03 · What You Need to Know
How to Maintain Local Research Independence When Using Globally Trained AI
Global Training Data Do Not Represent Every Research Context Equally
Generative AI systems are developed using information resources that may differ substantially in language, geography, disciplinary coverage, and institutional representation.
Some scholarly traditions and research environments may therefore be more visible than others.
For example, research from highly indexed English-language journals may be more prominent in the information available to an AI system than studies published in regional journals or institutional repositories.
However, the complete training composition of proprietary models may not be publicly available.
Researchers should distinguish plausible concerns about uneven representation from demonstrated biases in a particular model's outputs.
The mechanisms discussed in AI training-data bias provide a starting point for evaluating these concerns.
Training-Data Inequality and Retrieval Inequality Are Different Problems
Some AI systems generate responses primarily from patterns learned during training, while others retrieve information from external databases and websites.
These mechanisms may introduce different limitations.
A model may have incomplete knowledge of local research because relevant material was poorly represented during training.
A retrieval-enabled system may instead overlook relevant publications because they are absent from its search index or ranked below more prominent sources.
Researchers should identify which information sources and procedures contribute to the AI output where possible.
Corrective measures should address the actual limitation rather than assume that every omission originates in training data.
Research Questions Should Begin With the Phenomenon
Generative AI can suggest research questions, but the researcher should first establish what problem requires investigation.
For example, a study of asynchronous learning may arise from observed difficulties involving assessment schedules, employment obligations, or access to educational resources.
An AI assistant might instead propose a familiar question about technology acceptance because that topic appears frequently in scholarly literature.
Both questions may be legitimate, but they investigate different phenomena.
Researchers should not allow the availability of an AI-generated framework to redefine the original research problem without justification.
Local Relevance Should Be Established Through Evidence
A locally situated research question should be supported by appropriate evidence concerning the population and setting.
This may include previous studies, institutional records, policy documents, preliminary observations, or other relevant sources.
For example, claims about the experiences of working students should not be based solely on assumptions about employment or national culture.
Researchers should examine whether the proposed problem is documented and scientifically meaningful.
Local relevance is an evidentiary consideration rather than a label automatically attached to research conducted in a particular country.
AI-Recommended Theories Require Contextual Evaluation
Generative AI may frequently recommend established theories associated with particular research topics.
For example, technology acceptance models may be suggested for studies involving educational technology adoption.
These frameworks can be appropriate, but their assumptions should be examined.
A study concerned with institutional constraints may require additional theoretical perspectives beyond individual behavioral intention.
Researchers should compare candidate frameworks according to conceptual fit, empirical support, and explanatory value.
The possibility of AI favoring dominant scientific theories should encourage critical selection rather than automatic rejection.
Local Knowledge Should Not Be Treated as Automatically Superior
Locally developed concepts may provide important insights into particular populations and institutions.
However, their geographic origin does not independently establish scientific validity.
Similarly, internationally established frameworks should not be rejected merely because they originated elsewhere.
Researchers should evaluate all theoretical perspectives according to the research question, evidence, and methodological requirements.
The objective is contextual adequacy rather than geographic preference.
AI May Introduce Western-Centric Assumptions
Some AI-generated recommendations may reflect assumptions associated with highly visible Western academic traditions.
For example, an interpretation may emphasize individual motivation while giving limited attention to institutional or relational conditions.
However, Western scholarship is diverse, and individual-level explanations are not inherently inappropriate.
Researchers should examine whether the recommended framework adequately represents the evidence.
The issues discussed in Western-centric assumptions in AI-assisted research are relevant when evaluating theoretical and methodological fit.
Local Research Literature Should Be Searched Independently
AI-generated literature recommendations may omit relevant studies from regional journals, institutional repositories, or non-English sources.
Researchers should conduct independent searches using appropriate databases and local research resources.
For example, a study concerning Philippine higher education may require evidence from local journals, university repositories, and official educational policy documents.
International studies may remain essential for theoretical and comparative purposes.
The goal is to identify relevant evidence rather than achieve an arbitrary geographic distribution of citations.
Researchers should not interpret missing AI recommendations as proof that local research does not exist.
Research Gaps Should Not Be Determined by AI Visibility
A research gap is established through critical examination of relevant literature.
It is not established merely because AI fails to recommend studies on a particular topic.
For example, an AI assistant may claim that generative AI literacy has not been examined among Philippine university students.
Researchers should verify that claim through documented searches using appropriate terminology and sources.
The limitations of AI-assisted discovery of local literature can otherwise produce artificial research gaps.
Institutional Context Should Be Described Explicitly
Researchers can improve the relevance of AI suggestions by describing the actual institutional environment.
For example, a prompt may specify that a university program is fully online and asynchronous-first, that students may be employed, and that campus attendance is not required.
Such information may prevent the AI from assuming conventional classroom arrangements.
However, the researcher should verify that the system has interpreted these conditions correctly.
Detailed prompting improves contextual information but does not guarantee contextual understanding.
Local Regulations Require Authoritative Verification
Generative AI may provide incorrect descriptions of research ethics procedures, data protection requirements, or institutional approval processes.
For example, advice about processing personal information in the Philippines should be checked against applicable Philippine law and relevant institutional requirements.
Researchers should verify legal and administrative claims through authoritative sources.
The problem of AI misunderstanding local research systems can affect research feasibility and ethical compliance.
AI should assist with identifying questions rather than independently determining regulatory obligations.
Methodological Recommendations Should Match Available Resources
AI may recommend research designs requiring extensive funding, specialized equipment, or access to large datasets.
These designs may be technically defensible but impractical in the intended setting.
Researchers should examine participant access, data availability, budget, timeline, and methodological expertise.
However, resource constraints do not justify inappropriate analyses or unsupported conclusions.
Where necessary, researchers should revise the research question or select a defensible alternative design.
Contextual Adaptation Should Preserve Methodological Rigor
Adapting research procedures to local conditions can improve feasibility and validity.
For example, recruitment procedures may need to accommodate working students who cannot participate during conventional office hours.
However, such changes can affect sampling and the interpretation of findings.
Researchers should document adaptations and evaluate their methodological consequences.
Local suitability and scientific rigor should be treated as complementary requirements.
Research Instruments Need Contextual Validation
An instrument developed in one population may not measure the same construct equivalently in another.
For example, items measuring educational technology access may assume that device ownership implies independent access.
In some settings, shared devices or intermittent connectivity may be relevant dimensions.
Researchers should examine whether the instrument adequately represents the intended construct.
Where appropriate, adaptation should be followed by cognitive interviewing, pilot testing, or measurement evaluation.
AI-generated questionnaire items should not be treated as validated merely because they appear coherent.
Language and Translation Require Contextual Attention
AI may interpret local terminology through familiar English-language categories.
For example, culturally specific expressions may be translated into concepts that capture only part of their meaning.
Researchers should preserve original-language evidence when consequential interpretation depends on linguistic nuance.
Qualified bilingual or cultural review may be necessary.
The concern is particularly important when AI assists with interviews, questionnaires, or qualitative analysis.
AI-Assisted Qualitative Coding Should Be Verified Against Original Evidence
Generative AI may suggest codes that reflect dominant academic categories rather than participants' intended meanings.
For example, descriptions of institutional scheduling constraints may be classified as poor individual time management.
Researchers should examine whether the classification is supported by the source material.
Alternative interpretations should be considered when relevant.
The final coding framework should reflect the chosen qualitative methodology and the evidence rather than the AI's preferred categories.
Researchers Should Examine Alternative Explanations
AI-generated interpretations may emphasize explanations that are familiar or consistent with the researcher's prompt.
Researchers should ask whether other explanations are supported by the evidence.
For example, limited participation in online learning may involve motivation, connectivity, employment, institutional scheduling, or combinations of these factors.
However, requesting alternative explanations from AI does not independently validate them.
Each interpretation should be assessed against appropriate data and scholarship.
AI Can Help Identify Assumptions in Its Own Recommendations
Researchers may ask AI to explain the assumptions underlying a suggested theory or method.
For example, the system can be asked whether a proposed sampling strategy requires access to an institutional population registry.
This may reveal constraints that were not apparent in the initial recommendation.
However, AI self-critique may overlook important limitations or produce plausible but inaccurate explanations.
Independent verification remains necessary.
Matched-Prompt Testing Can Help Evaluate Contextual Bias
Researchers may compare AI advice across carefully constructed research scenarios.
For example, prompts can describe similar studies while varying the national or institutional context.
Qualified reviewers can evaluate methodological rigor, factual accuracy, and contextual appropriateness.
However, legitimate contextual differences may justify different recommendations.
Testing should identify unjustified disparities rather than demand identical advice across settings.
Researcher Positionality Influences AI Use
Researchers bring disciplinary assumptions, institutional experiences, and theoretical preferences to their interactions with AI.
Prompts may reproduce these assumptions.
For example, asking why students lack self-discipline already frames the phenomenon differently from asking how institutional arrangements influence learning practices.
Researchers should examine how their own questions shape the recommendations they receive.
Reflexive documentation can help distinguish researcher assumptions from claims supported by evidence.
Local Experts Can Provide Essential Contextual Verification
Researchers familiar with the relevant institution or population may identify errors that generic AI advice overlooks.
For example, local research officers may clarify data access requirements, while disciplinary specialists may identify relevant regional scholarship.
However, local expertise should also be evaluated critically.
Where possible, consequential decisions should be supported by documented evidence rather than informal authority alone.
AI Should Not Determine What Counts as Legitimate Knowledge
The prominence of a concept in AI-generated responses does not establish its scientific superiority.
Similarly, the absence of a local theory or publication from an AI recommendation list does not establish its irrelevance.
Researchers should evaluate knowledge claims according to appropriate epistemological and methodological standards.
AI can expand the range of possibilities considered, but it should not serve as the final authority over theoretical legitimacy.
Research Decisions Should Remain Traceable
When AI materially influences a research project, researchers should document consequential recommendations and decisions.
Relevant records may include prompts, model information, generated suggestions, alternative approaches, verification procedures, and final rationales.
For example, a researcher may record why an AI-recommended theory was supplemented with a locally relevant institutional perspective.
Documentation should reflect actual decision-making rather than provide retrospective justifications invented by AI.
AI Can Support Methodological Independence
Generative AI can explain methods, compare theories, and identify questions for further investigation.
These capabilities may help researchers working with limited methodological support.
However, researchers should use them to strengthen independent judgment rather than delegate final decisions.
The potential for AI methodological assistance is most valuable when researchers can critically evaluate the recommendations.
Institutional AI Policies Should Recognize Local Research Needs
Universities may establish policies governing acceptable AI use, research data protection, training, and disclosure.
These policies should consider the institution's research environment and disciplinary needs.
For example, researchers may require guidance about multilingual data, local policy sources, and the use of AI in qualitative analysis.
However, local policies should remain consistent with applicable legal, ethical, and scientific requirements.
Institutional governance should support contextual judgment rather than impose one universal AI workflow.
Research Quality Should Be Evaluated Independently of AI Use
A study does not become more rigorous merely because it uses AI.
Likewise, research conducted without advanced AI services is not inherently inferior.
Scientific quality depends on the relevance of the question, appropriateness of methods, validity of evidence, and defensibility of conclusions.
Researchers should evaluate AI contributions according to whether they improve these dimensions.
The objective is not technological conformity but reliable scientific inquiry.
