Skynet plays chess!

Nope. There are AIs for playing FPSs with only access to the pixels and current health/bullets. I think it's this one: arxiv.org/abs/1609.05521

This is a rather incredible thing. Some of the games it played, it didn't just beat Stockfish, it spanked it.

This is a big deal in the chess world. Whole new concepts may develop as a result of this. One game, for example, the AI managed to find a novel way to cripple a knight in a well known opening. Surely to affect chess theory. There are some giddy chess analysts out there this week.

There was some work at MIT doing this. Not sure about recent advances.
news.mit.edu/2011/language-from-games-0712

>(((HURR PANIC)))
Sorry Schlomo, we're not going to do any such thing. AI left out of the hands of (((evil))) men serves us just fine. And stop blaspheming thanks.

Sounds like it's not too far off to see an AI possess the capacity for rudimentary thought. I figured an reinforcement learning model that utilizes multiple competing reward mechanisms (or varying scope and priority) would achieve quite serious results. Though the real challenge would be to generate them through a stochastic process and have it evolve useful reward mechanisms (by evaluating their realism) from experience and stimuli. The idea would be treating reward mechanisms like "thoughts" an AI makes of its environment and stack up to build more generalized motivation and learning. In turn the core reward mechanism would be simplified to being: "Observe, imagine, interact, confirm" with no specific task assigned (except by whatever the other reward mechanism evolve towards). Personally I think such a design could result in it inheriting biases from experience giving it an actual "personality" of sorts and allow for higher level thinking. I'm curious if such an experiment has been attempted or this is even viable yet.

I'm genuinely worried about AI risk but this is just chess. This is not dangerous.

Isn't that from Hitchhiker's guide?

>chess.com/news/view/google-s-alphazero-destroys-stockfish-in-100-game-match

If you had actually read the article:
Stockfish was playing without opening book and the hardware was not the same.
So it's not that much meaningful yet.

op is probable being not exmplanatory enough, and I don't feel interested in reading the article. But, I think the gist of it is that Google Neural network is just a *simple* algorithm that has been trained for for hours(when in the past it could have been weeks or months) to produce something that beats an AI specialized in chess that was probably put a lot of effort into and well tought and such.

This means that given a correct interface, many thinking, office jobs could be just replaced with this thing, trained in few hours.

Imagine how fun would it be if the people losing their jobs first were the smart ones and not the truck drivers!

I don't remember the details, but I think Microsoft Research did something like that. Instead of a single AI solving whatever it was (go?) they divided it into several managers whose objectives weren't always aligned.

sounds like youre a redditard who has no capacity for rudimentary thought.

This article is jumping the gun. Nothing is "changed forever" until the Alpha team plays Stockfish using realistic settings and comparable hardware. They made Stockfish play hopping on one leg (low hash table RAM) with an arm amputated (opening book removed). Then, they gave themselves a supercomputer running against Stockfish on a desktop PC.

Would be interesting if they also layered rewards so instead of competing rewards representing goals, a bunch of smaller and sometimes conflicting rewards for simple behavior clustered together to form an objective reward function.

Sensationalist bullshit like this is why Ai can simultaneously be over-hyped and under-appreciated

No actually this is fun, let's see where it goes!

self driving trucks are already a thing though.

I for one welcome our new AI overlords

We will all lose our jobs very soon. And we are not prepared to deal with that.

California 13013
Akiba 3301
who will win?

Why is being replaced a bad thing? If bots are better than organisms at doing organism stuff, then what's wrong with them replacing us? The natural order of things is survival of the fittest. God would be proud of us.

Reading over my own reply, that was probably a very cuckly thing to say, but still. If we managed to make something good enough to replace us, then we'll at the very least die having accomplished something good. Maybe that's end-game evolution, just advanced species making stuff that replaced itself.

If you want a biological answer: because the whole point of an organism is to pass on its genes. For a human answer: because fuck you, robots can kiss my hairy flesh ass.
But this shit is nowhere close to being able to outsmart mankind. At best it's useful for (((some people))) to better control the rest of us, but it won't act out on its own and take over.

wow its nothing. Neural networks can't even jump over a fucking block. I'm not worried.

kinda off topic but anyone got or can find 2 ai's made by different companies then they try to make them converse? like very advanced ai's please

That's not what happened you retard. Chess is still far from solved.

...

Indeed. It may appear a simple problem that we would expect to have solved by now, but chess scales exponentially. The brute-force "calculate every possibility" approach is still beyond even our most powerful computers after a small number of moves. Same issue with the travelling salesman problem.

Being able to calculate potential outcomes of a chess move versus an opponent in a 2D grid where each chess piece has limited movement is lightyears away from general purpose reasoning and even basic sentience, both which are required for the "I am god" outcome you're worried about.