A ChatGPT author
Publishers follow different methods to deal with errors in the authorship of an article. One is to create a new version of the article with the correct list of authors. Another is to leave the article intact and publish an erratum in a later issue. Yet another is to do both.
Creating a new version of a published article is not a generally accepted practice, because it is seen as interfering with the history of science. With full transparency about older versions, clear change logs, and the CrossMark standard for marking versions, this should in principle be acceptable. But many publishers and databases resist it.
The erratum – or corrigendum – is a method dating back to the paper era. At the time, the only way to deal with an error was to broadcast a message in a subsequent issue. In the electronic age we started linking the erratum to the article, so that a reader could be notified. On a full-text publishing platform this is an acceptable method. But what about a secondary database?
Consider this case: an article was published with an AI tool listed as an author – a practice that several journals have since explicitly prohibited. The article's publisher subsequently issued a corrigendum stating that the AI tool is not an author of this article. On a full-text platform, linking the corrigendum to the original article is sensible: the reader can see both.
On a secondary database, the conventional rule is to follow what the publisher stated in the original article. The result: the AI tool remains listed as an author, and continues to appear in its author profile. The corrigendum achieved the opposite of what was intended.
What an erratum is for
I have always believed this is the wrong approach. Our mission is to build a representation of the world of research. A broadcast message has been sent stating someone is not an author. Our job is to act on what the erratum instructs us to do. That is what the erratum is for.
The fundamental problem is our mental model of the "record page" as something that represents the article as it arrived. In the evidence-based model, all the input – the original article, the corrigendum, any further updates – is treated as evidence from which the representation of the world of research is constructed. There is no single record that arrived and must be displayed as-is. There is a profile page that must represent our current best understanding of what happened.
The article profile page
A more useful way to think of what we need is an article's profile page: a page aimed at showcasing the article and all its details to the world. The profile page should present the article and the erratum together – it is a mistake to give errata their own separate pages, and errata's citation counts should not play a role in bibliometrics at all. The same principle applies to translations of articles: we should not spread the citations of one piece of research across two versions, diluting its true impact.
Such a profile page would contain:
- The details of all the publications that belong together, with explanations of the different exhibits and their provenance
- Related works: errata, translations, corrections, retractions
- Citation counts applied to the profile, not to individual versions
Searches for any of the related works would lead the user to the same profile. This also works for patent families, conference paper versions, and similar cases.
The profile page concept is a consequence of taking the evidence model seriously. It solves a cluster of related problems – errata, translations, versions, corrections – with one coherent approach.