Home/AI/AlphaGo Strategy Breakdown/Introduction

🎮 AlphaGo Strategy Breakdown

The AI that mastered Go through deep learning and tree search

Your Progress

0 / 5 completed
←
Previous Module
Actor-Critic Architectures

Introduction to AlphaGo

🏆 The Historic Achievement

In March 2016, AlphaGo defeated Lee Sedol, one of the world's top Go champions, 4-1. This breakthrough demonstrated that AI could master games previously thought to require human intuition and creativity. Go has 10^170 possible positions — more than atoms in the universe.

💡
Key Insight

AlphaGo combined deep neural networks for intuition with Monte Carlo Tree Search for strategic planning, creating a system stronger than any previous Go AI.

🧠
Policy Network

Predicts expert moves, selecting promising actions to explore

📊
Value Network

Evaluates board positions, estimating winning probability

🌳
Tree Search

MCTS explores move sequences, balancing exploration vs exploitation

🔄 AlphaGo Evolution

1
AlphaGo Fan (2015)

Defeated European champion Fan Hui 5-0. Trained on human games.

2
AlphaGo Lee (2016)

Beat Lee Sedol 4-1. Added value network and improved training.

3
AlphaGo Master (2017)

Won 60 online games against top pros. Enhanced self-play.

4
AlphaGo Zero (2017)

Learned from pure self-play, no human data. Defeated all versions.

✅ Breakthroughs

  • • Combined deep learning + tree search
  • • Self-play reinforcement learning
  • • Superhuman Go performance
  • • Discovered novel strategies

🎯 Impact

  • • Revolutionized game AI research
  • • Advanced neural network training
  • • Inspired AlphaZero, MuZero
  • • Applied to protein folding (AlphaFold)