The Second Apocalypse

Miscellaneous Chatter => Philosophy & Science => Topic started by: Madness on December 07, 2017, 03:39:32 pm

Title: AlphaZero "Learns Chess" in Four Hours
Post by: Madness on December 07, 2017, 03:39:32 pm
Entire human chess knowledge learned and surpassed by DeepMind's AlphaZero in four hours (http://www.telegraph.co.uk/science/2017/12/06/entire-human-chess-knowledge-learned-surpassed-deepminds-alphazero/)
Title: Re: AlphaZero "Learns Chess" in Four Hours
Post by: Wilshire on December 07, 2017, 04:30:05 pm
Love it. Can't wait for chess player's to pull a Genie in Aladdin.

Scene: Genie sets up a game of pool against Carpet. New rack.
"Your shot. Pool's a man's game so being a rug, you'll be at a disadvan... "
*Carpet shoots, gets every single ball in on a single hit*
"...Pool's a dumb game anyway"

(sorry, couldnt find the clip)

Chess is much less impressive than people are lead to believe, but especially from a 'difficult for a computer" perspective. There are a finite set of moves, and its a solvable game - ie if both players play a perfect game, the person who goes first will always win.

In truth, though this should be clear since it took 4 hours, this achievement is far, far less impressive than the Go AI.

Still a great breakthrough for machine learning though :).
Title: Re: AlphaZero "Learns Chess" in Four Hours
Post by: Madness on December 08, 2017, 01:48:20 am
Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm (https://arxiv.org/abs/1712.01815)

I was reading more about it, apparently, AlphaZero beat AlphaGo in four more hours after the initial four (the latter adapted to play chess instead of GO).

In one of the second or third articles I was reading about it tonight, there was an impressive seeming anecdote about AlphaZero blowing through the entire 1,400 years history (which isn't accurate but whatever) of human Chess development.
Title: Re: AlphaZero "Learns Chess" in Four Hours
Post by: Wilshire on December 08, 2017, 12:44:47 pm
Machine Learning is the present day future, and its delightful.
Title: Re: AlphaZero "Learns Chess" in Four Hours
Post by: Madness on December 08, 2017, 01:53:35 pm
"Better link" from Wilshire (https://arstechnica.com/gaming/2017/12/deepmind-ai-needs-mere-4-hours-of-self-training-to-become-a-chess-overlord/) ;).