Google's DeepMind Says Its AI Solved a Math Problem That Humans Were Stumped By

 

DeepMind says that for the very first time, an AI has cracked a famously hard math problem with a solution that eluded human mathematicians — which could be great if it holds up to scrutiny.

In interviews with MIT Technology Review and The Guardian, Google DeepMind scientists waxed prolific regarding their latest AI tool, which they says has created a brand new solution to what's known as the "cap set problem," which includes placing more and more dots without any of them ever creating a straight line.

The novel discoveries, which the scientists declared in a paper issued in the journal Nature, would mark the first time AI has made a rare scientific invention which, because it was lastly unknown, was not part of its training data. 

That would be a very huge deal considering that AI is known for conjuring up nonsense and made up useless even when its training data has the correct solutions.

DeepMind made the tool in question, called "FunSearch" in reference to mathematical operations (and not the other kind of fun) on the back of its AlphaZero AI, which cracks math problems as if it were playing a game. 

The LLM it uses is called Codey, which is trained and honed on computer code and programmed to deny wrong solutions and give right ones back into its model.

Feeding code into an AI is one thing, but having it spit out a brand-new solution to a known riddle — even though it took a few days, as MIT Tech points out — is a completely separate thing.

There is something of a mystical quality to what the DeepMind researchers are saying: that the LLM managed to — just maybe — think for itself.

"To be very clear with you, we have hypotheses, but we are not aware properly why this works," DeepMind scientist Alhussein Fawzi told MIT Tech. "In the starting of the project, we weren't sure whether this would function at all."

While there will obviously require to be lots more study to inspect the claims and try to understand exactly how FunSearch created its novel solution to the cap set problem, its producers are clearly stoked.

"When we begun the project there was no idea that it would create something that’s really new," Kohli told The Guardian. "As far as we know, this is for very first time that a real, latest scientific finding has been created by a big language model."