Then, I got another message. OpenAI was giving free access to frontier models to researchers at eligible institutions.

To me, this is the really upsetting thing. I'm at a liberal arts university, and have an active research program. This shuffles me solidly into the "have nots" category of mathematical researchers, even worse than not working at a research institution has in the past.

In one sense I understand. It may do bad things for tuning these models if people at institutions like mine participate. Fair, I suppose...and these are private companies that are trying to establish hegemony over competitors using mathematics research as a signal for investors.

The reason I'm here on MO this morning is to ask:

Question: What are we going to do about AI leading to an increase in inequity for mathematical researchers?

I realize this may be a very silly question, in the sense that if one views math as a science (for the purpose of analogy) requiring equipment, then it's completely analogous to asking why my small liberal arts college shouldn't be given its own particle collider.

There may be, however, an important difference. If Gowers is correct, the value of mathematics is as a widely distributed system of investigators. Indeed, it's perhaps more the fact that these investigators are all very cheap that makes this so valuable, but I think it is more that there is such a wide diversity of participants investigating so broadly. If it is possible for leading researchers to explore mainstream problems using powerful AI, this leaves little or no "cleanup" for the rest of us to do. There is very little incentive to even enter the field, and the diverse and broad participation is very likely to dry up.

The question for the community is whether changing this in light of AI models making it possible for a few of us to investigate more quickly is acceptable to us. If it is not acceptable, I'd like to know how we are going to protect equity in math research.

Edit: It occurs to me that it may be in the interest of the research community to want restricted access to these models in order to incentivize aspiring mathematicians to "earn" a place at an institution which has access to them. This would continue to allow for a tight quality control over AI use, keeping it in the hands of few and thereby preventing a flooding of publications from less trusted mathematicians. This is important to consider, since proof abundance and effective dissemination of new math is going to be a problem. This wouldn't prevent mathematicians from practicing mathematics the way we always have, or even augmented by less powerful or expensive models, but it definitely would create an even larger divide between the "haves" and "have nots". Indeed it's clear that baseline here wasn't equity (thanks Sam Hopkins for noting this), but do we really want to make this worse?

In weighing this, please take into account the claim of Gowers that the diverse and broad participation of mathematicians is a central value of the subject and a big reason to support it. In the event that all have access to frontier models, and thus an equal opportunity to have the force multiplier for investigation, since researchers who are better positioned in a traditional way (in better research groups/less isolated) will have the ability to resolve the current long standing conjectures and move into novel territory consequent to these conjectures, the right strategy for the general mathematician is to go for "blue ocean" problems, thereby expanding the mathematical landscape in even more creative ways.

It seems to me that the choice on how to deal with this might hinge on how we handle the adjudication of mathematical claims. If we make this more broadly available, expansive exploration (and I'd say mathematical progress) will be more likely, but it may accompany significant difficulty in evaluating novel claims.

(http://www.autoadmit.com/thread.php?thread_id=5890366&forum_id=2…id#50051385)