DeepMind’s AI Achieves Breakthrough in Solving Complex Mathematical Problems

Google’s AI company, DeepMind, announced that it has developed two specialized AI systems to solve complex mathematical problems involving advanced reasoning.

According to MIT Technology Review, the new AI systems, named “AlphaProof and AlphaGeometry 2” by Google DeepMind, worked together to successfully solve 4 out of 6 problems in the International Mathematical Olympiad, earning an award equivalent to a silver medal. This marks the first time an AI system has achieved such a high success rate in solving complex mathematical problems.

Pushmeet Kohli, Vice President of Research at Google DeepMind, who worked on the project, stated, “This is a significant advancement in machine learning and AI. Until now, no system has been developed that could solve such problems with this level of success and generality.”

There are several reasons why solving advanced reasoning mathematical problems is challenging for AI systems. These problems often require forming and utilizing abstractions, complex hierarchical planning, setting sub-goals, backtracking, and experimenting with new approaches. All these factors present significant challenges for AI.

The goal of creating AlphaProof, a learning-based system that trains itself to prove mathematical statements using Google DeepMind’s official programming language “Lean,” was to bridge this gap. AlphaProof improved its ability to tackle increasingly complex problems as it successfully solved more problems.

On the other hand, while AlphaProof is trained to address a wide range of mathematical topics, AlphaGeometry 2 is optimized to tackle problems involving equations with movements of objects, angles, ratios, and distances.

“BIG STEP FOR THE MATHEMATICS COMMUNITY”

Wenda Li, a lecturer in hybrid AI at the University of Edinburgh, emphasized that automating the translation of data into formal language is a significant step for the mathematics community. Li said, “If they can formulate this proof system, we can have much more confidence in the accuracy of published results and become more collaborative.”

Mathematician Tim Gowers also commented, “As a mathematician, I find it very impressive that AI answered 4 questions correctly in the Olympiad, and I see it as a significant leap from what was previously possible.