Google’s AI unit Deepmind has demonstrated two new AI systems, AlphaProof and AlphaGeometry 2, which can successfully solve complex mathematical problems, representing a significant milestone in AI development, Reuters reports .
Today’s AI models work by being able to recognize patterns and statistically predict, for example, the next word that should be generated. However, this is not enough to handle abstract mathematics, which instead requires a system that can reason in a way that more closely resembles human intelligence.
AlpaProof and AlphaGeometry 2 succeeded in solving four out of six questions in this year’s edition of the International Mathematical Olympiad (IMO). One question was solved in just a few minutes; others took up to three days to solve, which is longer than the competition allows. AlpaProof also managed to solve the competition’s most difficult problem, which only five of 600 human participants managed to do.