AI at the Olympiad
Our scientists comment on the success of AlphaGeometry, the latest AI programme from Google DeepMind, in tackling geometrical problems.
Can AI do maths? The answer is yes—at least when it comes to solving tough geometry problems, set as part of the annual International Mathematical Olympiad (IMO). Developed by AI specialists at Google DeepMind, the AlphaGeometry programme correctly answered 25 out of 30 such questions put to it, according to a paper in Nature. These results are “stunning and breathtaking”, said Gregor Dolinar, the president of the IMO.
Asked to comment by three separate publications, scientists at the London Institute sounded a more cautious note. For instance, Mikhail Burtsev told the FT that the "steeper challenge" will be if AI can “discover new mathematics to solve a question that has never yet been answered”.
In the New Scientist, Prof. Yang-Hui He observed that, to reach this milestone, AlphaGeometry must have access to far more mathematical knowledge, and be able to tell when a theorem is provable or interesting. In any case, Thomas Fink argued in The Times, AI lacks the near-mystical instincts for deep mathematical insight. “Contrary to popular belief, advancing mathematics requires a great deal of intuition.”