Can AI Make It Harder to Identify Each Human Contributor’s Intellectual Role?

When AI Helps Produce the Research, How Do We Know Who Contributed What?

A research team submits a manuscript containing a detailed conceptual framework, statistical analysis, and carefully developed discussion. The contribution statement identifies one researcher as responsible for methodology, another for formal analysis, and a third for writing the original draft.

During manuscript preparation, however, generative AI was used extensively. It suggested alternative research designs, generated analytical code, proposed interpretations, and drafted substantial portions of the discussion.

The final manuscript does not reveal which decisions originated with human researchers, which ideas were suggested by AI, or how extensively the researchers evaluated and revised the generated material.

Does the contribution statement still accurately represent the intellectual work performed?

This question concerns AI and human intellectual contributions in research. As automated systems become involved in increasingly substantive research activities, the visible output may provide less information about the reasoning, judgment, and responsibility exercised by individual collaborators.

Can AI Make Human Intellectual Contributions Harder to Identify?

Why AI Complicates Attribution of Scholarly Contributions

The Final Research Output Does Not Reveal the Entire Intellectual Process

Research outputs ordinarily present the results of many interconnected activities. A manuscript may contain conceptual arguments, methodological decisions, analytical procedures, figures, and interpretations developed by different contributors.

Even without AI, determining who contributed which intellectual elements can be difficult, particularly when collaborators revise one another's work.

Generative AI introduces an additional source of suggestions and generated material.

For example, a researcher may ask AI to propose theoretical explanations, another may select among those explanations, and a third may integrate the selected interpretation into the manuscript.

The final paragraph does not necessarily reveal these distinct activities.

Attribution therefore requires examining the process through which the research was developed rather than inferring contributions solely from the finished text.

Intellectual Contribution Is Different From Material Production

A central distinction is between producing an output and contributing the intellectual reasoning that makes the output scientifically meaningful.

A researcher may use AI to generate hundreds of lines of analytical code while making substantial contributions through model specification, testing, and interpretation.

Another researcher may generate a lengthy literature review without independently examining the publications or developing the synthesis.

Although both individuals produced material using AI, the nature and extent of their intellectual contributions differ.

These categories can overlap. Producing a manuscript section may itself involve substantial intellectual work, while reviewing generated material may range from superficial approval to major conceptual revision.

Researchers should therefore examine the actual activities involved rather than automatically assigning intellectual credit according to output volume.

Why AI Assistance Can Obscure Methodological Contributions

Methodological work often involves decisions that are not visible in the final methods section.

Researchers may compare alternative designs, assess assumptions, determine sampling strategies, or decide how variables should be operationalized.

AI can contribute suggestions at each stage.

For example, an AI assistant might propose a multilevel analytical approach. One researcher may accept the suggestion immediately, while another may independently evaluate the dependence structure, compare alternative models, and determine that the recommendation is suitable.

The resulting methods section could look similar in both cases.

However, the intellectual work performed by the researchers differs considerably.

A contribution statement identifying someone as responsible for methodology should therefore reflect their actual methodological contribution, not merely their use of an AI tool that supplied a recommendation.

How AI Can Complicate Attribution of Statistical and Computational Work

AI-generated code can blur distinctions among computational implementation, analytical design, and interpretation.

Suppose one collaborator specifies the statistical model, another uses AI to generate the code, and a third tests the implementation and interprets the results.

These activities may correspond to different contributor roles.

Generating code does not necessarily mean the researcher selected the analytical method. Similarly, selecting a model does not establish that the researcher verified its implementation.

Researchers should document these distinctions when they are relevant to contribution statements and accountability.

The question of understanding AI-generated research code is also relevant because a contributor may be unable to explain an implementation they nominally produced.

Computational contribution records should therefore identify substantive responsibilities rather than merely the person who operated the coding assistant.

Why Writing Contributions Are Particularly Difficult to Attribute

Generative AI can produce manuscript text that resembles ordinary scholarly writing.

A researcher may generate an initial discussion, critically revise its interpretation, and incorporate verified evidence. Another may accept the generated passage with minimal modification.

Both could report having prepared the discussion section, but their intellectual involvement differs.

Authorship assessment should distinguish between generating prose, critically developing an argument, and verifying the scientific claims expressed in that prose.

The International Committee of Medical Journal Editors (ICMJE) includes drafting or critically reviewing work for important intellectual content among its authorship criteria.

Under that framework, the significance of writing contributions cannot be established solely by counting words or identifying who first inserted a paragraph into the manuscript.

The broader issue of AI-assisted contributions and academic authorship therefore requires careful attention to substantive human work.

What Does the CRediT Taxonomy Contribute?

The Contributor Roles Taxonomy, commonly called CRediT, provides a standardized vocabulary for describing contributions to scholarly work.

Its 14 roles include Conceptualization, Methodology, Software, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Supervision, and Project administration.

CRediT can help research teams distinguish different forms of human contribution that might otherwise be grouped together.

For example, a collaborator who designed an analytical strategy may contribute to Methodology, while another who developed and tested computational implementations may contribute to Software and Formal analysis.

However, CRediT does not independently determine authorship eligibility.

Assigning a contributor role does not establish that the individual satisfies the applicable journal's authorship criteria.

Nor does a role label explain how AI assistance affected the work unless the team records that information separately.

Why AI Tools Should Not Be Treated as Human Contributors

Major publishing frameworks distinguish AI-assisted technologies from accountable scholarly authors.

The ICMJE states that AI-assisted technologies should not be listed as authors because they cannot assume responsibility for the accuracy, integrity, and originality of the work.

The Committee on Publication Ethics (COPE) similarly maintains that AI tools cannot meet authorship requirements involving responsibility and accountability.

These positions do not deny that AI systems can materially affect research outputs.

Instead, they distinguish technological assistance from the responsibilities associated with human authorship.

Researchers should therefore document AI assistance as required by applicable policies without treating the system as a human collaborator.

Human contribution statements should identify the people who performed qualifying work and accepted the relevant responsibilities.

Can Prompt Engineering Constitute an Intellectual Contribution?

Prompting can involve different levels of intellectual engagement.

A simple request to rewrite a paragraph may involve little substantive scholarly reasoning.

By contrast, developing detailed analytical specifications, identifying constraints, comparing generated alternatives, and validating results may involve considerable intellectual work.

However, prompt complexity alone does not establish the significance of a research contribution.

The relevant question is what the person contributed to the research problem, methodology, analysis, or interpretation.

For example, a researcher who specifies a theoretically justified model and uses AI to implement it may have made a substantial methodological contribution.

Another researcher who asks AI to choose the model without understanding the study design may not have performed equivalent intellectual work.

Authorship and contributorship decisions should reflect these differences without treating every prompt as either trivial or inherently substantial.

Why Verification Is Itself an Important Research Activity

Verification can involve substantial intellectual work, particularly when it requires specialized methodological judgment.

A researcher who discovers that AI-generated code incorrectly handles missing data may need to investigate the dataset, revise the analytical specification, and determine whether previous results remain valid.

Similarly, a researcher who critically evaluates an AI-generated literature synthesis may identify unsupported claims and reconstruct the argument from original publications.

These activities may contribute meaningfully to the research.

However, not every routine check constitutes a qualifying authorship contribution.

The significance of verification depends on its nature, intellectual demands, and relationship to the applicable contribution criteria.

Research teams should recognize substantive verification work without automatically treating every approval action as equivalent to intellectual authorship.

Why Contribution Records Should Be Maintained During Research

Reconstructing intellectual contributions after manuscript completion can be difficult.

Researchers may forget who proposed a methodological change, which collaborator corrected an analytical problem, or how an AI-generated interpretation was revised.

Records maintained during the project can provide a more reliable basis for later contribution statements.

Useful records may include research decisions, manuscript revisions, analytical documentation, code version histories, and agreed contributor responsibilities.

These records need not preserve every routine AI interaction.

The appropriate level of documentation should reflect the significance of the activity and applicable institutional or publication requirements.

For consequential contributions, the record should make clear which human researchers directed, evaluated, or approved the work.

How Should Research Teams Distinguish Human and AI Roles?

These examples illustrate possible distinctions rather than universal authorship rules.

A researcher's actual contribution may involve several activities, and some tasks may be performed jointly.

Contribution statements should describe what happened rather than force complex collaborative work into artificial categories.

Can AI Make Contribution Disputes More Likely?

AI assistance may introduce new grounds for disagreement when collaborators interpret generated work differently.

One researcher may regard producing an AI-assisted manuscript draft as a substantial contribution. Another may consider the intellectual work to lie primarily in conceptual development and critical revision.

Disagreements may also arise over who should receive recognition for an analytical method initially suggested by AI but subsequently refined by several collaborators.

These are plausible sources of dispute, although evidence is needed to establish whether AI use has increased the prevalence of authorship conflicts.

Research teams can reduce ambiguity by agreeing on contribution expectations early, recording substantive activities, and reviewing the allocation of roles as the project develops.

Where disagreements remain unresolved, institutional authorship procedures may provide an appropriate mechanism for review.

Why Contribution Transparency Matters for Research Integrity

The European Code of Conduct for Research Integrity emphasizes honesty, reliability, respect, and accountability.

Accurate contribution statements support these principles by identifying the people responsible for different aspects of scholarly work.

When AI assistance obscures who made consequential decisions, it may become harder to determine who should investigate an error or respond to questions during peer review.

For example, a coauthor may be unable to identify who verified an AI-assisted statistical analysis.

The problem is not simply inadequate credit allocation. It is a gap in the accountability structure of the research.

This connects to the need for collective accountability when AI contributes to collaborative research.

Reconstructing Contributions in an AI-Assisted Research Project

The example illustrates why AI-generated outputs cannot be used as direct measures of human intellectual contribution.

Researcher B may have performed substantial analytical work, but that cannot be established solely from generating code. Researcher C's verification and correction may also represent important intellectual contributions.

Researcher D's contribution depends on the substantive reasoning and revision performed, not merely the length of the AI-generated draft.

The final authorship decision requires application of the relevant criteria to the complete record of human contributions.

Misconceptions About Identifying Human Contributions

How to Make Human Intellectual Contributions More Transparent

Research teams should establish a practical method for identifying substantive human contributions throughout the project.

The objective is not to quantify every interaction with AI, but to make important scholarly decisions and responsibilities attributable.

Agree on Contribution Expectations Early

Collaborators should discuss anticipated intellectual responsibilities before substantial research work begins.

These expectations can be incorporated into research team agreements concerning AI use.

They should remain revisable because actual contributions may differ from initial plans.

Maintain a Proportionate Contribution Record

Research teams may document important conceptual decisions, analytical changes, code verification, manuscript revisions, and interpretive contributions.

Version histories, decision records, and contributor-role updates can support this process.

Documentation should not create unnecessary exposure of confidential prompts, participant information, or proprietary research material.

Review Contributions Before Submission

Before submitting a manuscript, collaborators should verify that contribution statements reflect actual work.

They should distinguish human roles from AI assistance and confirm that proposed authors satisfy the applicable criteria.

AI-use disclosures should be prepared according to relevant journal and institutional policies.

Preserve Accountability for Consequential Work

Researchers should know who is responsible for explaining and verifying consequential analyses, methodological decisions, and interpretations.

Where responsibilities are shared, the team should establish how questions or errors will be investigated.

Transparency is strongest when contribution records support both fair recognition and meaningful accountability.

Can Your Team Identify Its Human Intellectual Contributions?

Common Questions About AI and Human Intellectual Contributions

Make Human Intellectual Work Visible Even When AI Produces the Output

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.