Generated papers now compete with human work for journal space.
AI represents a multi-fanged threat to academia. It makes it trivial to spam the system with slop that passes surface checks... competing out the human-made academic slop that filled many journals and provided for the careers of many.
Real intellectuals ignore all of this and just go for truth or originality. But they rarely thrive in the academic world, where institutional incentives will be against them.
Users praise points linking AI slop in academic papers to institutional sloppiness and express optimism that anti-slop countermeasures could succeed.
No Digg Deeper questions have been answered for this story yet.
@s_r_constantin The vulnerability to AI slop is a wonderful measure for the preexisting sloppiness of an institution
I like AI, but I think institutions like Arxiv and the Granta competition need to be much tougher about cracking down on AI slop.
The real horror will start when Google Scholar gets tired of crawling/extracting, goes for AI Assist, and starts exporting corrupted/hallucinated citations.. 🤞
And the only one that can tell the difference is @anshulkundaje
It may add new kinds of noise: errors of kinds not found before, and harder to remedy. Hallucinated citations are not too bad, but automated citogenesis, emergent academic fashions and AI-perpetrated research fraud are obvious things. It is the non-obvious ones that will bite.
AI represents a multi-fanged threat to academia. It makes it trivial to spam the system with slop that passes surface checks... competing out the human-made academic slop that filled many journals and provided for the careers of many.
It also helps doing checking, potentially also revealing a lot of errors, mistakes and outright fraud. That has not even begun to bite, but may become even more devastating.
It may add new kinds of noise: errors of kinds not found before, and harder to remedy. Hallucinated citations are not too bad, but automated citogenesis, emergent academic fashions and AI-perpetrated research fraud are obvious things. It is the non-obvious ones that will bite.
AI undermines many forms of gatekeeping by making it easier to make surface-level or actually good outputs even for outsiders. Sure, way more nicely formatted crackpot papers, but also many more independent scholars.
Much of academia (like many human endeavors) is about maintaining status by claiming relevance, important truths, etc. and then gatekeeping access. At best this produces quality assurances and emergent epistemics, at worst just pompous academics.
As for checking references, that is at least something we can use AI to improve cheaply on top of our own ability... assuming our software can legitimately access the academic literature! As long as it is locked up, we should have lower hopes for human and AI truth-seeking.
I think the deep cruxes (beyond academic self-interest and defending self-image) are whether one cares whether knowledge is human-made, and whether academic structures (as they are and as traditions) have intrinsic value.
AI also provides ungated access. I might not know the finer points of ophthalmology but I can ask a cheap expert. It is like asking a grad student in a bar: not entirely reliable, but if you don't have easy access to serious experts it sure beats the alternative. Competition!
@Ronalfa dont worry I have the fix Turns out everything needs Validation now
@Ronalfa The fix is boring and works, link claims to sources, share code and data, and prereg confirmatory tests. If a launch or paper cannot meet that bar, treat it as marketing, not science.
@Ronalfa But marketing
@Ronalfa I think many others can and do but don't want to say anything publicly (which is understandable but unfortunate as well)
Much of academia (like many human endeavors) is about maintaining status by claiming relevance, important truths, etc. and then gatekeeping access. At best this produces quality assurances and emergent epistemics, at worst just pompous academics.
@rao2z Your "vibe CV"
@Plinz @s_r_constantin The same applies to individuals.
@Ronalfa @anshulkundaje Let's train an LLM on him and scale!
@Plinz @s_r_constantin That’s actually a very good point
@Ronalfa There is a difference between sophisticated insightful perspectives written with AI, and slop.
Your post uncritically treats it the same.
@Plinz @s_r_constantin I’ve been telling everyone about this “tracer dye for bs” phrase
