Google DeepMind's AI systems earn 'silver medal' at International Math Olympiad

Google DeepMind has announced its AI systems solved four out of six problems from this year's International Mathematical Olympiad (IMO), achieving a score equivalent to a silver medal.

The company claimed that this is the first time an AI has reached this level of performance in the prestigious math competition.

The feat was achieved by Google's AlphaProof, a new reinforcement-learning based system for formal math reasoning, and AlphaGeometry 2, an improved version of its geometry-solving system. 

"Together, these systems solved four out of six problems from this year’s International Mathematical Olympiad (IMO), achieving the same level as a silver medalist in the competition for the first time," Google said in a blog post.

The test scores were determined by Sir Timothy Gowers, an IMO gold medalist and Fields Medal winner, and Dr Joseph Myers, a two-time IMO gold medalist and Chair of the IMO 2024 Problem Selection Committee.

Google said its combined system earned 28 out of 42 possible points, just shy of the 29-point gold medal threshold which is a perfect score on the competition's hardest problem, and which only five human contestants solved this year.

"The fact that the program can come up with a non-obvious construction like this is very impressive, and well beyond what I thought was state of the art," Sir Timothy Gowers said.

According to Google, AlphaProof solved two algebra problems and one number theory problem, and AlphaGeometry 2 cracked the geometry question.

Two combinatoric problems remained unsolved. Google claimed its model solved one problem within minutes while others took days.

To make them readable, Google first translated the IMO problems into formal mathematical language for its AI model to process, which is different from the official competition, where human contestants work directly with the problem statements in two 4.5-hour sessions.

AlphaGeometry 2 could solve 83 per cent of IMO geometry problems from the past 25 years, up from its predecessor's 53 per cent success rate, before this year's competition.

Google said the new system solved this year's geometry problem in 19 seconds after receiving the formalised version.

Sir Gowers in a thread on X expounded on the limitations of the AI and concluded that however impressive the AI's performance, it was far from having "solved mathematics."

"The main qualification is that the program needed a lot longer than the human competitors for some of the problems over 60 hours and of course much faster processing speed than the poor old human brain," Sir Gowers wrote on X.

"If the human competitors had been allowed that sort of time per problem, they would undoubtedly have scored higher.

"Are we close to the point where mathematicians are redundant? It's hard to say. I would guess that we're still a breakthrough or two short of that," he added.

Image Source: Unsplash