Chapter
How Neural Networks Learn to Play Games
45:34 - 53:51 (08:16)
The narrow sense of learning through self-play by systems can lead them to achieve superhuman level performance in games like chess and go without human supervision. There's a lot of surprising and beautiful things that neural networks in the deep learning revolution can reveal.
Clips
The capabilities of neural networks in deep learning have been surprising and beautiful, particularly with regards to achievements in chess play without human supervision.
45:34 - 48:07 (02:32)
Summary
The capabilities of neural networks in deep learning have been surprising and beautiful, particularly with regards to achievements in chess play without human supervision. Despite this, there is still work to be done in solving problems in computer vision and vision for robots.
ChapterHow Neural Networks Learn to Play Games
Episode#217 – Rodney Brooks: Robotics
PodcastLex Fridman Podcast
Learning is a complex term that has different meanings depending on its context.
48:07 - 49:39 (01:32)
Summary
Learning is a complex term that has different meanings depending on its context. While machine learning systems can become less dumb at the task they are assigned to, they often fail when the conditions change even slightly.
ChapterHow Neural Networks Learn to Play Games
Episode#217 – Rodney Brooks: Robotics
PodcastLex Fridman Podcast
The game of Go can be learned to beat the best people in the world without human supervision through self-play by the system playing itself.
49:40 - 52:13 (02:32)
Summary
The game of Go can be learned to beat the best people in the world without human supervision through self-play by the system playing itself. This can achieve superhuman level performance through learning alone, although a human player would still be able to play effectively.
ChapterHow Neural Networks Learn to Play Games
Episode#217 – Rodney Brooks: Robotics
PodcastLex Fridman Podcast
Summary
Arthur Samuel is a name you should know. He was an AI pioneer back in the 1950s, and he invented the concept of a machine-learning algorithm decades before anyone used the term. Samuel's major accomplishments include the first checkers program that could defeat a world champion, a seminal machine-learning paper, and pioneering work in machine vision, language understanding and more.