AlphaGo Zero Beats the Original AlphaGo 100-0

by Developer.com Staff

The new artificial intelligence program relies on reinforcement learning.

The Google DeepMind team has created a new artificial intelligence (AI) program that is even better than the original AlphaGo program that beat human experts at the board game Go. AlphaGo Zero, as the new program is called, beat the original AlphaGo in 100 straight Go matches.

While the original AlphaGo was trained using data from human games of Go, AlphaGo Zero was trained using reinforcement learning, meaning that it did not have human input in developing its intelligence. As a result, AlphaGo Zero "is no longer constrained by the limits of human knowledge," blogged DeepMind's David Silver and DeepMind CEO Demis Hassabis.

View article

This article was originally published on Friday Oct 20th 2017
Mobile Site | Full Site