---
title: "Self-play"
kind: topic
topic_kind: concept
url: https://raskell.io/tags/self-play/
graph_id: t:self-play
broader: [reinforcement-learning, game-ai]
source: https://raskell.io/map.json
license: https://creativecommons.org/licenses/by/4.0/
---

# Self-play

A program improving by playing itself: generating its own training data, learning from the outcomes, and testing each generation against the last.

## Broader topics

- [Reinforcement learning](https://raskell.io/tags/reinforcement-learning.md): Learning from the outcomes of one's own actions: self-play, rewards, and feedback loops that turn experience into skill.
- [Game AI](https://raskell.io/tags/game-ai.md): Programs that play games: search, evaluation, learned networks, and how their strength is measured, from chess engines to Go and real-time strategy.

## Articles

- [The Engine Plays Itself](https://raskell.io/articles/the-engine-plays-itself.md) (2026-10-28): Self-play as reinforcement learning: the engine makes its own training data, a network learns to predict the outcome, and one unfitted constant cost 176 Elo.
