Google’s Alpha Go is now self-taught
First AlphaGo, created by DeepMind, a subsidiary of Google specialized in artificial intelligence, beat the world best Go players. The new version, AlphaGo Zero, using a single neuronal network, learnt to systematically beat its predecessor without human assistance. It was simply taught the rules and developed the winning strategies from scratch to beat the former version 100 in 0.
AlphaGo Zero takes AI to the next level since the software is not based on data from human games. According to its creators, it is no longer confined by the limits of human knowledge. But humans are still present: they’re the ones who programmed the algorithms allowing AlphaGo Zero to learn. With this research, Deepmind’s goal is not only to play but to apply them to scientific and medical issues, such as protein folding.
Spotted in 2018
- United Kingdom