Google DeepMind's AlphaGeometry AI algorithm solves complex geometry problems at the level of a maths Olympiad champion

DeepMind

An artificial intelligence system called AlphaGeometry, developed by Google DeepMind, a British company, has demonstrated an ability to solve geometry problems comparable to the level of gold medallists at the International Mathematical Olympiad (IMO), a prestigious competition for high school students.

Here's What We Know

According to DeepMind researchers, AlphaGeometry was able to solve 25 out of 30 IMO test problems in the allotted 4.5 hours. This result is comparable to the average performance of human Olympiad winners.

Comparison of the productivity of AlphaGeometry, the previous model and the IMO medallists

The system combines two different approaches:

• a neural language model that generates intuitive ideas;
• a symbolic deduction mechanism that verifies ideas using formal logic and rules.

The language model is based on the same technology as Google's search engine and natural language understanding systems. The deduction mechanism is inspired by a method developed by Chinese mathematician Wen-Tsün Wu in 1978.

Experts say the findings demonstrate progress in the development of machine intelligence capable of logical reasoning and discovering new mathematical insights. However, they recognise the limitations of AlphaGeometry.

Nevertheless, in the future, such systems could find widespread use to accelerate scientific research in maths, science and other fields, DeepMind believes. For this purpose, they have posted the AlphaGeometry source code on GitHub.

Source: Nature