative Methods in Organizational Research.
In Annual Review of Organizational Psychology and Organizational Behavior (Vol. 12).
hps://doi.org/10.1146/annurev-orgpsych-111722-032953
Putri, S. H. (2022). Pengaruh Brand Image terhadap keputusan pembelian produk
skincare. Science, 7(1).
Pysmennyi, V., & Rudan, V. (2024). Methods of nancial management in local
communities under the conditions of the russian-Ukrainian war. Economic Analysis,
34(3). hps://doi.org/10.35774/econa2024.03.351
RAHMAN, G. O. H. S. E. Y. T. A. (2020). DETERMINANTS OF FOOD DELIVERY
PLATFORMS THROUGH CUSTOMER LOYALTY TOWARD SUSTAINABILITY OF
BUSINESS. Human Relations, 3(1).
Rahmawati. (2025). EFEKTIVITAS PELAYANAN DUKCAPIL ONLINE PADA DINAS
KEPENDUDUKAN DAN CATATAN SIPIL DI KOTA MAKASSAR. SKRIPSI
UNIVERSITAS NEGERI MAKASSAR, 11(1).
Raan, P., Penrice, D. D., & Simoneo, D. A. (2022). Articial Intelligence and
Machine Learning: What You Always Wanted to Know but Were Afraid to Ask.
In Gastro Hep Advances (Vol. 1, Issue 1).
hps://doi.org/10.1016/j.gastha.2021.11.001
Rejter, A. (2024). STUDIES IN ONOMASTICS AND THE NEW HUMANITIES.
Onomastica, 68. hps://doi.org/10.17651/ONOMAST.68.2
Rincon Soto, I. B., Soledispa-Cañarte, B. J., Sumba-Bustamante, R. V., del Carmen
Burbano-Gómez, Z., & Jiménez-Granizo, F. P. (2023). METHODOLOGICAL
139
APPROACHES AND PARADIGMS IN SCIENTIFIC AND TECHNOLOGICAL
RESEARCH: A BIBLIOMETRIC REVIEW. Bibliotecas, Anales de Investigacion, 19(1).
Rincon, I. B., Soledispa-Cañarte, B. J., Sumba-Bustamante, R. V., del Carmen
Burbano-Gómez, Z., & Jiménez-Granizo, F. P. (2023). Methodological
Approaches and Paradigms in Scientic and Technological Research: a
Bibliometric Review | Abordajes Metodológicos Y Paradigmas En La
Investigación Cientíca Y Tecnológica: Una Revisión Bibliométrica. Bibliotecas,
Anales de Investigacion, 19(1).
Roberts, J., Baker, M., & Andrew, J. (2024). Articial intelligence and qualitative research:
The promise and perils of large language model (LLM) ‘assistance.’ Critical Perspectives
on Accounting, 99. hps://doi.org/10.1016/j.cpa.2024.102722
Rodríguez Flores, E. A., & Sánchez Trujillo, M. de los Á. (2025). Scientic research and
articial intelligence in graduate students. A qualitative analysis. European Public and
Social Innovation Review, 10. hps://doi.org/10.31637/epsir-2025-1049
Rodríguez Flores, E. A., Garcés Giraldo, L. F., Valencia, J., & Valencia-Arias, A.
(2025). Research trends in the use of articial intelligence techniques in scientic
research. Revista Venezolana de Gerencia, 30(109).
hps://doi.org/10.52080/rvgluz.30.109.23
Ronzon, T., Gurria, P., Carus, M., Cingiz, K., El-Meligi, A., Hark, N., Iost, S., M’barek, R.,
Philippidis, G., van Leeuwen, M., Wesseler, J., Medina-Lozano, I., Grimplet, J., Díaz, A.,
Tejedor-Calvo, E., Marco, P., Fischer, M., Creydt, M., Sánchez-Hernández, E., … Miras
Ávalos, J. M. (2025). Understanding Material Recovery Facility Policies in the Philippines:
A Scoping Review. Sustainability (Swierland), 2(1).
Rosenberg, J. (2024). Automated Transcription of Interviews in Qualitative Research
Using Articial Intelligence A Simple Guide. Journal of Surgery Research and Practice.
hps://doi.org/10.46889/jsrp.2024/5204
Saheb, T., Sidaoui, M., & Schmarzo, B. (2024). Convergence of articial intelligence with
social media: A bibliometric & qualitative analysis. In Telematics and Informatics Reports
(Vol. 14). hps://doi.org/10.1016/j.teler.2024.100146
Sampaio, R. C., Chagas, V., Sanchez, C. S., Gonçalves, J., Borges, T., Alison, M. B.,
140
Tigrinho, C. S., de Souza, J. R., & Paz, F. S. (2024). AN ARTIFICIAL INTELLIGENCE
(AI)-ASSISTED SCOPING REVIEW OF EMERGING USES OF AI IN QUALITATIVE
RESEARCH AND ITS ETHICAL CONSIDERATIONS. Revista Pesquisa Qualitativa, 12(30).
hps://doi.org/10.33361/RPQ.2024.v.12.n.30.729
Sani, A., Maharani, A. R., & Nazimah. (2025). Peran Komunitas Dosen Dalam
Peningkatan Kualitas Pembelajaran Di Perguruan Tinggi. Journal Genta Mulia, 15(1).
Saragih, B., & , Ika Karyati, D. S. (2010). Pengaruh Pewarna Ekstrak Cair Alami Bawang
Tiwai (Eleutherine americana Merr) Terhadap Mutu Selai Kulit Pisang Kepok (Musa
paradisiaca Linn). Jurnal Tenknologi Pertanian, 6(2).
Sart, G. (2017). HOW DOES THE MAKER MOVEMENT AFFECT THE SCHOOL
ADMINISTRATIONS’ ROLES ON IMPROVING THE STEM EDUCATION? INTED2017
Proceedings, 1. hps://doi.org/10.21125/inted.2017.2452
Saurbayev, R. Zh., Zhetpisbay, A. K., & Yerekhanova, F. T. (2024). Philological
Studies: Computer Aspects of Stylometry Automation. Iasaýı Ýnıversıtetіnіń
Habarshysy, 133(3). hps://doi.org/10.47526/2024-3/2664-0686.72
Schmidt, D. A., Alboloushi, B., Thomas, A., & Magalhaes, R. (2025). Integrating articial
intelligence in higher education: perceptions, challenges, and strategies for academic
innovation. Computers and Education Open, 9. hps://doi.org/10.1016/j.caeo.2025.100274
Schneider, R., Angwin, C., Jenkinson, C., Jones, A. S. K., Jennison, C., Henley, W.,
Farmer, A., Saar, N., Holman, R. R., Pearson, E., Shields, B., al., et, Blackwell, K.,
Donskih, R., Jones, C. M., Nixon, A., Vidal, M. J., Nakov, R., Singh, P., … Ewel, C.
(2021). Eectiveness of olive oil for the prevention of pressure ulcers caused in
immobilized patients within the scope of primary health care: study protocol for
a randomized controlled trial. Trials, 21(1).
Sentioningtias, I. (2025). Pemanfaatan Aplikasi Math Kids Untuk Mengembangkan
Berhitung Anak Usia Dini Di TK Islam Nrhasanah Bandar Lampung. In Sustainability
(Swierland) (Vol. 11, Issue 1).
Shakil, M., Ali, M., Illahi, T., & Ahmed, F. (2025). FinTech 5.0 and the Future of Global
Finance: Harnessing Articial Intelligence, Blockchain, and Big Data to Build Inclusive,
Sustainable, and Resilient Financial Ecosystems. Inverge Journal of Social Sciences, 4(4).
141
hps://doi.org/10.63544/ijss.v4i4.179
Shimray, S. R., & Subaveerapandiyan, A. (2025). AI and workforce dynamics: a
bibliometric analysis of job creation, displacement and reskilling. Global Knowledge,
Memory and Communication. hps://doi.org/10.1108/GKMC-12-2024-0824
Shivarama Krishna, K., & Sathish Kumar, K. (2015). A review on hybrid renewable.
Renewable and Sustainable Energy Reviews, 52.
Sipa, M. (2023). Trends and challenges in talent management – mind map. Scientic
Papers of Silesian University of Technology Organization and Management Series, 2023(164).
hps://doi.org/10.29119/1641-3466.2022.164.34
Skoczkowski, T., Bielecki, S., Wołowicz, M., & Węglarz, A. (2025). Redening Energy
Management for Carbon-Neutral Supply Chains in Energy-Intensive Industries: An EU
Perspective. Energies, 18(15). hps://doi.org/10.3390/en18153932
Skrypnyk, S. v. (2025). Key Concepts and Models of Strategic Innovation Management
in Ukraine: Modern Approaches. The Problems of Economy, 2(64).
hps://doi.org/10.32983/2222-0712-2025-2-77-85
Skyrme, D., & Amidon, D. (2010). Business research methods. Journal of Knowledge
Management, 14(1).
Soni, A., Jain, A., & Kaur, B. (2022). A bibliometric analysis of eect of articial
intelligence on customer satisfaction in banking sector. International Research Journal of
Modernization in Engineering, 4(3).
Sovia Lenny. (2006). RINGKASAN Puding merah (Gruptophyllum pictum L Gri).
Jurnal Farmasi Galenika (Galenika Journal of Pharmacy) (e-Journal), 1(1).
Spivakovsky, O., Omelchuk, S., Kobets, V., Poltoratskyi, M., Malchykova, D., &
Lemeshchuk, O. (2025). INSTITUTIONAL REGULATION OF USING AI-BASED
SERVICES THROUGH THE KANO MODEL: THE CASE OF KHERSON STATE
UNIVERSITY. Information Technologies and Learning Tools, 108(4).
hps://doi.org/10.33407/itlt.v108i4.6269
Sposato, M. (2025). Introducing AI to qualitative research: Theoretical and practical
implications. In Exploring Qualitative Research in Business: Approaches, Strategies, And
Methods. hps://doi.org/10.4018/979-8-3693-7698-0.ch010
142
Starre, A., Yow, J., Loer, C., Irvin, M. J., Adams, P., Guillemot, F., Lacroix, F., Nocus,
I., Assen, J. H. E., Oing, H., Jin, X., Tigelaar, D., van der Want, A., Admiraal, W., Peschl,
H., Deng, C., Larson, N., Kuno, C. B., Hein, S., … León, J. (2022). Editorial Board. Teaching
and Teacher Education, 90(3).
Sydorovych, O., Bukovskyi, M., & Yaroshovych, A. (2024). DIGITAL TOOLS OF TAX
ADMINISTRATION: CURRENT TRENDS AND PERSPECTIVES. INNOVATIVE
ECONOMY. hps://doi.org/10.37332/2309-1533.2024.4.26
T. P., & . V. G. (2025). INVESTIGATION OF DEEP LEARNING FRAMEWORK
TO ACCELERATE AI FOR IMAGE PROCESSING APPLICATIONS. International
Journal For Multidisciplinary Research, 7(3).
hps://doi.org/10.36948/ijfmr.2025.v07i03.45548
Taha Kandil, T. (2025). Articial intelligence (AI) and alleviating supply chain
bullwhip eects: social network analysis-based review. Journal of Global
Operations and Strategic Sourcing, 18(1). hps://doi.org/10.1108/JGOSS-04-2023-
0038
Tàpias, M. (2019). Clothing as Solar Radiation Protection. Autex Research Journal, 2(1).
Tierney, A., Peasey, P., & Gould, J. (2025). Student perceptions on the impact of AI on
their teaching and learning experiences in higher education. Research and Practice in
Technology Enhanced Learning, 20. hps://doi.org/10.58459/rptel.2025.20005
Tong, L., Niu, Y., Zhou, L., Jin, S., Wang, Y., & Xiao, Q. (2025). Perceptions and
experiences of generative articial intelligence training to support research for Chinese
nurses: A qualitative focus group study. International Journal of Nursing Sciences, 12(3).
hps://doi.org/10.1016/j.ijnss.2025.04.010
Torres-Sainz, R., de-Zayas-Pérez, M. R., Trinchet-Varela, C. A., Pérez-Vallejo, L. M., &
Rodríguez, R. P. (2025). INTELLIGENT PREDICTIVE MAINTENANCE: A
BIBLIOMETRIC APPROACH THROUGH THE SCOPUS DATABASE. Bibliotecas, Anales
de Investigacion, 21(1).
Trejos-Gil, C. A., & Gómez-Monsalve, W. D. (2024). Articial Intelligence in
Media and Journalism. Systematic Review on Spain and Latin America in Scopus
143
and Web of Science Databases (2018-2022). In Palabra Clave (Vol. 27, Issue 4).
hps://doi.org/10.5294/pacla.2024.27.4.1
Tschisgale, P., Wul, P., & Kubsch, M. (2023). Integrating articial intelligence-based
methods into qualitative research in physics education research: A case for
computational grounded theory. Physical Review Physics Education Research, 19(2).
hps://doi.org/10.1103/PhysRevPhysEducRes.19.020123
Tursunbayeva, A., Chalu-Ben Gal, H., Santos Fonseca, S. S., Benito, P. A., Piña Ramírez,
C., Rubeis, G., Filz, M.-A., Bosse, J. P., Herrmann, C., Cavalieri, A., Reis, J., Amorim, M.,
Santarsiero, F., Carlucci, D., Schiuma, G., Sukums, F., Wamala-Larsson, C. W., Kisenge,
R., Ravi, S. J., … Kazancoglu, Y. (2024). A Spotlight on Environmental Sustainability in
View of the European Green Deal. Sustainability (Swierland), 16(1).
UU Republik Indonesia, Munawir, M., Nasional, B. S., Susanta, G., Jatimnet.com,
MajaMojokerto.net, Radarmojokerto.jawapos.com, Wijayakusuma, D. M. S., Nahman, A.,
Godfrey, L., Jacobsen, R., Buysse, J., Gellynck, X., Bayu, D., Depkes, R., Soekidjo
Notoatmodjo, Boulanger, L., Ismoyo, I. H., Tarigan, R., … Sanyal, S. (2022).
PENENTUAN ALTERNATIF LOKASI TEMPAT PEMBUANGAN AKHIR (TPA)
SAMPAH DI KABUPATEN SIDOARJO. Energies, 15(1).
van Merrienboer, J. J. G., Paas, F., Traxler, A., Brewe, E., Hazari, Z., Cass, C., Beaie, C.,
Kelly, A. M., Wilcox, B. R., Lewandowski, H. J., Bao, L., Xiao, Y., Koenig, K., Han, J.,
Eisinga, R., Grotenhuis, M., Pelzer, B., Cronbach, L. J., Bus, M. M., … VanLehn, K. A.
(2016). Evaluation of Colorado Learning Aitudes about Science Survey. Physical Review
Physics Education Research, 12(2).
Venter, I. M., Blignaut, R. J., Craneld, D. J., Tick, A., & Achi, S. el. (2025). AI versus
tradition: shaping the future of higher education. Journal of Applied Research in Higher
Education, 17(7). hps://doi.org/10.1108/JARHE-12-2024-0702
Verier, S. D., & Drive, A. H. (2015). Simulink ® Verication and Validation TM Reference.
ReVision.
Vicente, N. R. (2010). Sistemas agroorestais sucessionais como estratégia de uso e
conservação de recursos orestais em zonas ripárias da Microbacia Arroio Primeiro de
Janeiro, Anchieta-SC. Journal of Mental Science, 7(37).
Vollmar, H. C., Kramer, U., Müller, H., Griemmert, M., Noelle, G., & Schrappe, M. (2017).
144
Digitale Gesundheitsanwendungen - Rahmenbedingungen zur Nuung in Versorgung,
Strukturentwicklung und Wissenschaft - Positionspapier der AG Digital Health des
DNVF. Gesundheitswesen, 79(12). hps://doi.org/10.1055/s-0043-122233
VOLOSHYN, R. (2024). COMMODITY EXPERTISE IN E-COMMERCE. “EСONOMY.
FINANСES. MANAGEMENT: Topical Issues of Science and Practical Activity,” 4 (70).
hps://doi.org/10.37128/2411-4413-2024-4-4
Wang, C., Nan, J., Kang, Y., Xu, X., Li, T., & Zhu, Z. (2025). Design of a
comprehensive teaching experiment for mechanoluminescent human
biomechanical detection. Experimental Technology and Management, 42(9).
hps://doi.org/10.16791/j.cnki.sjg.2025.09.028
Wang, S., & Kim, R. Y. (2025). Exploring the role of generative AI in enhancing critical
thinking in multicultural classrooms: a case study from korea. Multicultural Education
Review, 17(1). hps://doi.org/10.1080/2005615X.2025.2467787
Wang, Y., Kang, M., Liu, Y., Li, J., Xue, K., Wang, X., Du, J., Tian, Y., Ni, Q., & Wang, F.
Y. (2023). Can Digital Intelligence and Cyber-Physical-Social Systems Achieve Global
Food Security and Sustainability? IEEE/CAA Journal of Automatica Sinica, 10(11).
hps://doi.org/10.1109/JAS.2023.123951
Wang, Z., Wang, D., Xu, Y., Zhou, L., & Zhou, Y. (2025). Intelligent Computing Social
Modeling and Methodological Innovations in Political Science in the Era of Large
Language Models. Journal of Chinese Political Science. hps://doi.org/10.1007/s11366-025-
09917-6
Warhurst, A. (2002). Sustainability indicators and sustainability performance
management. Sustainability Indicators and Sustainability Performance Management, 43(43).
Waryono, T. (2019). RINGKASAN Puding merah (Gruptophyllum pictum L Gri). Jurnal
Farmasi Galenika (Galenika Journal of Pharmacy) (e-Journal), 1(1).
Wijanarti, H. L., & Anisyah, T. D. A. (2022). Hubungan Antara Kualitas Tidur, Beban
Kerja Fisik Terhadap Perasaan Kelelahan Kerja Pada Perawat Rawat Inap Kelas 3 Di Rs
Pku Muhammadiyah Gamping. Jurnal Lentera Kesehatan Masyarakat, 1(1).
Williamson, B. (2020). New digital laboratories of experimental knowledge
production: Articial intelligence and education research. London Review of
Education, 18(2). hps://doi.org/10.14324/LRE.18.2.05
145
Yang, Y., & Ma, L. (2025). Articial intelligence in qualitative analysis: a practical guide
and reections based on results from using GPT to analyze interview data in a substance
use program. Quality and Quantity, 59(3). hps://doi.org/10.1007/s11135-025-02066-1
Yang, Y., & Siau, K. (2018). A Qualitative Research on Marketing and Sales in the
Articial Intelligence Age. Midwest Association for Information Systems
ConferenceAssociation for Information Systems Conference, 5(17-18 May 2018).
Yıldırım, H. M., Güvenç, A., & Can, B. (2025). Reporting Quality in Qualitative Research
Using Articial Intelligence and the COREQ Checklist: A Systematic Review of Tourism
Literature. International Journal of Qualitative Methods , 24.
hps://doi.org/10.1177/16094069251351487
Yubo, S., Ramayah, T., & Hongmei, L. (2022). Research on the Impact of Intelligent
Manufacturing Technology on Business Model Based on CNKI’s Evidence. Global
Business and Management Research, Suppl. Special Issue, 14(3s).
Zavalii, T., & Lehenchuk, S. (2025). DEVELOPMENT OF RESEARCH METHODS IN
MARKETING IN THE CONDITIONS OF FORMING THE DIGITAL MARKETING
PARADIGM. Economic Scope, 204. hps://doi.org/10.30838/ep.204.101-109
Zhang, B., Liu, Q., Li, X., Liu, L., Yang, B., Letu, H., Gao, L., Zhang, W., Zhang,
H., Bian, Z., Qi, M., Chen, C., & Shang, H. (2025). The core concepts and
fundamental issues of remote sensing science. National Remote Sensing Bulletin,
29(1). hps://doi.org/10.11834/jrs.20244503
Zhao, R., Yang, X., Zhang, D., Li, J., Huang, Y., Xian, G., Kou, Y., & Sun, T. (2025).
Articial Intelligence for Agricultural Science (AI4AS): Key Elements, Challenges and
Pathways. Smart Agriculture, 7(3). hps://doi.org/10.12133/j.smartag.SA202502019
Zhao, T., Zhou, Y., Shi, R., Cao, Z., & Ren, Z. (2024). GWAI: Articial intelligence
platform for enhanced gravitational wave data analysis. SoftwareX, 28.
hps://doi.org/10.1016/j.softx.2024.101930
Zhao, X., Wang, X., Research, R. C.-A. in R. (2025). Research on optimizing water
resource management and improving crop yield using big data and articial
intelligence. Jstage.Jst.Go.Jp, 5(2).
Zharov, A. M. (2024). THE DYNAMICS OF KNOWLEDGE:AN EVOLUTIONARY
ANALYSIS OF SCIENTIFIC DEVELOPMENT THEORIES. Vestnik Tomskogo
146
Gosudarstvennogo Universiteta. Filosoya, Sotsiologiya, Politologiya, 80.
hps://doi.org/10.17223/1998863x/80/3
Zheldibayeva, R. (2025). THE IMPACT OF AI AND PEER FEEDBACK ON RESEARCH
WRITING SKILLS: A STUDY USING THE CGSCHOLAR PLATFORM AMONG
KAZAKHSTANI SCHOLARS. Scientic Journal of Astana IT University.
hps://doi.org/10.37943/21tbax7694
Zheng, Z., He, Z., Khaab, O., Rampal, N., Zaharia, M. A., Borgs, C., Chayes, J.
T., & Yaghi, O. M. (2023). Image and Data Mining in Reticular Chemistry Using
GPT-4V. ArXiv.
Zozulov, O., & Boboshko, Y. (2025). Classication of product development and market
entry models. Marketing and Digital Technologies, 9(2).
hps://doi.org/10.15276/mdt.9.2.2025.4
Zuo, H. (2025). Human-Environment Relationship Research from a Multidisciplinary
Integration Perspective in Geography. Frontiers in Sustainable Development, 5(8).
hps://doi.org/10.54691/64t83m16
Игнатьев, М. Е., & Степанов, В. В. (2025). Методология диссертационных
исследований в юридических науках: отдельные аспекты и проблемы.
СОВРЕМЕННОЕ ПРАВО, 7. hps://doi.org/10.25799/ni.2025.32.84.001
Ксения Юрьевна, Б., & Елена Иосифовна, Л. (2024). Identication and regulation of
the macro-regional ecosystem’ human resources potential development trends: global
and national experience. STATE AND MUNICIPAL MANAGEMENT SCHOLAR NOTES,
1(4). hps://doi.org/10.22394/2079-1690-2024-1-4-28-33
This edition of "Scientific research methodology applied to artificial intelligence
and data science: General approach", was completed in the city of Colonia
del Sacramento in the Eastern Republic of Uruguay on October 3, 2025
ISBN 978-9915-698-43-4