I just had a fascinating discussion with a buddy who works in the non-theoretical area of AI/ML and is very successful. They made a pretty bold claim about authorship in publications:
“In AI/ML, it’s basically first author or nothing.”
I’m likely to take this person’s perspective seriously since they are from a prestigious university and have over 2,000 citations. They said, “Sometimes beyond third authorship, they don’t even touch the codebase.”
I’m interested in what other people think about this. Is it accurate to say that in AI/ML publications, only the first author is given substantial weight? In what way does this differ from other fields?
Has this happened to you at work or in your studies? Any advice would be very appreciated, particularly from those who are now employed in the business or in academics.It’s possible that this could be the case.
7 Likes
Only I touched the code in a three-author study (ECCV paper) where I was the first author. In order to target the ideal framing and set of tests for selling my concept, the second author provided a great deal of assistance with ideation and framing. They were essentially vital. They know what sells since they have extensive experience chairing and evaluating for prestigious conferences.
My PI didn’t do too much, other than provide broader guidance at a weekly check-in and helped structure some parts of the paper and helped with figures. All incredibly critical to the success of the paper but the bulk of the work, coding, and the idea came from me.
From the standpoint of recognition, he’s not entirely incorrect. however untrue.
Consider the Transformers paper, where every person is familiar with each other.
Essentially, the only trustworthy measure of effort is found on the first author. That was almost probably the individual who worked the hardest on the project. The second author’s effort might be anything from working in the trenches to just offering advice at the water cooler since we only conduct ordinal rating. It’s possible that the last author gave the instructions or that they are unaware of the topic of the article.
as a benchmark for proficiency in an ML-based position? Most likely, yeah. Ironically, this is also the reason why, in my opinion, this isn’t sustainable—there aren’t enough articles available.
No, for a tonne of other metrics. People seldom have the opportunity to read a work in its entirety, therefore at some point throughout the decision-making process, they will base their opinion on things like co-authorship at prestigious venues, citation counts in your Scholar account, etc. Having co-authorships in reputable publications increases your overall trend of publications and, therefore, you as a researcher.
Basically, it depends on your circumstances. It’s possible that just contributions made by the first author count for a PhD. You should strive for final authorship if you are a postdoc or an early professor.
Reputable senior writers with a lot of experience in the subject will be at ease with their own role in the work and will decline co-authorship if they don’t believe they added anything.
The project’s first author is its most crucial contributor. Since the last author is the senior researcher and the one who suggested the larger direction and confirmed the accuracy of the work, they are more significant than the intermediate authors. However, as others have noted, this is only important if you want to know who was involved in a particular study; otherwise, for a more comprehensive picture of a researcher, they would just look at your profile, citations, or h-index, where it doesn’t matter how many papers you ranked first, middle, or last.
Only my teachers were the second and third authors on any of my articles. read the work and offered ideas for improvement; nothing was changed. Unfortunately, scholarly papers really state this.