AlphaZero
Google DeepMind AI program generalising AlphaGo to chess and shogi; developed by David Silver's team.
Last refreshed: 1 May 2026 · Appears in 1 active topic
How does AlphaZero's label-free learning connect to what Ineffable Intelligence is building?
Timeline for AlphaZero
Ineffable lands $1.1bn seed, SAIU rides minority
UK Startups and Innovation- What is AlphaZero and how is it different from AlphaGo?
- AlphaZero is a 2017 DeepMind system that mastered chess, shogi and Go from scratch using only self-play, with no human game data. AlphaGo learned from millions of human Go games first; AlphaZero needed none. It surpassed the world's best chess engine, Stockfish, within 24 hours.Source: https://en.wikipedia.org/wiki/AlphaZero
- How did AlphaZero beat Stockfish at chess?
- AlphaZero played 100 games against Stockfish 8 at fixed time controls and won 28, drew 72, and lost none. It had trained for only 24 hours using self-play from random initialisation, developing an attacking style noticeably different from conventional computer chess.Source: https://www.science.org/doi/10.1126/science.aar6404
- Why does AlphaZero matter for understanding Ineffable Intelligence?
- AlphaZero's designer, David Silver, founded Ineffable Intelligence in 2025 to extend the same idea, learning without human data, from bounded games to open-ended real-world problems. Sequoia and Lightspeed backed that vision with a $1.1 billion seed in April 2026.Source: https://techcrunch.com/2026/04/27/deepminds-david-silver-just-raised-1-1b-to-build-an-ai-that-learns-without-human-data/
Background
AlphaZero is an AI programme developed by Google DeepMind, published in December 2017 and in Science in December 2018. It generalised the AlphaGo approach into a single algorithm that taught itself chess, shogi and Go by playing against itself from random initialisation, without any opening books, endgame databases, or human game records. Within 24 hours of training it surpassed the world's strongest specialised chess engine, Stockfish 8, winning 28 games, drawing 72 and losing none from 100 games at fixed time controls. It similarly surpassed the best shogi programme (Elmo) and the AlphaGo Master version of itself.
AlphaZero was written by David Silver, Thomas Hubert and Julian Schrittwieser at DeepMind. The critical contribution was showing that a single reinforcement learning algorithm, given only the rules of a game and a self-play mechanism, could exceed any domain-specific engineered system. This was a conceptual break from decades of chess AI: Stockfish, Deep Blue and their predecessors relied on hand-crafted evaluation functions; AlphaZero learned evaluation entirely from scratch. The Science paper attracted widespread coverage and became one of the most influential AI publications of the decade, with over 10,000 citations by 2025.
AlphaZero's significance for Ineffable Intelligence, founded by David Silver in 2025, is direct. Ineffable's mission, building models that learn without human-generated training data, is an extension of the AlphaZero hypothesis from bounded game environments to open-ended real-world problems. Sequoia Capital and Lightspeed Venture Partners backed that extension with a $1.1 billion seed in April 2026, the largest in European history, treating AlphaZero's track record as a proof of concept for what an unconstrained system might achieve.