Assertion Supported AI assessment confidence: 98% certainty 3/5 debate potential 2/5

Marcus: DeepMind's deep RL required millions of data points per game

Gary Marcus · Can A.I. Become More Human? // Gary Marcus, Geometric Intelligence (Hosted by FirstMark Capital) · Jan 25, 2016 · at 14:02

Gary Marcus, cognitive scientist and CEO of Geometric Intelligence, discusses the high data requirements of DeepMind's reinforcement learning techniques.

0:00 / 0:10exact quote · 10.6s

▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark

“This is what DeepMind used in their systems, and they used a version called deep reinforcement learning. And what they did is they collected billions or, you know, probably millions of data points for each game.”

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

DeepMind used deep reinforcement learning that required millions of data points per game to train its game-playing AI systems.

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

DeepMind's landmark published research, including Mnih et al. (Nature 2015), confirms that their Deep Q-Network (DQN) AI systems utilized deep reinforcement learning and required 50 million agent steps (corresponding to 200 million raw frames) per Atari game during training.

Mnih et al. (Nature, Feb 2015) 'Human-level control through deep reinforcement learning' established that DQN agents were trained on 50 million frames/steps per Atari game.
Mnih et al. (NIPS 2013) 'Playing Atari with Deep Reinforcement Learning' demonstrated training on 10 million frames per game using deep reinforcement learning.

Checked against

[1] arxiv.org [2] researchgate.net [3] stanford.edu [4] toronto.edu [5] stanford.edu [6] github.io [7] medium.com [8] medium.com

DeepMind Human level control through deep reinforcement learning Nature 2015 50 million frames Mnih et al Nature 2015 Deep reinforcement learning millions frames data points Atari "50 million frames" OR "10 million frames" OR "38 million" OR "50 million" DQN "Mnih et al" 2015 "Mnih et al" 2015 "50 million frames" OR "50,000,000" OR "38 million" OR "10,000,000" OR "50 million steps"

assessed as of 2026-07-30

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

deep reinforcement learning DeepMind

⚑ Spot an error? Report it

More from Gary Marcus

Opinion

Marcus challenges Andrew Ng: AI cannot automate every one-second mental task

“I don't think it's true at all, so there are some things that people can do in a second that AI can automate, and there are other things that it can't”

Gary Marcus Sep 17, 2019 ▶ 1:18 Rebooting AI // Gary Marcus, Robust AI (FirstMark's Data Driven NYC)

Opinion

Marcus: Drivers should not trust Tesla Autopilot because AI remains unready

“So you should not be trusting a Tesla. Elon might get on 60 minutes and take his hands off the wheel, but you should not be doing the same because the technology is not really there yet.”

Gary Marcus Sep 17, 2019 ▶ 2:25 Rebooting AI // Gary Marcus, Robust AI (FirstMark's Data Driven NYC)

Opinion

Marcus: Current AI techniques are too brittle to earn public trust

“Most of why I don't think we should trust AI right now is that the current techniques that people are using are simply too brittle to earn the trust that we're giving them.”

Gary Marcus Sep 17, 2019 ▶ 3:42 Rebooting AI // Gary Marcus, Robust AI (FirstMark's Data Driven NYC)

Assertion Not checkable as stated

Marcus: Deep learning models are giant correlation machines without causal understanding

“These systems are just giant correlation machines. They have no understanding of causation.”

Gary Marcus Sep 17, 2019 ▶ 12:56 Rebooting AI // Gary Marcus, Robust AI (FirstMark's Data Driven NYC)

Insight

Marcus: Deep learning cannot perform common sense, planning, analogy, or reasoning

“Perception is what deep learning does, and actually it only does a small part of perception. There's lots of parts of perception where we use our knowledge about the world that it doesn't capture. And then there are all these other things like common sense and…”

Gary Marcus Sep 17, 2019 ▶ 13:33 Rebooting AI // Gary Marcus, Robust AI (FirstMark's Data Driven NYC)

Insight

Marcus: Deep learning is a better ladder but cannot reach the moon

“Deep learning is a better ladder for sure, but a better ladder doesn't necessarily get you to the moon.”

Gary Marcus Sep 17, 2019 ▶ 19:27 Rebooting AI // Gary Marcus, Robust AI (FirstMark's Data Driven NYC)

Marcus: DeepMind's deep RL required millions of data points per game

More from Gary Marcus

Turn any episode into a week of clips.