As a stupid human, I'd like to claim this 3-1 scoreline is a statistical tie.

The million $ still not coming our way. Why is it that when we win, it is a "statistical tie" and when the computer wins it is just "computer pwns humanity"? Doug Polk and the guys should take a lesson from the Amazon team. If we win, we freaking just win.

After killing Go those shenanigans will now try to bring down poker:

http://www.theguardian.com/technolog...omputer-casino

It's all for the good of the ~ecosystem~ humanity guys!

"And unlike Go, victory in that field could probably fund itself – at least until humans stopped playing against the robot"

Wat
Are they gonna launch it on iPoker skins?

http://www.aaai.org/ocs/index.php/WS...File/8811/8351

same guy (David Silver)

Most of NVG wants that computer to get started on the sex ratios problem. Namely, how to increase theirs above zero.

wp

the programs have already done relatively well. with the resources that google has they could easily create an unbeatable poker program.

https://www.technologyreview.com/s/6...through-poker/

Bumping for AlphaGo Zero

Cliffs: 100 games between AlphaGo Zero and AlphaGo Master (the one that crushed Lee Sedol)
AGZ 100, AGM 0

Here's a few links:

This is the Nature article, unfortunately it's paywalled and I couldn't find an arXiv version

A brief story in Nature about this, not paywalled

Another story that I like, from the Atlantic

I guess that's a wrap for games of perfect/complete information...

thx quantph

"AlphaGo Zero took a mere three days to reach the point where it was pitted against an older version of itself and won 100 games to zero."

Wow. We'll see if it's still around in 3 days.

Until this thread, I thought the game Go was Japanese (since "go" is the Japanese word for "five"). My parents have a pretty sick set, handed down a few generations from a ancestors in Kagoshima.

I think the stones are made of slate or something like that. You can tell they're a different grade than some store-bought set. My great grandparents thought enough of them to take them to the internment camps during WWII, at least that's what I've been told.

Anyway, pretty cool to learn that Go is a) this big a deal in some circles, and b) so hard to master. Never knew either!

Link to paper: https://deepmind.com/documents/119/a...ted_nature.pdf

I wouldn't' call it a wrap (yet) for game of perfect information, there are still plenty of games that have a higher complexity than Go. Go is a landmark because it's the most complex of all the classics games out there.

^which ones? I thought GO is the most complex by far

Off the top of my head, any complex complete information board/war game such as Europa Universalis, World in Flames, etc. but those are not "classics" like Go and Chess are.

It would also be quite easy to design a much more complex version of Go or Chess that would resist AI in its current state today, but that's not the point. Those games are useful as benchmarks for algorithms because we know human expertise is very high as we've played them for centuries.

One opening move in Game One has many players stumped. Says Redmond, “I think a natural reaction (and the reaction I’m mostly seeing) is that they just sort of give up, and sort of throw their hands up in the opening. Because it’s so hard to try to attach a story about what AlphaGo is doing. You have to be ready to deny a lot of the things that we’ve believed and that have worked for us.”

AlphaGo confirmed baby ragequitter.

SC1 and SC2 will be very interesting. I think if the AIs APM is limited in some way the bot will have a very hard time beating top players. SC1 and SC2 is also infinetely more complex than Go for a computer to understand and learn, I think anyone who says there isn't a huge gap between them is crazy.

One could argue whether limiting the AIs APM is necessary or not, I think it would be uninteresting to not limit it. I don't care about the computer's mechanical ability, I know it is far superior to humans. I am interested about its intelligence in understand the game and solving confrontations, using mechanical ability comparable to human limitations.

A better example to compare this would be any tactical FPS like CS GO. I would not care at all in seeing an AI battle a pro team if the AI would be mechanically perfect. It would not be a competition since the AIs mechanical perfection would far outweigh any tactical skill the human team could gain an edge over the AI.

Come with me if you want to live.

Rule of bot is that whatever you say is impossible will be the next thing they break.

There was a bot already who beat every single pro in LoL (Dendi) 1v1 every single time - created by Elon Musk's team btw. I actually don't think SC1/2 would be that big of a challenge for these things. (https://www.youtube.com/watch?v=9_cS7oWtp1o)

It was dota2 actually and it was only one character. Still it was really impressive.

I've been reading some of the recent comments and I think that some posters are missing the point. The achievement here isn't that the bot did learn Go, it's how it learned Go. It's just like poker, don't be results oriented - the process that was used to develop this bot is what's impressive.

You could take the process that was used for the bot and develop algorithms for any number of games/problems/tasks. At some point, it just becomes a question of developing the machinations to feed the game rules. The algorithm takes over from there and teaches itself and it teaches itself very quickly - that's quite powerful.

1v1 DotA2 is also infinetely less complex than 1v1 BW/SC2. A regular 5v5 game of DotA2 would be a better comparison to 1v1 BW/SC2 than a 1v1 DotA2 (this is not meant as a knock on DotA2, just a way of saying DotA2 1v1 is not comparable to 1v1 BW/SC2).

Regarding sc1 and 2 bots have vastly inferior strategy at the moment in certain areas, but they make up for it with a much higher amounts of actions per minute - it's quite a big multiple.

They have a pretty big advantage there.