Matrix Games Forums

A self-taught computer has become the world’s best player of Go, the fiendishly complex board game, without any input from human experts. DeepMind, Google’s artificial intelligence subsidiary in London, announced the milestone in AI less than two years after the highly publicised unveiling of AlphaGo, the first machine to beat human champions at the ancient Asian game. Details are published in the scientific journal Nature.
“It learns to play simply by playing games against itself, starting from completely random play,” said Demis Hassabis, DeepMind chief executive. “In doing so, it quickly surpassed human level of play and defeated the previously published version of AlphaGo by 100 games to zero.”

This is a major milestone in AI in my view - I understand it became the best player in the history of the game within some 70 hours!

Lets hope they don't let Mr Grigsby borrow it.......

With the amount of time a turn conclusion takes I wouldn't expect it to learn very quickly.

Also AIs are a lot better at perfect information games than at partially hidden games. WITPAE with its partial and flawed (aka fog of war) information is perhaps even worse.

ORIGINAL: geoffreyg
This is a major milestone in AI in my view - I understand it became the best player in the history of the game within some 70 hours!

https://deepmind.com/blog/alphago-zero- ... g-scratch/
It's about 20 days of training

Go is a game with simple rules, simple turns and, as Rafid noted, perfect information. So it is far easier to train the AI. AFAIR in most those kinds of games AI already surpassed humans, and we would see further progress of AI players there.

But practical applications are limited. You don't have a lot of real world scenarios that you can replay indefinitely with no outside interference

Let it play Ichi-Go now and we will see if this AI is any good.

Many thanks for the correction - in three days it "surpassed the abilities" of an earlier version that beat world champion Lee Sedol 4 out of 5 games.
I take the point of Go having full information but it will be interesting to see what progress AI makes in evaluating possibilities where it lacks this.

I've always advocated a "learning" AI for these complex war games we all enjoy. It can be done, but no one wants to bother to do it.

Regardless of the FOW in this game, the machine AI could learn to calculate probabilities and become statistically more likely to win a game.
Fine- let computers use their vast memories and lightning calculations to do things that require cold calculation.
Just don't hook them up to real-world weapons with no human control.

What!! Don't you want an AI to ask you 'Shall we play a game?" [:D]

Actually, BB...the game would be getting better because the players would be getting better. If we played an AI game as I advocate we'd be playing the BEST of our club members.

No one would hold a gun to your head saying you HAD to play AI, but the AI would actually present an challenge.

While I agree with you in principal, I think implementing a learning AI in WitPAE would present a seriously non-trivial programming problem. I recall a thread on here recently where the level of effort to re-architect this game was discussed, to include a state-of-the-art AI. Seem to remember the consensus was an investment of high 6-figures to low 7-figures would be required to accomplish the task.

If I manage to win the Powerball I would write the check. [;)]