AlphaGo
A Go-playing AI developed by Google DeepMind, trained using reinforcement learning within a closed game environment. AlphaGo is a landmark achievement in game AI, demonstrating that reinforcement learning techniques can achieve and surpass professional human-level performance in complex board games.
Core Facts
AlphaGo combines deep learning with Monte Carlo Tree Search (MCTS), using two key neural networks: a policy network and a value network.
In March 2016, AlphaGo and Lee Sedol played a five-game match in Seoul, with AlphaGo winning 4:1.
AlphaGo defeated Go world champion Lee Sedol in 2016.
Timeline (last 90 days)
AlphaGo defeated the Go world champion, publicly proving that AI reaches superhuman levels in specific domains.
AlphaGo's success demonstrated the practicality of MCTS in a huge state space (Go's branching factor is about 250), but it also exposed its dependence on computational resources—a large number of GPUs/TPUs must be invoked during play.
Go and chess are perfect-information games, with all piece positions public and in theory a definite search space, and systems like AlphaGo can rely on Monte Carlo tree search plus a value network to approach the optimal solution.
AlphaGo's Move 37 appeared in the 37th move of the second game of DeepMind's 2016 match against Lee Sedol, demonstrating that AI can employ strategies humans would never conceive of.
The RLVR paradigm is essentially a linguistic transfer of the AlphaGo approach, because both Go outcomes and math problem answers can be adjudicated by machines.
Complexity itself does not breed consciousness; chess engines, AlphaGo, Google's search system, and the Linux kernel are all highly complex but non-conscious computational systems.
The Go match between AlphaGo and Lee Sedol in South Korea ignited a nationwide AI enthusiasm across the country.
DeepMind developed AlphaGo and AlphaStar (StarCraft), while OpenAI developed a Dota 2 team battle AI
AlphaGo's policy network and value network were trained through massive self-play, conducting search and evaluation within a closed system with clear rules and objectives.
The reason the reinforcement learning paradigm underlying AlphaGo was able to produce superhuman strategies lies in the clarity and density of the reward signal, as each game provides unambiguous win/loss feedback.
28 more timeline events
All Facts (20)
AlphaGo combines deep learning with Monte Carlo Tree Search (MCTS), using two key neural networks: a policy network and a value network.
90%VerifiedIn March 2016, AlphaGo and Lee Sedol played a five-game match in Seoul, with AlphaGo winning 4:1.
90%VerifiedAlphaGo defeated Go world champion Lee Sedol in 2016.
90%VerifiedDeepMind's AlphaGo learned Go strategies by playing millions of games against itself.
75%VerifiedAfter the integration of deep reinforcement learning, AlphaGo defeating the human Go champion and OpenAI Five became its landmark applications.
75%VerifiedAlphaGo's defeat of the human Go champion is a landmark application of reinforcement learning.
70%VerifiedMonte Carlo Tree Search (MCTS) is one of the core algorithms for Go AI to search game positions, and was the mainstream technology in Go programs before AlphaGo.
65%VerifiedDeepMind is an artificial intelligence company under Google, known for developing AlphaGo, which defeated the world Go champion.
65%VerifiedReinforcement learning is a branch of machine learning in which a model learns by continuously interacting with an environment and adjusting its strategy based on reward signals; both AlphaGo and ChatGPT's RLHF training rely on this framework.
65%VerifiedAlphaGo Zero completely abandoned human data, learned purely through self-play, and ultimately reached a higher level than AlphaGo Lee.
65%VerifiedDeepMind is developing an agent framework that integrates different specialized models such as AlphaFold and AlphaGenome.
85%VerifiedAlphaGo defeated world champion Lee Sedol in 2016, with its core breakthrough being the combination of deep convolutional neural networks with Monte Carlo tree search.
75%Unverified在AlphaGo之前，最强的围棋AI仅能达到业余中等水平，与职业棋手之间存在巨大鸿沟
90%Unverified过去打造AI Agent主要依赖强化学习（RL）算法，需要为每一个任务训练一个专门的模型
90%Unverified全球围棋人口在AlphaGo之后显著增长
75%UnverifiedDeepMind's AlphaGo overturned the research direction of Go in 2016 using deep neural networks combined with Monte Carlo Tree Search.
70%UnverifiedThe AlphaGo series improved its playing strength through self-play.
60%UnverifiedMCTS is the planning core of systems such as AlphaGo and MuZero, and its core loop consists of four steps: selection, expansion, simulation, and backpropagation.
60%UnverifiedGoogle DeepMind merged the two major research teams of the original DeepMind (AlphaFold, AlphaGo) and Google Brain (birthplace of the Transformer architecture)
60%UnverifiedGoogle possesses proprietary TPU chip clusters, DeepMind provides fundamental research capabilities, and it has billion-daily-active-user products such as Search, Gmail, and Google Docs
60%