Kevin Buzzard maps mathematicians' reaction to AI onto the five stages of grief

Not a mathematician, but I'd genuinely like to see what the AI comes up with here. Could be interesting.

Buzzard is such a good writer. Read it all the way to the end, it gets even better as it goes.

I share the optimism. Connecting dots across knowledge, lived experience and purpose is still a deeply human pursuit, and the more AI discovers, the more we get to connect those discoveries to questions that matter to us.

I don't see much evidence of AI connecting dots though. If there were dots to connect between a war and some geological event, the model has to brute force an O(N) scan at inference time. It never learned that during training, and I want to know why.

These idealistic pieces mostly come from senior, established or wealthy people who aren't exposed to the labor market. For labor this is a survival question, not a philosophical one, and it's landing when the scales are already tilted hard toward capital.

Most people who oppose AI don't think it's too good and will replace us. They think it's hype that doesn't work. First we'd have to agree it can actually do the work, then we can talk about protecting livelihoods.

Is mathematics research even the kind of job you keep pursuing when survival is the question?

That's right but fatalistic. Nothing is set in stone and physical leverage is always available, just at a steep price. A lot of the bad in the world runs on PR insisting resistance is futile so you do nothing.

The end of ZIRP and the arrival of AI at the same time was not ideal. As labor, the one thing I appreciate is that the level of bullshit inside companies dropped a lot. Everyone is hyper focused and results oriented now, though maybe that's just my bubble.

Labor won't get far by pretending this isn't happening or making moral arguments. These models don't get tired, they keep improving at rates humans can't match, and they're already good enough for most intellectual tasks. Asserting the primacy of meat-based thinking won't get us very far.

Calling governments bought out is cynical fatalism. They've been run by neoliberal zealots for decades, but that isn't a law of nature. If we don't want decisions concentrated in a handful of AI oligarchs, we need strong democratic institutions again.

AI doesn't destroy labor. If anything it makes the average laborer more productive.

What I've actually seen: programmers made redundant with code of dubious quality, QA and design the same, people replacing friendship with bots, people skipping their own brain to ask an LLM. Best thing I got out of it is search, since Google turned the web into an SEO hellhole.

The five stages of grief keep showing up in AI discourse. Programmers are at bargaining right now, some still at denial telling themselves it's hype like NFTs and the bubble will pop.

The real crisis in mathematics is a labor one. Mathematicians traded proofs for employment, that was the currency, and now the currency is gone. You can't argue anyone into a new definition of mathematics, you just follow your own curiosity and it sorts itself out.

Sounds like bargaining to me.

I'm an optimist too, though the labor side worries me. Current top models do at least seven different things at once: natural language, formal reasoning, informal reasoning, encyclopedic recall, filling gaps in vague instructions, acting as a proxy for a person, and translation. Untangling those will take decades of work, and that work is interesting.