DeepMind said its AlphaGeometry AI model was able to solve complex geometry problems at a comparable level to an Olympiad gold-medalist, showcasing the ability of AI models to use reasoning skills when solving problems.
Google-owned DeepMind has shared details about its latest AI model that is pushing the boundaries of machine-based reasoning in mathematics.
The company said its AlphaGeometry AI system can solve complex geometry problems at a level approaching a human gold-medalist of the International Mathematical Olympiad, a “modern-day arena” for the world’s brightest high-school mathematicians.
In a study published in the scientific journal Nature, DeepMind put its AI model through a benchmarking test where it attempted to solve 30 Olympiad geometry problems. The company claims that AlphaGeometry managed to solve 25 within the standard Olympian time limit.
“For comparison, the previous state-of-the-art system solved 10 of these geometry problems, and the average human gold medalist solved 25.9 problems,” DeepMind said in a blogpost.
DeepMind said AI systems usually struggle with complex problems in geometry and mathematics due to a “lack of reasoning skills and training data”. The company claims AlphaGeometry combines a neural language model with a “rule-bound deduction engine” to find solutions to complex geometry problems.
“By developing a method to generate a vast pool of synthetic training data – 100m unique examples – we can train AlphaGeometry without any human demonstrations, sidestepping the data bottleneck,” DeepMind said.
The company said the score its AI model achieved demonstrates the growing ability for AI to reason logically and to “discover and verify new knowledge”.
“Solving Olympiad-level geometry problems is an important milestone in developing deep mathematical reasoning on the path towards more advanced and general AI systems,” DeepMind said.
“We are open-sourcing the AlphaGeometry code and model, and hope that together with other tools and approaches in synthetic data generation and training, it helps open up new possibilities across mathematics, science, and AI.”
Last month, DeepMind claimed one of its AI models – FunSearch – found a new answer for an unsolved mathematical problem. The company said this AI model has an automated “evaluator” to prevent hallucinations, allowing the model to find the best answers for advanced problems.
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.
