DeepMind's AI Nears Human-Level Geometry Skills

Google's DeepMind AI, AlphaGeometry, nearly matches top students in solving complex math Olympiad problems.

Can AI solve mathematical problems?

Google DeepMind's latest advancement in artificial intelligence, AlphaGeometry, has significantly narrowed the gap between AI and human capabilities in solving complex geometry problems. In a recent Nature publication, the system's near-gold-medal performance at the International Mathematical Olympiad is highlighted.

Overview of the neuro-symbolic AlphaGeometry
Overview of the neuro-symbolic AlphaGeometry (Source: Nature)

AlphaGeometry: AI's Leap In Problem-Solving

AlphaGeometry, Google's AI endeavor, impressively solved 25 out of 30 questions from the high school level International Mathematical Olympiad. This performance, verging on the standards of top human contestants, underscores AI's advancing proficiency in complex mathematics. DeepMind researcher Quoc V Le views this as a pivotal step towards developing an General Artificial Intelligence (AGI) capable of matching or surpassing human intelligence.

AI 101: General AI
General AI, poised to transform many industries and research, faces complex technical and ethical challenges in its development.

Bridging Language Learning And Deductive Reasoning

The AI system employs a neuro-symbolic approach, combining language learning with deductive reasoning. This method mirrors Daniel Kahneman's concept of "Thinking, Fast and Slow," where quick pattern recognition complements logical thinking. DeepMind's Trieu H Trinh asserts that this hybrid technique uniquely equips AlphaGeometry to tackle geometric problems, balancing everyday spatial understanding with intricate mathematical theory.

Benchmarking Against Human Competitors

To train AlphaGeometry, researchers created a dataset of 100 million synthetic geometry examples. The system's score of 25 nearly matched the 25.9 average of human winners in mathematical Olympiads from 2000 to 2022 and far surpassed the previous best automated system's score of 10. However, the AI found certain problems, like a 1979 Olympiad conundrum by Vietnamese mathematician Lê Bá Khánh Trình, particularly challenging.

The Pursuit Of AI Excellence In Mathematics

DeepMind's broader objective extends beyond matching human abilities; the aim is to enable AI to tackle mathematical problems yet unsolved by humans. While Mikhail Burtsev of the London Institute for Mathematical Sciences acknowledges DeepMind's significant progress, he emphasizes the ultimate challenge: Developing AI capable of discovering new mathematics for unsolved problems. The vision of AI achieving a landmark victory over human rivals in mathematics, akin to Deep Blue's chess triumph over Garry Kasparov in 1997, remains an aspiration as DeepMind continues to explore the realms of advanced mathematics.

IBM’s Deep Blue: A Chess Match That Transformed AI
Deep Blue, IBM’s chess-playing AI, not only won a game but transformed the field of artificial intelligence.

Subscribe to our newsletter and follow us on X/Twitter.

Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to REX Wire.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.