AI Superintelligence
I've been reading Noah Smith's blog for years. I enjoy more of his econ posts than political posts, but it's been interesting.
Recently though, more of his posts have been around AI (whose aren't?). Noah is pretty positive towards AI and the abilities of AI models. He's made reference a number of times to AI "super intellegence" and the ability to solve tricky maths problems as evidence of that.
Specifically, he points to OpenAI claims that AI solved a very hard problem in 88 hours.
I'm not in a position to read any mathematical proof nor am I even able to understand the problem that's been claimed to have been proved.
In 2006, I read Fermat's Last Theorem by Simon Singh. It's a popular science book about a theorem devised by a 17thC lawyer called Pierre de Fermat. Fermat claimed to have found a proof to the problem, but it wasn't proved for 300 years until a chap called Andrew Wiles provided a proof. Wiles worked on the proof for six years and revealed his proof in 1993.
It then took another two years for the proof to be formally pubished and accepted because of some flaws.
I suppose the punch line is: how do we know that OpenAI's proof of Navier-Stokes problem is sound? Did we ask Copilot?
Whilst I am being facetious, I do wonder how much we, as humans, understand some of these problems, much less the solutions. There comes a point that even though I am atheist and "believe" in science and the scientific method, I have to defer to a higher authority. In some ways, if feels like we all believe in God to a greater or lesser extent.
Is believing in the Big Bang any different from believing in a divine creator? I'm not sure it is.
Existenial crisis aside, AI producing a solution to a hard maths problem fast, doesn't seem particularly impressive given how much AI can produce right-looking answers AND how difficult it is to verify hard maths problems.
Or maybe I'm just an AI cynic and letting that bias cloud my judgement.