Assertion Supported AI assessment confidence: 98% certainty 3/5 debate potential 1/5

Domingos: AlphaGo Trained on 30 Million Human Moves Before Self-Play

Pedro Domingos · #13 Pedro Domingos: The Rise of The Machines · Aug 30, 2016 · at 47:47

Computer scientist Pedro Domingos explains the architectural differences and training methodology behind Google DeepMind's AlphaGo to host Shane Parrish.

0:00 / 0:20exact quote · 20.9s

720p mp4 · rendered on demand · StarZero watermark

“So AlphaGo, the first thing that AlphaGo did was it learned from all the existing, the entire existing database of Go matches played by human masters. That was the first thing it did was learn from those, right? Thirty million moves or something like that is the figure I think I heard. And then the next thing that it did was actually it learned by playing against itself.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

AlphaGo was trained first by learning from an existing database of approximately 30 million expert human Go moves, followed by reinforcement learning through self-play.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

The claim accurately describes the training pipeline of AlphaGo as documented in DeepMind's January 28, 2016 Nature paper ('Mastering the game of Go with deep neural networks and tree search'). The paper specifies that AlphaGo's policy network was initially trained via supervised learning on 29.4 million board positions (~30 million moves) from 160,000 expert human games played by 6-dan to 9-dan players, followed by reinforcement learning through self-play games against previous iterations of itself.

DeepMind's January 2016 Nature paper confirms AlphaGo was trained via supervised learning on 29.4 million positions from expert human Go matches on the KGS Go Server.
The paper confirms that after the supervised learning phase, AlphaGo's policy network was refined through reinforcement learning via self-play.

Checked against

[1] nih.gov [2] baulab.info [3] researchgate.net [4] semanticscholar.org [5] airev.us [6] medium.com [7] scispace.com

Mastering the game of Go with deep neural networks and tree search 30 million positions moves supervised learning self play AlphaGo 30 million positions KGS Go Server Nature paper Silver et al 2016

assessed as of 2026-09-03

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

Mentioned

AlphaGo Go Google DeepMind

⚑ Spot an error? Report it

More from Pedro Domingos

Opinion

Domingos: Doctors Gatekeep Medical AI to Prevent Automating Their Own Jobs

“In the particular case of medicine, it's not used more already because, of course, the doctors are also the gatekeepers Of the system, and they're not very interested in replacing themselves or their job that they like best by machines.”

Pedro Domingos Aug 30, 2016 ▶ 25:20 #13 Pedro Domingos: The Rise of The Machines

Prediction Not checkable as stated

Domingos: AI Will Plateau Rather Than Experiencing Infinite Runaway Intelligence Growth

“And I think they will be in the case of AI, but I don't think we're going to see this, you know, infinite growth that, you know, goes completely beyond you know, what humans can imagine.”

Pedro Domingos Aug 30, 2016 ▶ 39:35 #13 Pedro Domingos: The Rise of The Machines

Prediction Not checkable as stated

Domingos: Computers Will Soon Discover and Store Most Knowledge on Earth

“So in the not too distant future, the vast majority of the knowledge on earth Will be discovered and will be stored in computers.”

Pedro Domingos Aug 30, 2016 ▶ 3:34 #13 Pedro Domingos: The Rise of The Machines

Assertion Partly supported

Domingos: A Venture Capital Fund Appointed an Algorithm as a Voting Director

“For example, there's this venture fund recently that announced that one of their directors is now going to be an algorithm. There's seven directors on the board, and one of them's an algorithm. So their algorithm doesn't decide anything all by itself, but it d…”

Pedro Domingos Aug 30, 2016 ▶ 4:29 #13 Pedro Domingos: The Rise of The Machines

Prediction Not checkable as stated

Domingos: Researchers Can Develop a Single Universal Master Algorithm for ML

“But what I and others believe is that we can Develop a true master algorithm, meaning an algorithm that is able to solve all the different kinds of learning problems that these different algorithms can.”

Pedro Domingos Aug 30, 2016 ▶ 7:58 #13 Pedro Domingos: The Rise of The Machines

Opinion

Domingos: Achieving a Universal Master Algorithm Requires Entirely New AI Paradigms

“And my gut feeling is that actually it's more the latter. I do believe that we have made a lot of progress, but I think we are still missing some important ideas.”

Pedro Domingos Aug 30, 2016 ▶ 21:47 #13 Pedro Domingos: The Rise of The Machines

Domingos: AlphaGo Trained on 30 Million Human Moves Before Self-Play

More from Pedro Domingos

Turn any episode into a week of clips.