Hacker News Reader

Comment by eviks | Hacker News Reader

Check the article after this quote "Their tricks are so crude that anyone who takes a look can easily spot them"

As I'm sure you know, there is a big difference between something a human can do when they pay attention to something, vs a completely automated procedure that can be done at scale.

Yeah, I didn't catch that you have a strict requirement for full flawless automation (which the current system fails to satisfy), I thought the adjusted algorithms/indices etc would still involve some humans before having an impact

I think there is huge potential to use AI for detecting many types of academic fraud. Citations of unrelated articles should be an easy one. Flag them and then refer to human university administrators for review.

You hardly need AI for this; you only need to construct a graph of researchers with the number of citations as weights, and compare metrics of connectedness to detect cliques and other fraudulent behavior.

Cliques can easily occur just by a few people researching the same thing. If you are maybe two or three research groups in the works that use a particular method to try to solve a particular problem, of course you are all going to cite each other excessively.

You're going to get a lot of false positives with such a simplistic approach. Sometimes the same group of researchers legitimately cites each other over and over because they're working on similar topics and collaborating. AI can help to reduce the false positives by looking at content to identify citations that might be out of place. Of course there will still be some false positives with any algorithm so human academic committees will still have to make the final decisions.

Unfortunately a citation, especially as a prestige or impact metric, carries associations of quality to observers outside its domain. A resource having been selected for citation implicitly signals that it wad the best choice among a field of rival candidates, yet in many cases such as you describe the truth is that it is the _only_ such work available.

This is especially problematic as some researchers build entire careers out of gatekeeping preprint access to unique, single- source primary data that only they or their lab/ staff can collect- not because of their superior ability to analyze it but because theirs is the only facility with the appropriate ensemble of apparatus on site or close at hand, or because they are the latest in a close-knit (often nepotistic) chain of custodians presiding over a certain exclusive primary source, such as a nature reserve or archaeological site.

Yes, but we all know that the "human review" is only going to be a temporary step on the road to furthering the dystopia of 0-accountability institutions that are capable of damaging your life on the imperceptible whims of the black box and the people who train and maintain it.

I think that instead of turning over every opportunity for a human job to robots that will fail and be captured, we should be giving people those jobs. The kind of flesh bodies that are afraid of being jailed for years at a time.

Reply on news.ycombinator.com