.

I think people are missing the really big thing here.

It is a great achievement to be able to design a program, with the right hardware, to win a game of Go against a top human player. Of course comparable to Deep Blue beating Kasparov in chess or Watson winning in Jeopardy. But AlphaGo differs from Deep Blue and Watson. Quote from Wikipedia;

Of course AlphaGo has unique code written for it. But at its core it seem to use code that is applicable to a lot of stuff and digital problem solving. The program is learning new things and then applying things it has learned to better accomplish future tasks. I think this is a huge milestone (although the step might not be huge compared to other AI-technology at this time) from which on digital systems will become more autonomous and less and less possible to predict.

As someone noted above, the singularity is near. Whether that is a good or bad thing and if we really want this technology to be developed by competing, private companies is questions now more urgent than ever.

Starcraft and poker is interesting games since it is games without perfect information and this is of course another big obstacle for AI to overcome. For Starcraft it would seem scouting/gathering information would be of huge importance for the AI. In poker scouting isn't possible, on the other hand the possibilities in the game is far fewer than in Star Craft (or chess or go).
It would seem a game with many possible actions and without the possibility of gathering information about your opponent would be the hardest for an AI to master. Although there might not be such a game which is based on skill and not just random luck. Someone care to enlighten me?

Elon Musk, Amazon and a bunch of other tech heavy hitters recently created a nonprofit called OpenAI headed by a former Google scientist to supposedly pursue AI for the benefit of humankind, or something. I would think that was a move to compete with Google rather than anything having to do with the welfare of humanity.

Found a video of DeepMind solving Breakout in 240 minutes and doing it in a very clever fashion!

https://www.youtube.com/watch?v=V1eYniJ0Rnk

You have absolutely a zero ****ing idea about who Lee Sedol is. The guy reigned for over 10 years and is considered a living legend.

Human takes game 4.

AlphaGo made a mistake according to the tweet by DeepMind CEO

Sedol winning game 4 is very interesting. I don't understand the game well enough to appreciate what kind of mistake was made but it may be that the style of the bot is exploitable after all.

How much more complex is NLHE poker compared to go? Seems like to could apply the same kind of principles they used for go now they can solve a game of imperfect information?

The point is moot because even if it's making exploitable mistakes right now, it will self-adapt to patch it up.

Nice, the program has even developed the ability to show pity.

Go is a game of perfect information. Poker a game of imperfect information. Disregarding the perfect/imperfect information asymmetry go is many times more complex from my understanding. But since many games, and many decisions made in the real world, is made with imperfect information that is quite a big thing to disregard.
I do believe that if a big company like Google DeepMind with a lot of computer resources put the effort in they would in 6-12 months be able develop a program (with appropriate hardware) that in the long run would beat all the best players. But that is me speculating with not that much to stand on.

Maybe. But as the developers explained during the press conference after the fourth game, AlphaGo uses millions of games to develop itself. A few hundreds or even thousand games won't make much of a difference, and of course one game even less.

Google program beats human GoPro

https://en.wikipedia.org/wiki/Go_and_mathematics

Pretty cray we exist in a world where this is simultaneously occurring while Hillary/Bernie and Trump/Cruz are the remaining choices for president

lol

yea the candidates are all like 1990's MACs

old, out of date garbage.

Why wouldn't they be able to? Too few pixels to spread out to? I'm sure they'd be able to do stuff, after all when it comes to absolute perfection humans aren't even remotely close. When you practice a LOT you might be able to control like 5 units very well at once, but any more than that humans are painfully slow and inaccurate compared to bots.

Well, according to Go math experts: number of possible positions in Go is 10^761 compared to the estimated 10^120 in chess ... a difference which exceeds the number of atoms in the known universe

I'd say the human race just crossed into the DANGER ZONE

All these 10^700 game states types of metrics are irrelevant now that computers beat humans with deep learning. They're not doing it by searching game states brute force, they're doing it by literally learning.