By Ana Marušić

In their editorial on generative AI and authorship, Brian Earp and colleagues offered replacing authorship with the concept of contributors as a possible solution. My first reaction of this proposal was: we have been here before.

In 1997, Drummond Rennie and colleagues proposed such a split. They argued that an “author” does two jobs at once – handing out credit for a contribution and assigning accountability to someone who can stand behind the work and answer for it. They proposed that that anyone who contributed to the work should be named as a contributor, and a smaller group of guarantors would take formal responsibility. It never really caught on, but AI has brought the same tension back. This time, I think it should be taken seriously.

Here is the argument I make in my commentary: if accountability is the real reason to grant authorship, then a contributor who dies before the manuscript is finished cannot be authors on a published article as they can no longer approve anything or answer to a journal editor. But journals keep deceased authors on the byline, without controversy. That tells us that accountability was never really why someone deserved credit for the work/manuscript. Rather, accountability is something that is layered on authorship after the work is done, as a practical safeguard for readers and editors. The actual basis for credit is retrospective, to recognise that someone did the intellectual work.

Once you see it that way, the question of AI authorship gets easier, but not entirely comfortable. An AI cannot be held accountable so it cannot be an author – that part is straightforward. But the argument that “it cannot be accountable” is not a sufficient reason for withholding credit in the first place, as illustrated by the practices around deceased authors. So, if an AI system does something substantial, like proposing a hypothesis, drafting an interpretation, flagging an error that changes the conclusions, the fact that it cannot answer for the work is not a reason to hide that it happened.

I want to be careful here, as the analogy to a deceased author is not perfect and cannot be stretched too far. A person who died once had a stake in being recognised. This interest is protected by posthumously giving credit to the person. An AI model has no such interest as it does not care whether it is mentioned. The case for disclosing a substantial AI contribution is not about owing AI anything. It is about the honesty to the reader, who has a reasonable interest to know how the research and the article came together.

However, I have concerns about disclosing AI contribution. They do not come from the AI debate itself but more from my experience running empirical studies on how authorship gets decided. Over the years, our studies asked students to resolve authorship disputes, asked authors why they thought they deserved credit on an article, checked whether researchers’ answers about their contributions stayed consistent over time. They did not, not reliably, thus leaving authorship more in a moral rather than a conventional domain.

This leaves me with a caution: even if we agree that AI contributions should be disclosed, getting humans to disclose anything accurately, consistently and without self-interest has proven genuinely difficult. Adding a non-human contributor to that mix will not make the bookkeeping of authorship/contributorship any easier. The moral case for disclosure of AI contribution is fairly clean but building a system that reliably captures it is a much harder problem that we have not yet solved even for ourselves.

Paper title: Forward to the past: let’s revive contributorship

Author: Ana Marušić

Affiliation: University of Split School of Medicine, Split, Croatia

Competing interests: None declared

Social media: X: @Meta_Split_AM