Yesterday a mathematician at Anthropic announced that, working with Claude, he believes he has proved the Riemann hypothesis, the 167-year-old problem many consider the most important in pure mathematics. (Nobody has checked it yet, but the proof is written in Lean, so a computer can verify it line by line.) If it holds, it comes barely a month after OpenAI released hundreds of AI-generated proofs, including a partial result on the same problem. People are calling it the "mathpocalypse". It's not hard to see why.

Trinity's mathematicians are some of the best of their age in the world, and they're taking it in different ways: some are slowly losing hope of pursuing a PhD; some insist there's still enough for humans to do, though they struggle to say what; some dismiss the AI proofs as ugly, saying they don't give us any understanding; and most carry on with their problem sheets as if nothing had happened.

But something has happened. AI "solved" coding first, which was easy enough to see coming, and now it's mathematics. What's next?

That raises a bigger, more frightening question. If we assume AI will keep becoming more capable (and given the current investment and rate of progress, I think that's a reasonable assumption), what will be left for us humans to do?

Results and experiences

I think the answer hinges on a distinction: (1) some things we value for their results (working code, say, or a diagnosis); and (2) others we value for the experience of doing them. AI is getting very good at results, and I see no reason it won't keep beating us there, in maths and then in whatever comes next. Maybe that's for the better: nobody would complain about a cure for cancer, after all.

Experiences are different. It might seem too obvious to state, but it's worth keeping in mind that we are not AI, and AI is not us. AI can't feel our grief or joy for us, or the cold on a morning walk, or what it's like to be consoled by a friend. I'm not making a claim about AI consciousness. Maybe AI will experience things one day. Maybe, to some extent, it already does! But even then, its experiences wouldn't be mine. I still have my own life to live, to enjoy and to suffer, and no machine can live it for me.

That's why chess didn't die when Deep Blue beat Kasparov in 1997. In fact, more people play now than ever. If we played only for the quality of the moves, we'd have given up two decades ago. Instead, we play for the concentration, the opponent across the board, the pleasure of an idea that's ours. The same goes for my attempts at Greek verse. Claude could probably (/definitely) write a better line of dactylic hexameter than I could, but the point was never the line: it's what goes into making it.

Maybe some of mathematics will go the same way, done for the love of understanding and not to push the frontier. I suspect that's already why most of the mathematicians here do it. And consider our bodies: AI can design the drug, but it can't do the exercise for us, at least for now.

The catch

There are, of course, practical problems with living this way. Most of us aren't paid for our experiences. Rather, we're paid for results; and results are exactly what AI is taking over. It doesn't matter that I'd happily spend my days writing Greek verse, if nobody is going to pay me for it.

And I don't think AI can answer that one for us, because it isn't a technical question. If AI does the work, the wealth it creates goes to whoever owns it, unless we decide otherwise: through taxes, a basic income, or whatever else we come up with. That's a choice about what kind of society we want, a political question.

(Of course, this assumes money still means anything. In a world where AI does all the work, it might not, and that's before considering the possibility that AI ends up making these decisions instead of us.)

What we want

In any case, there's one thing I think we're still best placed to work out: what we want. And if AI does end up making the decisions, that's what we'd need it to want too. I don't doubt that AI will become wiser than us in many ways. It may even give better answers to "how should one live?" in general. But for my own life, there's no better answer than mine, because a good deal of what makes an answer right for me is that I arrived at it myself. A life lived by someone else's answer, however wise, wouldn't be my life. That's the theory, anyway. In practice, knowing what we really want is getting harder. Anyone who has lost an evening (or, who am I kidding, most evenings) to their phone knows how easily our wants can be shaped for us.

My own answer

So what do I want? My best answer, for now, is this: to do something only I could do, and to look back and say "I was there, thinking about all this as it happened."

That's partly why I'm happy reading Classics, even though, given I want to get into AI, it might seem the least practical choice of all. I chose it partly because there's no pressure for it to be useful. And because it's all of the humanities in one (language, literature, history, philosophy), it keeps leading me to unexpected places. Philosophy is one, and philosophy is what got me thinking about AI like this. So for now I'm reading, going to philosophy lectures and keeping up with AI, waiting to figure out my own small way to contribute.

I'm not anxious about it, mostly excited: if the machines are going to do the maths (as they've already demonstrated), I'd like to help think about what they should be for, and what we're for.