The International Mathematical Olympiad is not only a terrifying litany of words for a math moron like me, it is also a world competition in mathematics in which high school students from over 100 countries compete. Each year, students compete in selected host countries to showcase their mathematical prowess, each aiming to solve problems that would make the rest of us cringe in horror.

Google DeepMind announced that it took on this year's contest problem as a combined system of its two AI systems, AlphaProof and AlphaGeometry 2. The AI had its answers scored by previous gold medalist winners, Professor Timothy Gowers and Dr. Joseph Myers, the latter chairing the IMO 2024 problem selection committee.

AI not only scored 28 out of 42 points, one point shy of the 29 needed for the gold medal, but also achieved a perfect score on the competition's most difficult problem (via Ars Technica), as two combination problems were unsolved. Still, let's focus on what we're good at.

However, I am a little concerned. In a Twitter thread, Professor Timothy Gowers noted that while the AI did indeed score higher than many, it took much longer than its human competitors to do so. Human test takers submitted their answers in two four-and-a-half hour sessions, while the AI solved one question in a few minutes, but took as long as three days to solve the others.

“If human competitors had been allowed that kind of time per question, they would undoubtedly have scored higher,” wrote Lord Gowers.

“Nevertheless, (i) this is far beyond what the automatic theorem provers were able to do before, and (ii) it is a very good idea.

Not only that, but the AI didn't sit down in front of a test paper and start chewing on a pencil. The questions were manually translated into Lean, a proof assistant and programming language, so the automatic formalization of the questions was done by old-fashioned humans.

Still, as the professor points out, what AI has accomplished here is far more complex and subtle than simply solving problems brute-force:

“We may be approaching having a program that allows mathematicians to get answers to a variety of questions

“We may be getting close to having a program that allows mathematicians to get answers to a variety of questions.

[18] “Are we getting to the point where mathematicians are redundant? I suspect we are still missing that one or two breakthroughs.”