Reinforcement Learning: The AI Technique That Taught Machines to Master Chess, Go, and the Real World
When DeepMind AlphaGo defeated Lee Sedol: the world best Go player: in March 2016, it was a watershed moment in AI history. Go had long been considered the game AI would struggle with most, given the astronomical number of possible positions and the extent to which expert play depends on intuition built over decades. Yet AlphaGo won four games out of five, making moves that stunned Go experts: moves that looked bizarre at first glance and turned out to be profound.
The technique behind AlphaGo: and behind a long series of subsequent AI breakthroughs: is reinforcement learning. And while its game-playing achievements capture headlines, reinforcement learning is increasingly being applied to real-world problems with enormous practical significance.
Reinforcement learning is a type of machine learning where an AI agent learns to make decisions by interacting with an environment and receiving feedback in the form of rewards or penalties. Unlike supervised learning where the model learns from labeled examples of correct answers, reinforcement learning learns from the consequences of its own actions.
The basic setup involves three elements:
An agent: the AI system making decisions
An environment: the world the agent interacts with: a game, a simulation, a robot body, or a real-world system
Please enable JavaScript to read the full article.