A new Artificial Intelligence (AI) model was released by Dr. Trieu Trinh: AlphaGeometry. This system can solve geometry problems from the International Mathematical Olympiad at almost the same level as a human gold medalist.
On January 10, 2024, Dr. Trinh defended his doctoral dissertation on AlphaGeometry at New York University. He originally pitched the AI system to research scientists at Google and became a resident. AlphaGeometry is just one of Google’s AI systems; it joins the Google DeepMind fleet. While other AI systems such as AlphaZero for chess have been developed, AlphaGeometry is significant as math has almost infinite ways to reach a solution. On the other hand, chess remains finite.
The International Mathematical Olympiad (IMO) is a highly esteemed competition for the brightest high-school students across the globe. When AlphaGeometry was tested in the competition, it solved 25 out of 30 Olympiad questions within the time limit. On average, a bronze medalist solves 19.3 problems within the limit, while silver and gold medalists solve 22.9 and 25.9 problems respectively. AlphaGeometry, at 25 problems, is only slightly behind the gold-medalist standard.
This breakthrough remains significant as AI systems lack reasoning skills, which leads them to be unable to solve geometry problems. AlphaGeometry employs the predictive power of a neural language model along with a rule-bound deduction engine to find solutions to the problems. So, it combines fast ideas with deliberate, rational decision-making. This develops over 100 million unique examples, allowing AlphaGeometry to be trained without humans.
This impressive accomplishment was recognized by Fields Medalist and IMO Gold Medalist Ngô Bảo Châu. Châu spoke, “It makes perfect sense to me now that researchers in AI are trying their hands on the IMO geometry problems first because finding solutions for them works a little bit like chess in the sense that we have a rather small number of sensible moves at every step. But I still find it stunning that they could make it work. It’s an impressive achievement.”
AlphaGeometry was also highly regarded by another former Olympiad gold medal winner, Evan Chen. Chen explained that AlphaGeometry is both verifiable and clean, which contrasts with earlier complex geometric proofs. The previous proofs were hard for human reviewers to understand, but AlphaGeometry was similar to the thought process of a human mathematician.
Over the past few years, Olympiad math problems have been a benchmark for AI. In November 2023, the Artificial Intelligence Mathematical Olympiad Prize was announced, with $5 million going to the first AI to win gold in the Olympiad. This AI discovery is one of the many breakthroughs occurring with AI. While AlphaGeometry has had positive results, other AI platforms have had less success. It was found recently that Life Corporation, a Texas Company, created a voter robocall that impersonated Biden using AI. Because of this spread of disinformation, Meta has called for cooperation in labeling AI-generated content, so users are aware of what content is real and which isn’t. While the innovative future of AI lies ahead, the world will need to be wary of potential negative consequences.
