Alphazero

December 13, 2017

Last week, on the 6th of December, learning has changed forever
I am not using a hyperbole. I feel changed by this, this has validated so many things in my life
Stockfish is the leading Chess program out there. It is the World champion of computers, far better than any other humans. It is a continuation of Deep Blue who beat Kasparov 20 years ago
Stockfish wins by brute force. It has been taught by humans all openings, end game tables, millions of games,etc... this plus its computation power makes it a monster.
Last week it played Google Deep Mind’s Alphazero
Alphazero is not a Chess program. It is a generalized AI. It means it is not created with a specific purpose in mind, but with a capacity to learn anything through Reinforcement Learning
What is reinforcement learning?
Read this if you want to familiarize yourself with it: Click Here
Alphazero was given 4 hours to teach itself Chess. You read that right.That means no human games to analyse, no openings, no end game tables. Only Alphazero playing itself

100 games were played. The results? Alphazero won 28-0 with 72 draws
Stockfish was annihilated. I watched the games and i can tell you this is not an overstatement
How Alphazero won is actually the story there. It played like a human, albeit a superior one. Intuition, positional, long-term advantage,etc...
Mind-blown does not describe how i feel right now

The main thing i took from this is this: for a while we have been trying to think like computers, thinking they were better
DeepMind just showed us that the most important thing is to teach yourself, to try random things, to forget the past once the lessons are learned and move forward
DeepMind has just taught us the most important lesson in the last 50 years: we were right all along, being more human is the way to go