Which Contributions Deserve Credit? Perceptions of Attribution in Human-AI Co-Creation

TL;DR

The paper tackles the challenge of attributing authorship in human-AI co-creation, showing that credit is not binary but varies by contribution type, amount, and initiative. Using a $2\times3$ factorial, scenario-based survey with $N=155$, it demonstrates that AI is consistently credited less than humans for equivalent inputs, even when AI contributes substantively. It identifies factors shaping attribution decisions, including originality, quality, and human-led accountability, and it proposes design and policy avenues—such as granular AI contribution statements—to improve transparency and alignment with creator values. The findings underscore the need for AI-specific attribution frameworks as transparency requirements expand across domains.

Abstract

AI systems powered by large language models can act as capable assistants for writing and editing. In these tasks, the AI system acts as a co-creative partner, making novel contributions to an artifact-under-creation alongside its human partner(s). One question that arises in these scenarios is the extent to which AI should be credited for its contributions. We examined knowledge workers' views of attribution through a survey study (N=155) and found that they assigned different levels of credit across different contribution types, amounts, and initiative. Compared to a human partner, we observed a consistent pattern in which AI was assigned less credit for equivalent contributions. Participants felt that disclosing AI involvement was important and used a variety of criteria to make attribution judgments, including the quality of contributions, personal values, and technology considerations. Our results motivate and inform new approaches for crediting AI contributions to co-created work.

Figures (5)

• Figure 1: Means and 95% confidence intervals of authorship credit scores across all writing partners and writing contexts for (a) contribution type, (b) contribution amount, and (c) initiative. We observe that ratings generally favor the self for contribution type, ratings are distributed across the spectrum for contribution amount, and ratings are polarized for initiative.
• Figure 2: Means and 95% confidence intervals of authorship credit scores across human and AI partners for different types of contribution. X-axis labels correspond to Self credit / Partner credit (details in Table \ref{['tab:credit-assignment-scale']}). Types marked with an asterisk (*) indicate a statistically significant difference between human and AI partners (details in Table \ref{['tab:authorship-summary']}).
• Figure 3: Means and 95% confidence intervals of authorship credit scores across human and AI partners for different amounts of contribution. X-axis labels correspond to Self credit / Partner credit (details in Table \ref{['tab:credit-assignment-scale']}). Amounts marked with an asterisk (*) indicate a statistically significant difference between human and AI partners (details in Table \ref{['tab:authorship-summary']}). The "no writing" condition is omitted from this chart since it was used as the attention check, and all filtered responses selected "Sole/n."
• Figure 4: Means and 95% confidence intervals of authorship credit scores across human and AI partners for different levels of initiative. X-axis labels correspond to Self credit / Partner credit (details in Table \ref{['tab:credit-assignment-scale']}). Levels marked with an asterisk (*) indicate a statistically significant difference between human and AI partners (details in Table \ref{['tab:authorship-summary']}).
• Figure 5: Design exploration for crafting an AI attribution statement, modeled from the Creative Commons License Chooser cc2024license. On the left, the user specifies (1) whether AI was used to create this content, (2) which AI model or application was used, (3) the types of contributions AI made, (4) the proportion of work created or modified by AI, and (5) the initiative taken by AI in creating the work. Then, the user (6) indicates whether AI-generated content was reviewed and approved by a human. On the right, an attribution statement -- AIA model-name CeNc PAI Pm R 1.0 -- is created based on the user's selections, with AIA indicating the start of the AI Attribution statement and 1.0 indicating the version number of that statement. A longer attribution statement and explanations are also produced to more clearly explain the AI's contribution. Attribution statements can then be used to label co-created work, such as the statement we include in Acknowledgments.