Google's AlphaGo Zero: The Beginning of the End for Humanity's Cognitive Superiority?

Two hundred thousand years ago, evolution invented a frontal lobe for a small group of primates in the African Savannah. That frontal lobe coupled with an opposable thumb have reshaped the planet and dominated every other species living on it, even driving more than a few into extinction.

A new technology inspired by that same frontal lobe has surpassed it in at least one domain: the most complex game ever invented. The Chinese game of Go was invented thousands of years ago and has more than 10^170 possible board positions (that is 1 followed by 170 zeros). As recently as a few years ago, it was believed that mastery of Go will remain dependent on human intuition and creativity for at least another decade because that level of complexity defied even the most advanced computers in existence.

Over the past two years, however, first the European champion of Go, followed by one of its best international players and finally, in May 2017, the reining world champion were all beaten by AlphaGo, an artificial intelligence program developed by Google's Deep Mind subsidiary.

Impressive and unexpected as this was, this is not the reason for the question in the title. The versions of AlphaGo that achieved these prior wins were based on the accumulated human knowledge of Go developed over 3000 years by masters of the game in China and Korea. Frontal lobes could still claim that it was their smarts at the heart of the machine.

The reason for the title is an article recently published by Deep Mind scientists in Nature. The article reports on a new version of AlphaGo that relied on NO HUMAN KNOWLEDGE of the game. Named AlphaGo Zero, the program was only provided with the basic rules of the game and learned everything else on its own. After just 40 days of self-play AlphaGo Zero was able to beat 87-13 the version of the program that beat Ke Jie, the world champion just 5 months prior! AlphaGo Zero achieved a rating of 5185, about 1400 points higher than Ke Jie. What that means is that AlphaGo Zero is as far ahead of the world champion as he is from a beginner. It makes it almost impossible that he would be able to beat AlphaGo Zero. An amazing discovery was that when AlphaGo Zero was trained using human knowledge, it learned faster but never got as good. Human knowledge hindered the program's progress! And it achieved that in 40 days compared to 3000 years for humanity's Go knowledge.

AlphaGo Zero is a narrow artificial intelligence (good at only one task: Go) compared to the general intelligence frontal lobe. It's probably going to be two to three decades before AlphaGo Zero's general artificial intelligence descendent can achieve the same level of cognitive superiority across the spectrum of tasks that a frontal lobe can tackle. But given the implications for the losing species the last time around, it might be worth it for humanity to prepare!

For some other implications that this rate of artificial intelligence development could have on the economy and employment check this article.

Please let me know if you liked this post using thumbs up Like button below.