Authorship Capabilities of AI Models

Determine whether generative AI models such as ChatGPT possess the capacities for semantic understanding, consent, and accountability for the text they generate, sufficient to satisfy academic journal authorship criteria that require understanding of all written content and the ability to take responsibility for it.

Background

In response to early instances of listing AI systems (e.g., ChatGPT) as co-authors on scholarly articles, many publishers adopted policies disallowing AI as authors while encouraging acknowledgment of AI tool usage. These policies commonly reference standard authorship criteria, including the need for an author to understand the manuscript’s content and to assume responsibility for it.

The paper highlights an unresolved issue underlying such policies: whether AI systems can meet the authorship requirements of understanding and responsibility. Clarifying this point is essential for determining if and how AI systems could ever be credited as authors under prevailing scholarly norms.