AI-DEMIA

Why does it feel that the leaders of AI companies aren’t expressing their awareness of how afraid we are? Jensen Huang provided us this pseudo comfort: “You’re not going to lose your job to AI. You’re going to lose your job to somebody who uses AI.” What he may not realize is that “somebody” may someday be a no-body; a non-human. First you would lose your job to some being who uses AI, and they will in-turn lose theirs to fully autonomous AI. Arguments like these give an optimistic view of the immediate future, but fail to carry the point to the next likely future. 

Do we believe these tools will truly empower us? The latest conference I’ve applied to saw over 6000 research paper submissions; clearly a consequence of AI speeding up the implementation, evaluation, data interpretation, and writing processes. How on earth will the reviewers handle this, even with the possibility of excess desk-rejections? One would have to be living in the clouds to believe they wouldn’t also use some form of AI support to keep up with the effort it takes for triage and review. 

Papers that make it to the finish line are scrutinized by humans since these works become state-of-the-art, and should reflect high research quality. I wonder if reviewers don’t feel a sense of futility in picking apart works that may have been completed mostly by AI. So… if AI is multiplying researcher “productivity” enough for reviewers to need it, why not just remove the human from the loop?

I propose AI-Demia, or let’s fast-forward academia with game theory. Here’s what we do: encode conferences as a set of acceptance patterns based on their prior academic publications. Model publication incentives. Model as much of the rules and context as possible. Then, design the digital agents to be the researchers and reviewers working in a specific research direction that – at this point – does not require physical work. Next, give the virtual researchers access to AI and the internet, and have them come up with the next best works based on some strategy. Strategies with varying assumptions can be initialized to observe the different innovations and research directions favored over many simulations. Let their research be scrutinized by virtual reviewers with access to the same tools. Run this until “we” invent free energy.