10 years of AlphaGo: The turning point for AI | Thore Graepel & Pushmeet Kohli

2026-03-10 · Source: Google DeepMind · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, extended

Summary

The Google DeepMind podcast reflects on the 2016 AlphaGo match against 18-time Go world champion Lee Sedol, marking a pivotal moment in AI development. AlphaGo, a neural network-based AI system utilizing reinforcement learning, defeated Sedol 4-1 in a highly complex game previously thought impossible for machines to master. This victory, particularly AlphaGo's "Move 37," demonstrated AI's capacity to generate counter-intuitive, yet optimal, strategies that surpassed human understanding. The discussion highlights how AlphaGo's success, achieved through a combination of "fast thinking" (intuition via deep learning policy networks) and "slow thinking" (explicit planning via search algorithms), laid the groundwork for subsequent AI breakthroughs like AlphaFold for protein folding and AlphaTensor for optimizing matrix multiplication. The evolution to AlphaZero, which learned Go from scratch without human data and discovered superior strategies, further underscored AI's potential to transcend human knowledge.

Key takeaway

For research scientists developing advanced AI, the AlphaGo legacy underscores the importance of designing systems that can both learn from and transcend human knowledge. Focus on creating AI agents that combine intuitive pattern recognition with rigorous search algorithms, and integrate robust verification mechanisms to distinguish genuine breakthroughs from errors. Your work can lead to "Move 37" moments in scientific discovery, expanding human understanding in fields like biology, chemistry, and mathematics.

Key insights

AlphaGo's victory in Go demonstrated AI's ability to combine intuition and calculation, surpassing human expertise and fostering new scientific discoveries.

Principles

• AI can learn beyond human knowledge.
• Combine intuition and calculation for complex problem-solving.
• Reinforcement learning excels in verifiable domains.

Method

AlphaGo combined a deep learning-based policy network for "fast thinking" (intuition) with a search algorithm for "slow thinking" (explicit planning) to navigate vast combinatorial spaces in Go.

In practice

• Use AI to discover novel algorithms.
• Apply AI to optimize complex scheduling problems.
• Employ verifiers to filter AI "hallucinations."

Topics

• AlphaGo
• Reinforcement Learning
• Algorithmic Discovery
• AI for Science
• Deep Learning

Best for: Research Scientist, AI Researcher, AI Scientist, Machine Learning Engineer

Related on AIssential

• 10 Years of AlphaGo: The Turning Point for AI | Thore Graepel & Pushmeet Kohli — AlphaGo's victory marked AI's transition beyond human-level intelligence, combining intuition and calculation to solve complex problems.
• From games to biology and beyond: 10 years of AlphaGo’s impact — AlphaGo's victory in Go catalyzed modern AI, demonstrating systems can surpass human intuition and accelerate scientific discovery.
• Google Drops Veo 3.1 Lite — AI systems can surpass human intuition and knowledge, driving scientific discovery in complex domains.
• AI and the long game — AI's evolution, exemplified by AlphaGo's victory, shifts focus from human-machine competition to understanding intelligence through continuous scientific exploration.
• Move 37 — AlphaGo's Move 37 redefined AI's reasoning capabilities, surprising human experts.
• AI is rewiring how the world’s best Go players think — AI has transformed professional Go, shifting training paradigms and democratizing access while raising questions about creativity and human understanding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Google DeepMind.