Which Research Competencies Should Remain Independently Demonstrable in an AI-Rich Research Environment?

What Should Researchers Still Be Able to Do When AI Can Perform Much of the Research Workflow?

A doctoral student submits a thesis containing a sophisticated literature review, a complex statistical analysis, and a carefully written discussion. Generative AI contributed to nearly every stage of the work.

During the thesis examination, the student is asked to explain why the statistical model was selected, how the literature supports the theoretical framework, and whether the findings justify the conclusions.

The student can describe what the AI system recommended but struggles to reconstruct the reasoning behind those recommendations.

The thesis may contain technically plausible work, yet the examination reveals uncertainty about the student's own research competence.

This raises a fundamental question for research education: which competencies should researchers remain able to demonstrate independently when AI can assist with literature searching, methodological decisions, data analysis, interpretation, and scientific writing?

The answer requires distinguishing intellectual independence from technological self-sufficiency. Researchers have never been expected to conduct every aspect of scholarship without tools or collaborators. Nevertheless, they remain responsible for understanding and defending the scientific decisions attributed to them.

Which Research Abilities Should Remain Independently Demonstrable?

Defining Independent Research Competence in an AI-Rich Environment

What Does Independently Demonstrable Competence Mean?

Independent competence refers to a researcher's ability to demonstrate relevant knowledge and judgment without depending entirely on another agent to supply the reasoning.

For example, a researcher may use statistical software to estimate regression coefficients while independently explaining why the model is appropriate, what the coefficients represent, and which limitations affect interpretation.

Similarly, a researcher may use AI to improve manuscript language while retaining the ability to explain the argument and identify the evidence supporting its claims.

The important distinction is between assistance with completing research activities and dependence on assistance to understand those activities.

Both forms of independence can matter, but their importance differs by task.

A computational researcher may need to demonstrate programming competence, while another researcher may appropriately collaborate with a programmer.

Both should understand the scientific purpose and implications of computational work incorporated into their research.

Why the Final Research Product Is Not Enough

Generative AI can help produce polished manuscripts, analytical scripts, literature summaries, and methodological explanations.

These outputs may be useful and, under appropriate conditions, scientifically sound.

However, the quality of an AI-assisted product does not establish the competence of the person who submitted it.

A student may produce a correct statistical interpretation without understanding the underlying model. Another may write a convincing literature synthesis without examining the original publications.

Conversely, a capable researcher may produce an imperfect first draft while demonstrating strong methodological reasoning.

Research assessment should therefore distinguish product quality from individual understanding.

This is particularly important in thesis supervision, research-methods education, and situations where the researcher's competence is itself an intended learning outcome.

1. Formulating and Justifying Research Questions

Researchers should be able to explain what their study investigates, why the question matters, and what evidence would help answer it.

AI can generate numerous possible research questions, but a researcher should determine whether a question is conceptually coherent, feasible, and relevant to the discipline.

For example, a study examining AI literacy among university faculty should specify what is meant by literacy and why the proposed research design can investigate it.

The researcher should also distinguish between an interesting topic and a researchable problem.

Independent competence does not require inventing every possible question without assistance. It requires evaluating and justifying the selected question.

2. Searching for and Evaluating Scholarly Evidence

Researchers should understand how relevant literature can be identified and how the quality and applicability of evidence can be evaluated.

Foundational information-literacy competencies include identifying search concepts, selecting appropriate information sources, verifying references, and examining original publications.

The Association of College and Research Libraries' Framework for Information Literacy for Higher Education emphasizes the strategic and contextual nature of information searching.

AI can support discovery, but generated reference lists do not automatically establish adequate coverage or source accuracy.

Researchers should therefore retain the ability to evaluate whether a search is appropriate for its purpose.

This relates to independent literature-search competence.

3. Critically Appraising Research Claims

Researchers should be able to evaluate the relationship between a publication's research question, design, methods, results, and conclusions.

For example, a researcher should recognize when an observational study is being interpreted as though it established a causal relationship.

Critical appraisal also involves examining measurement, sampling, uncertainty, and relevant limitations.

AI-generated summaries may help identify important information, but they can omit qualifications or misrepresent findings.

Researchers should remain capable of evaluating the original evidence rather than accepting a generated interpretation solely because it appears coherent.

This is central to maintaining critical reading skills.

4. Explaining Methodological Choices

Researchers should be able to justify why their selected methodology addresses the research question.

This includes understanding relevant assumptions, design constraints, and alternative approaches.

For example, a student using a mixed-methods design should explain why combining qualitative and quantitative evidence is appropriate and how the components contribute to the research objectives.

Similarly, a quantitative researcher should understand why a particular statistical model fits the data structure and inferential purpose.

AI can suggest methods, but the final choice should not rest solely on the system's apparent confidence.

Independent methodological reasoning is especially important when a researcher must defend a study during peer review or thesis examination.

5. Understanding Research Data and Their Limitations

Researchers should understand what their observations represent, how data were collected, and which transformations or limitations affect analysis.

For quantitative research, this may include variable definitions, measurement scales, missing-data patterns, repeated observations, and data provenance.

For qualitative research, it may involve familiarity with participant accounts, contextual details, analytical records, and the circumstances in which material was generated.

AI can summarize datasets or transcripts, but these summaries should not become the researcher's only understanding of the evidence.

The ability to explain the structure and meaning of one's own data remains essential to responsible interpretation.

6. Interpreting Results and Representing Uncertainty

Researchers should be able to explain what their findings support and what remains uncertain.

For quantitative studies, this may involve interpreting effect estimates, confidence intervals, model assumptions, and the distinction between statistical and practical importance.

For qualitative studies, it may involve explaining how interpretations were developed and substantiated through the original material.

The American Statistical Association's statement on p-values emphasizes that scientific conclusions should not be reduced to whether a conventional significance threshold has been crossed.

AI-generated interpretations may overlook such qualifications.

Researchers should therefore remain able to explain the findings without depending on generated narratives.

7. Constructing Evidence-Based Scientific Arguments

Researchers should be able to develop claims that correspond to the available evidence.

This includes distinguishing findings from speculation, explaining how studies relate to one another, and acknowledging alternative interpretations.

AI can improve clarity and organization, but a polished paragraph does not establish that its argument is defensible.

For example, a researcher should be able to explain why a particular finding supports the discussion's central claim.

Independent scientific writing competence is therefore closely connected to reasoning rather than merely sentence production.

The concern about AI reliance and scientific writing skills is particularly relevant when students delegate entire sections without developing their own interpretations.

8. Recognizing Errors and Evaluating Automated Recommendations

Researchers should understand that AI can produce inaccurate references, unsuitable methodological recommendations, incorrect code, and misleading interpretations.

They should know how to verify consequential outputs using appropriate independent evidence.

For example, a researcher using AI-generated code should be able to determine whether the implementation matches the intended analysis or obtain qualified computational review.

A researcher using AI-generated citations should verify that the publications exist and support the attributed claims.

These competencies are central to appropriate reliance on automation.

Lee and See (2004) explain that trust should correspond to the actual capabilities of an automated system rather than produce indiscriminate acceptance or rejection.

9. Upholding Research Ethics and Integrity

Researchers should understand their responsibilities concerning honesty, transparency, confidentiality, participant protection, and accurate reporting.

The European Code of Conduct for Research Integrity identifies reliability, honesty, respect, and accountability as fundamental principles.

These principles remain applicable when AI contributes to research.

Researchers should recognize when AI assistance creates risks involving fabricated evidence, unauthorized data disclosure, misleading methodological descriptions, or unacknowledged intellectual contributions.

They should also understand applicable institutional, journal, funder, and professional requirements.

Research integrity cannot be delegated to an AI system.

10. Recognizing When Independent Expertise Is Necessary

Independent competence includes recognizing the limits of one's own expertise.

A researcher may understand the scientific objective of a study while requiring specialist assistance with advanced statistical modeling, clinical interpretation, or technical implementation.

Seeking expertise is not evidence of inadequate independence.

Research is frequently collaborative, and responsible judgment involves knowing when a decision requires additional knowledge.

AI can help researchers prepare questions for consultation, but it should not automatically replace independent expert advice.

Which Competencies Must Be Independent, and Which Can Be Supported by Tools?

The appropriate distinction depends on the researcher's responsibilities.

The table is a practical framework rather than a universal standard.

For example, a computational researcher may need to demonstrate substantial independent programming competence, while another researcher may appropriately rely on a qualified collaborator for technical implementation.

Both remain responsible for accurately representing their contributions and ensuring that the resulting research is appropriately reviewed.

Why Independent Competence Does Not Require Manual Reproduction of Everything

Researchers have long used tools to extend their capabilities.

Statistical software, reference managers, laboratory instruments, and computational libraries perform activities that would otherwise require substantial manual effort.

Generative AI introduces additional assistance with reasoning-intensive tasks.

The presence of automation does not automatically invalidate the researcher's contribution.

For example, a statistician need not calculate every coefficient manually to demonstrate understanding of regression analysis.

However, the statistician should understand the model, relevant assumptions, and interpretation of its results.

Likewise, a researcher using AI-generated code should know what the code is intended to accomplish and how its behavior has been verified.

Independent competence concerns intellectual control over the research process, not the absence of technological assistance.

How Should Competence Be Assessed When AI Is Widely Available?

Assessment should align with the specific knowledge and abilities researchers are expected to demonstrate.

If the objective is statistical interpretation, students might explain an unfamiliar output without AI-generated assistance.

If the objective is literature appraisal, students might examine an original publication and identify its methodological strengths and limitations.

If the objective is scientific writing, students might construct a short evidence-based argument and explain its supporting sources.

These tasks provide evidence of independent competence that a polished AI-assisted manuscript may not reveal.

However, assessments should not require artificial manual work unrelated to the intended competency.

For example, prohibiting calculators during an assessment of methodological interpretation may add unnecessary difficulty without improving the validity of the assessment.

Why Oral Examinations Can Be Useful but Are Not Sufficient Alone

Oral examinations can provide opportunities for students to explain methodological decisions, respond to questions, and clarify their understanding.

They may help distinguish memorized explanations from deeper reasoning when examiners use appropriate follow-up questions.

However, oral performance can be influenced by anxiety, language proficiency, communication style, and familiarity with examination conditions.

Consequently, oral examination should not automatically be treated as a perfect measure of research competence.

Combining oral discussion with written justification, practical demonstrations, and other aligned assessments may provide a more comprehensive picture.

The broader challenge of assessing students' own research abilities when AI assistance is allowed requires attention to validity, accessibility, and fairness.

What Does Research on Cognitive Offloading Suggest?

Risko and Gilbert (2016) describe cognitive offloading as the use of external resources to reduce cognitive demands.

Such assistance can improve performance without necessarily reducing competence.

However, generative AI can offload activities that are central to research judgment, including interpreting evidence and proposing methodological decisions.

Researchers should distinguish between offloading routine operations and delegating the reasoning they are expected to understand.

Reduced independent practice may create concerns about skill maintenance, but it does not automatically establish deterioration.

The question of whether heavy AI use weakens research skills requires evidence of changes in independent performance rather than assumptions based on frequency of use.

How Should Expectations Differ Across Research Career Stages?

Beginning research students may need structured opportunities to develop foundational competencies.

Advanced doctoral students may be expected to demonstrate greater methodological autonomy and the ability to defend complex decisions.

Experienced researchers may use AI extensively for tasks within their expertise while remaining responsible for verification.

However, academic seniority does not guarantee competence in unfamiliar methods.

The appropriate standard should reflect demonstrated task-specific understanding.

AI-use expectations should therefore be responsive to learning objectives and responsibilities rather than determined solely by academic rank.

Can AI Help Researchers Demonstrate Competence?

AI can support formative assessment by generating practice problems, suggesting alternative explanations, or identifying possible weaknesses in a researcher's reasoning.

However, an AI-generated evaluation is not automatically an accurate measure of competence.

For example, an AI system may incorrectly criticize a valid statistical interpretation or fail to identify a methodological problem.

Educators should verify consequential feedback and use assessment methods appropriate to the intended learning outcomes.

AI can contribute to the learning process without becoming the sole authority for determining whether a researcher is competent.

Why Independent Judgment Remains Central to Scientific Accountability

Scientific research depends on the ability to explain how conclusions were reached and why they are justified.

AI may contribute to the production of research outputs, but it does not assume the professional responsibilities of human researchers.

The International Committee of Medical Journal Editors states that AI-assisted technologies should not be listed as authors and that human authors remain responsible for the accuracy and integrity of their work.

Other publishers and institutions may establish additional requirements.

Researchers should consult the policies applicable to their work.

Ultimately, independent competence matters because researchers must be able to defend the evidence and reasoning underlying the claims they publish.

Assessing Research Competence When AI Helped Produce the Thesis

The example illustrates how independent competence can be assessed without requiring the student to reproduce every technical operation manually.

The committee evaluates whether the student understands and can defend the intellectual work represented in the thesis.

The specific assessment criteria should reflect the doctoral program's learning outcomes, disciplinary standards, and applicable institutional requirements.

Misconceptions About Independent Research Competence

How to Define and Assess Independent Research Competence

Research programs and supervisors should identify which competencies students must demonstrate and what forms of assistance are compatible with those objectives.

The requirements should reflect disciplinary expectations and the actual intellectual responsibilities of the researcher.

Define Competencies Before Designing Assessments

Research educators should specify whether an activity measures methodological understanding, literature appraisal, data interpretation, programming, or scientific writing.

The permitted use of AI should follow from the intended competency rather than determine it.

Distinguish Assistance From Intellectual Responsibility

Researchers may use AI to generate code, organize literature, or improve language.

However, they should remain able to explain the consequential decisions and verify the resulting work.

Where specialized expertise is required, responsibilities should be assigned transparently.

Use Multiple Forms of Competence Evidence

Written justifications, practical demonstrations, oral discussions, and independent analytical tasks can provide complementary evidence of understanding.

The assessment design should be appropriate to the student's discipline, level of training, and learning objectives.

Support Learning Before Judging Performance

When students cannot explain AI-assisted work, supervisors should identify the relevant knowledge gap and provide opportunities for learning.

Inadequate understanding does not automatically establish that AI caused deterioration or that academic misconduct occurred.

Where integrity concerns arise, they should be addressed through applicable institutional procedures and evidence rather than assumptions based solely on the presence of AI assistance.

Preserve Human Accountability

Researchers should be able to explain why their methods, interpretations, and conclusions are defensible.

AI assistance may change how research is conducted, but it does not remove the responsibility to evaluate evidence and accurately represent scientific work.

Which Research Competencies Can You Demonstrate Independently?

Common Questions About Independent Research Competencies

Preserve the Competence Needed to Defend Scientific Decisions

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.