# 10 years of AlphaGo: The turning point for AI | Thore Graepel & Pushmeet Kohli Source: https://www.youtube.com/watch?v=qoinGjj60Fo Recap page: https://rapidrecap.app/video/qoinGjj60Fo Generated: 2026-03-10T17:33:53.692+00:00 --- ## Quick Overview Ten years after AlphaGo defeated Lee Sedol 4-1 in March 2016, the participants reflected on how that match catalyzed the modern AI revolution, leading to advancements in reinforcement learning and large language models, while also highlighting the initial shock and eventual acceptance of AI creativity in complex domains like Go. **Key Points:** - The AlphaGo vs. Lee Sedol match took place in Seoul, South Korea, in March 2016, where AlphaGo ultimately won 4-1 against the 18-time world champion. - Move 37 in Game 2 was a key moment, deemed a 'surprising move' and 'creative' by commentators, which initially caused confusion but proved to be a pivotal, non-human-like move. - Thore Graepel, a Distinguished Research Scientist at Google DeepMind, noted that the core technique used was reinforcement learning, which allowed AlphaGo to learn from self-play, often discovering moves that human experts initially considered mistakes or hallucinations. - Pushmeet Kohli, who leads DeepMind's science work, mentioned that the AI's ability to solve complex problems like protein folding was underpinned by similar techniques pioneered in the AlphaGo project. - The match generated intense public interest, leading to a massive surge in AI research and public awareness regarding the potential of systems that operate beyond direct human intuition or prior knowledge. - The participants emphasized that AlphaGo's success wasn't just about winning, but about discovering new, effective strategies and opening up new avenues for human understanding in complex domains. ![Screenshot at 00:04: The match between Lee Sedol \(South Korea\) and AlphaGo \(Google DeepMind\) is displayed on a large screen in Seoul in March 2016, marking a major event in AI history.](https://ss.rapidrecap.app/screens/qoinGjj60Fo/00-00-04.jpg) **Context:** This video is an interview/discussion on the Google DeepMind podcast, hosted by Professor Hannah Fry, featuring two key figures from the AlphaGo project: Thore Graepel, Distinguished Research Scientist, and Pushmeet Kohli, VP of Science Research. They discuss the impact and legacy of the 2016 match where AlphaGo defeated world champion Lee Sedol in the game of Go, reflecting on the AI's unique strategies and how the project influenced subsequent AI development, particularly in reinforcement learning and large language models. ## Detailed Analysis The discussion centers on the AlphaGo vs. Lee Sedol match from March 2016, which served as a turning point for modern AI, particularly deep reinforcement learning. Professor Hannah Fry interviews Thore Graepel and Pushmeet Kohli about the match's impact and the AI techniques that drove AlphaGo's success. Graepel describes the initial shock among professional players due to AlphaGo's seemingly counter-intuitive moves, such as Move 37 in Game 2, which experts initially dismissed but later realized were strategically superior. Kohli emphasizes that the underlying methodology—self-play reinforcement learning coupled with policy and value networks—allowed the system to discover novel strategies beyond human intuition, leading to breakthroughs in other areas like protein folding. The participants highlight that the challenge was not just solving the game but finding novel, verifiable solutions that expanded human understanding of Go. They also note the immense public and scientific interest generated, which spurred further research into explainable AI, as the system's superior moves often lacked immediate intuitive justification for human observers. ### AlphaGo Match Context - The Google DeepMind Challenge Match against Lee Sedol occurred in Seoul, March 2016 - AlphaGo won the series 4-1, with Move 37 in Game 2 being a standout move that surprised commentators. ### Reinforcement Learning & Novelty - AlphaGo’s strength came from reinforcement learning via self-play, allowing it to discover moves humans considered mistakes or hallucinations, leading to novel strategies. ### Broader AI Impact - Techniques developed for AlphaGo, like those involving policy and value networks, became foundational for subsequent AI advancements, including large language models and protein folding prediction. ### Explainability Challenge - A key issue discussed is the lack of human interpretability for AlphaGo's best moves, contrasting its calculated success with human intuition. ### Post-Match Legacy - The match spurred significant interest in AI research globally, demonstrating the power of deep learning approaches to tackle complex problems with massive search spaces. ![Screenshot at 00:04: A billboard in Seoul advertises the AlphaGo vs. Lee Sedol match scheduled for March 9th.](https://ss.rapidrecap.app/screens/qoinGjj60Fo/00-00-04.jpg) ![Screenshot at 00:25: A close-up view of a monitor displaying the AlphaGo game interface, showing the current board state and variations/log data.](https://ss.rapidrecap.app/screens/qoinGjj60Fo/00-00-25.jpg) ![Screenshot at 00:44: David Silver, a lead researcher, discusses the surprising nature of AlphaGo's moves with another team member while observing the game.](https://ss.rapidrecap.app/screens/qoinGjj60Fo/00-00-44.jpg) ![Screenshot at 01:02: A Korean newspaper headline reflects the global reaction to AlphaGo's victory over Lee Sedol.](https://ss.rapidrecap.app/screens/qoinGjj60Fo/00-01-02.jpg)