AlphaZero AI Beats Champion Chess Program after Teaching Itself in Four Hours

Megalith

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
Google's latest artificial intelligence, AlphaZero, has defeated one of the best chess programs in the world after learning the game from scratch in just four hours. The “superhuman” AlphaZero AI played 100 games against rival computer program Stockfish 8, and won or drew all of them.

“Starting from random play, and given no domain knowledge except the game rules, AlphaZero achieved within 24 hours a superhuman level of play in the games of chess and shogi [a similar Japanese board game] as well as Go, and convincingly defeated a world-champion program in each case,” said the paper’s authors that include DeepMind founder Demis Hassabis, who was a child chess prodigy reaching master standard at the age of 13.
 
Now use this to solve all of mankind's problems, that would be great.
 
removing humanity should be the first priority. the human brain is f terrible at judging reality:
conspiracy_theories.png
 
Creig said:
Be careful what you wish for. The biggest problem facing mankind right now is man.
Click to expand...
Specifically not man, but man's free will.

"Wait what? You want a different outcome than the collective deems the best one? Time to make up a fresh batch of soylent green!!"
 
This is basically pavlovian conditioning for computer programs. Learning with reinforcement. It didn't choose to learn chess. The AI is an algorithm that was pre-programmed with the rules of the game, and that winning is the preferred outcome, then it learned trough trial and error. The reason it learned so fast is because it can do millions if not billions of iterations in 24 hours. Put the same ai against a human opponent and it wouldn't beat him in 1000 years, because they can't play enough games for the AI to learn from enough outcomes. Deeplearning is basically giving it such a high sample count that it will guess right with remarkable accuracy, but I wouldn't call that true intelligence, it's an imitation trough the sheer computational capability. It's nothing like human intelligence.
 
I wonder if it plays differently enough to discover interesting new lines, or if it's just better than Stockfish at the same methodology.

Stockfish and the like have an openings book, amd endings book, and everything in the middle is basically just running all future possible permutations to a certain depth (e.g. 8 moves deep) and scoring them. Highest score line gets played. It's a near-perfect tactician, but not much else.
 
c_porter said:
I wonder if it plays differently enough to discover interesting new lines, or if it's just better than Stockfish at the same methodology.

Stockfish and the like have an openings book, amd endings book, and everything in the middle is basically just running all future possible permutations to a certain depth (e.g. 8 moves deep) and scoring them. Highest score line gets played. It's a near-perfect tactician, but not much else.
Click to expand...

The Go version has apparently changed the pro scene of the game by coming up with new moves and strategies that humans never thought to play.
 
This is awesome. It's awesome because it illustrates exactly what human-designed AI should be the best-suited for i.e., solving problems where the rules and variables are finite and complete.
 
gxp500 said:
Now use this to solve all of mankind's problems, that would be great.
Click to expand...
Chess and Go are trivial to master compared to real world problems where all the rules are not known and the results of potential actions can't be reliably predicted and scored.
 
Creig said:
Be careful what you wish for. The biggest problem facing mankind right now is man.
Click to expand...

The biggest problem facing certain people, is certain other people. This AI will get used to help certain people, to the detriment of others.
 
James P Hogan wrote a pretty good book on AI development; it's called "The Two Faces Of Tomorrow"

Has a happy sappy ending, but examines the basic AI questions in a much less metaphysical way than "Destination Void", by Herbert.

Like Humanity, we learn to work together or Die. :)
 
You must log in or register to reply here.
Back
Top