AI has already begun impacting all manner of fields, but there are some that could fall before the rest.
Former Google CEO Eric Schmidt believes that mathematics and programming could be the first fields in which AI will become significantly more capable than the best humans. “Problems where the data already exists or where the verification is very easy will be the problems that are solved first. And there are two that are quite obvious. One is computing, that is programming, and another one is math,” he said in an interview.
“Because with a computer program, you can keep generating programs until you find one that actually works. You know what success looks like. And with math, you can keep generating proofs until you find proofs that are proven, because we know how to prove proofs,” Schmidt said. Schmidt seemed to be saying that once AI is close to human-levels at these fields, it can just keep grinding away at problems day and night without getting tired or demotivated like humans. This could mean that it’ll eventually reach breakthroughs that humans weren’t previously able to achieve.
But Schmidt says there is also something unique about programming and math that makes them ideal for AI to master. “Also, the language of those two sub disciplines of science are relatively simple. They don’t require reading all of the novels of the world, whereas human language is much more complicated. So we expect gains in the next year to two years, which will produce superhuman mathematicians, superhuman programmers, at least in some areas,” he continued.
It’s an interesting point. While ChatGPT had initially wowed people with its writing ability when it was first launched, its written outputs currently haven’t yet come close to bettering human creativity — it’s been two years since ChatGPT was released, but there have been no best-selling books written by AI. On the other hand, AI is rapidly getting very good at math and programming. OpenAI’s latest o3 model scored a 2727 rating on the Competition Code test, placing it at par with the 175th-best competitive human programmer — this means that AI is, at least at competitive programming, better than all but 174 human beings. Also, on the Competition Math test, which is a feeder exam for the US Math Olympiad, o3 had an accuracy of 96.7 percent, which is getting quite close to a perfect score. And if these trends continue, Eric Schmidt says that it might not be long before the best programmers and mathematicians around aren’t humans, but AI models.