Google New AI Surpasses Olympians at Solving Geometry Problems

Google DeepMind advanced AI system AlphaGeometry2 (AG2) is capable of solving difficult geometry problems more effectively than the International Mathematical Olympiad (IMO) gold medallists. With an average success percentage of 84%, AG2 outperformed its top human opponents, who had an average success rate of 81.8%.

The Future of AI in Mathematics

DeepMind wants to improve AG2 even further by increasing its speed, accuracy, and capacity to tackle increasingly challenging issues. Error-free, totally automated geometry problem solving is the aim.

Although AG2 is primarily concerned with geometry, its technology may have applications in engineering, robotics, medicine, and genetic research.

AG2 is still a specialised tool as of right now. However, its success suggests that AI may eventually be able to tackle difficult human problems, one equation at a time.

Limitations and the Process of Fixing Problems

AG2 solves geometry issues using a combination of symbolic engines and neural language models.

The symbolic engine tests the recommendations made by the language model, which also interprets the issues and makes possible construction suggestions.

Thanks to a faster symbolic engine and an improved data set, AG2 is both faster and more accurate than its predecessor. It also looks for geometric proofs using a specific method.

Potential Applications of the Technology

Although AG2 is currently only a specialised tool for geometry, its technology may find application in other fields such as genetic research, engineering, robotics, and pharmaceuticals.

This suggests that AG2’s achievement signifies AI’s increasing capacity to address difficult human problems.

But for the time being, it is only a specialised tool that focusses on solving challenging geometry issues.

AlphaGeometry: More Intelligent, Fast, and Advanced

The California-based company said that in order to improve the system’s capabilities, it expanded the original AlphaGeometry language to address more challenging problems that involved object movements and problems that involved linear equations of angles, ratios, and distances.

“This, together with other additions, has markedly improved the coverage rate of the AlphaGeometry language on IMO 2000-2024 geometry problems from 66 per cent to 88 per cent.”

Google also enhanced AlphaGeometry2’s search function to boost language modelling using its cutting-edge Gemini AI tool.