AlphaGeometry: An Olympiad-level AI system for geometry

Researchers from Google DeepMind, New York University (NYU), and the Computer Science Department of NYU have developed a new algorithm called AlphaGeometry, which utilizes classical geometry rules and angles to solve mathematical problems. The new approach is capable of solving six problems, out of the four problems typically focused on in traditional mathematical reasoning with AI systems such as the Large Languae Model (LLM) – the most widely used AI model for symbolic reasoning problems in mathematics. Their approach differs from previous methods which focus solely on verifying and checking whether an LLM-based system can perform mathematical reasoning, but still rely primarily on visual inspection of images or diagrams. AlphaGeometry utilizes classical geometry rules, such as angles and angles between parallel lines, to reason about problems in mathematics, thereby expanding the frontiers of human knowledge. This research is a collaboration between Google DeepMind’s AI research group and NYU’s Computer Science Department, with support from Google Research. The authors acknowledge that their information will be used in accordance with Google’s Privacy Policy.