ALPHAGO → ALPHAZERO → INEFFABLE
THE DEAL
Silver helped turn reinforcement learning from a specialist field into one of modern AI's defining ideas: let a machine act, get feedback and improve through experience.
SOURCES: Google DeepMindGoogle DeepMind
Silver led the systems that beat Go, then taught themselves chess and shogi from self-play. His thesis now: experience — not human text — is the road to superintelligence — heir to his PhD advisor Rich Sutton's 'bitter lesson' that general methods beat hand-built knowledge. He left DeepMind in 2026 to build a lab around it.
2004
Silver returned to academia in 2004 for a PhD focused on reinforcement learning with Rich Sutton, after earlier work in games and software.
SOURCES: UCLreported
2011
Silver received a Royal Society University Research Fellowship and joined UCL, where he built an academic program around reinforcement learning.
SOURCES: UCLreported
2013
After consulting for DeepMind from its early days, Silver joined the company full time in 2013 and became one of the leaders of its reinforcement-learning program.
SOURCES: UCLreported
Jan 27, 2016
Silver led the AlphaGo work combining neural networks, search and reinforcement learning, published in Nature before the system's landmark match against Lee Sedol.
SOURCES: Google Researchfirst-party
Oct 18, 2017
Silver and DeepMind introduced AlphaGo Zero, which learned Go from self-play without training on human game data and surpassed the previous AlphaGo versions.
SOURCES: Google DeepMindfirst-party
Dec 6, 2018
Silver and the DeepMind team published the full AlphaZero evaluation, showing one self-play reinforcement-learning system mastering chess, shogi and Go from scratch without human game examples.
SOURCES: Google DeepMindfirst-party
Jan 15, 2026
Silver announced Ineffable Intelligence, a new lab built around his thesis that superintelligence should learn continually from experience rather than primarily from human-generated data.
SOURCES: Ineffable Intelligencefirst-party
IDEA
Reinforcement learning trains an agent through the consequences of its actions rather than by handing it the right answer every time. Silver built his career around making that loop work at increasingly difficult scales.
SOURCES: UCL
IDEA
AlphaGo Zero started from the rules of Go and improved by playing itself, without training on human game records. AlphaZero generalized the same basic recipe across Go, chess and shogi, showing how far a system could travel when the environment itself supplied the feedback.
SOURCES: Google DeepMindGoogle DeepMind
IDEA
Silver's new company, Ineffable Intelligence, is explicitly betting that superintelligence will come from continual learning through experience rather than primarily from human-generated data. That is a thesis, not an established result, but it makes his post-DeepMind project unusually clear about what it thinks today's AI is missing.
SOURCES: Ineffable Intelligence
RECORD
Silver led the AlphaGo work that combined deep neural networks, search and reinforcement learning, culminating in a system that defeated top human Go players and became a landmark in modern AI.
SOURCES: Google Research
RECORD
The AlphaZero work showed that a single self-play learning approach could master chess, shogi and Go without human game examples, extending the central AlphaGo Zero idea beyond one game.
SOURCES: Google DeepMind
NOW
In January 2026 Silver announced Ineffable Intelligence, a lab organized around continual experiential learning and the claim that this route can produce superintelligence. After AlphaGo and AlphaZero, he is now trying to turn the reinforcement-learning worldview into a company.
SOURCES: Ineffable Intelligence
