Like many other mathematicians I have been grappling with how AI is going to affect the future of our discipline. I am at a really junior stage of my career and there is so much uncertainty around these things, and there is also a great deal of pressure, coming from many people in mathematics, even to highest level, to start making AI an active part of our research process. Inspired by this post, all of whose points I agree with, I have come to a decision that I am going to stop all interaction with AI. I have just cancelled my $20 a month duck.ai subscription.

I had largely been playing around with AI out of morbid curiosity, to try to test just how good it's become at mathematics, and in particular to see whether or not it is able to do things in my area, model theory (I found the AI versions I was using on duck.ai, Claude 4.6 and Claude 4.8, largely unable to talk competently about advanced topics in model theory, although I have been informed by several people that in fact the more expensive and advanced models are really actually able to). In my most recent paper I used it for a very specific thing that fell far outside my expertise (I explain the use in detail in my AI declaration in the paper), and interacting with the AI about that question led me to a reference of something in the literature that did precisely what I needed. It was a bit remarkable.

But I am absolutely convinced that using AI more in my research would destroy the biggest things I love about mathematics, which are (1) the deeply human and social aspect of it, and (2) the sensation of personal pride that comes from making a big breakthrough after spending months consumed by a question. (Let alone years... I am too young to have experienced the feeling of spending a decade thinking about a problem and finally resolving it!) I am aware it may perhaps be a bit damning for my career to not use it, but I think I would rather leave mathematics than be forced to interact with AI.

On point (1), the idea of replacing interactions with humans by interactions with a computer is something that removes essentially all of the appeal of mathematics for me. As an example, regarding the AI contribution to my work that I allude to above: it is true that AI probably allowed me to find the reference I needed many orders of magnitude faster than I would have been able to pre-AI. However, in the pre-AI world, it is something I would have asked an expert in topological dynamics about, and discussing it with them would been a vastly more enriching human experience. Mathematics for me is a completely social endeavour, and I struggle completely to see what the point of mathematics research is if those social aspects begin to be lost.

Point (2) is surely an almost universal experience among mathematicians. There is enormous gratification and personal satisfaction in the sensation of having resolved a difficult question and obtained a fantastic result, and knowing how hard one worked to do it. I have heard some people try to argue that this is just an artifact of our attachment to our egos, and maybe there's a certain amount of truth in that, but nonetheless, for me at least, it would cheapen and devalue my sensation of accomplishment and pride entirely if I substituted my long periods of being stuck in great frustration, turning a problem over again and again in my mind, by interactions with AI.

None of this even touches on the enormous societal and environmental damage AI will have more broadly, which is far outside the scope of anything I can talk about here. But I am including this declaration here as a blanket statement about my research, which is that moving forward I will not interact with AI on it.