Google's new AI outperforms Olympiad gold medalists in geometry problems
What's the story
With its latest creation, AlphaGeometry2 (AG2), Google DeepMind has taken a giant leap in the field of artificial intelligence (AI).
The advanced AI system has shown exceptional problem-solving skills, beating International Mathematical Olympiad (IMO) gold medalists.
The results were published on February 7 in the arXiv preprint database.
AG2 solved 84% of IMO geometry problems, exceeding the 81.8% achievement rate of gold medalists.
AI capabilities
AG2's unique approach to complex geometry
AG2 is not merely a pattern recognition tool, but a system that combines logic and creativity. This innovative approach makes it stand out from other AI systems.
Unlike Microsoft's rStar-Math, launched last month, and using small language models (SMLs) for broad mathematical reasoning, AG2 specializes in complex geometry.
The updated version of AG2 has demonstrated a 30% improvement over its predecessor, first introduced in January 2024.
Challenges
Problem-solving process and limitations
AG2 employs a combination of neural language models and symbolic engines to solve geometry problems.
The language model interprets the problems and suggests potential constructions, while the symbolic engine tests the suggestions.
However, despite its success, experts warn against seeing AG2 as a step toward Artificial General Intelligence (AGI).
"AlphaGeometry2 shows intelligence, but it's not human-like," said John Bates, CEO of SER Group.
Evolution
DeepMind's future plans for AG2
Despite its impressive performance, AG2 has its limitations. It struggles with 3D geometry, non-linear equations, and problems involving variable/infinite points. Plus, it can't explain its solutions in human language.
However, DeepMind plans to further refine AG2 to improve speed and accuracy while enhancing its problem-solving capabilities.
The ultimate goal is full automation of geometry problem-solving without errors.
Impact
Potential applications of the technology
While AG2 is just a specialized tool for geometry at the moment, its technology could prove useful in other areas like engineering, robotics, pharmaceuticals, and genomic research.
This indicates that the success of AG2 marks the growing potential of AI to solve complex human challenges.
However, for now, it is just a specialized instrument with its main focus on mastering complex geometry problems.