Google’s advanced AI system outperforms top human mathematicians
Google has unveiled a revolutionary artificial intelligence (AI) system capable of outperforming human gold medalists at the International Mathematical Olympiad (IMO). The second-generation AI system, AlphaGeometry2, has demonstrated an 84% success rate in solving complex geometry problems, surpassing the average 81.8% accuracy of gold medal-winning participants.
A leap in AI-driven mathematical reasoning
Developed by DeepMind, AlphaGeometry initially performed at the level of silver medalists when introduced in January last year. However, Google now claims that its latest iteration has exceeded the capabilities of top human competitors, marking a significant milestone in AI-driven mathematical problem-solving.
To enhance the system’s capabilities, Google expanded AlphaGeometry’s core language, enabling it to tackle more intricate problems involving object movements, 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% to 88%,” Google stated.
Integration of Gemini AI for enhanced problem-solving
Leveraging its advanced Gemini AI tool, Google also refined AlphaGeometry2’s search process, making it more efficient in language modeling and mathematical reasoning.
“The team also introduced the ability for the AI to reason by moving geometric objects around a plane—allowing it to move a point along a line to change the height of a triangle, for instance—and for it to solve linear equations,” Google explained.
Challenges and future developments
Despite its remarkable performance, Google acknowledges that AlphaGeometry2 still has limitations. The AI currently struggles with problems involving a variable number of points, non-linear equations, and inequalities.
“Our domain language does not allow talking about a variable number of points, non-linear equations, and problems involving inequalities, which must be addressed to fully solve geometry,” the company noted.
DeepMind researchers aim to further refine the system, with the long-term goal of achieving full automation of geometry problem-solving without errors. Future developments will focus on improving inference speed, reliability, and the overall scope of mathematical reasoning capabilities in AI.
As AI continues to bridge the gap between human and machine intelligence, Google’s advancements in mathematical problem-solving signal a transformative era for artificial intelligence in academia and beyond.
