Altherion Posted March 12, 2016 Share Posted March 12, 2016 In the past few days, there was a 5 game match between AlphaGo (a program of Google's DeepMind AI) and Lee Sedol, one of the best Go players in the world (perhaps the best one still actively playing). Computers have beaten top humans at games where they can't simply brute force their way through every possibility -- the archetypal example is chess. However, chess took half a century to "solve" and the way it was done didn't really bring one much closer to winning at games which are not similar to chess. For example, until AlphaGo came along, computers had only beaten human professionals when given a handicap. DeepMind is different in that it uses a neural network that essentially "learns" from both the games of its competitors and by endlessly playing itself. It had beaten the European champion last October, but given that Go is much more popular in Asia, that wasn't much of an indication of how this would go. Before the match, Lee believed that at the moment, he could still win: Quote Lee, who is 32, is confident that he'll beat AlphaGo either 5-0 or 4-1 this time around, but he's also smart enough to know that that won't always be the case. “I have heard that Google DeepMind’s AI is surprisingly strong and getting stronger, but I am confident that I can win at least this time," Lee said in statement on February 4. But later, in a February 22 interview with the Associated Press, he added: "But if artificial intelligence continues to advance, in a year or two years, it will be really hard to guess the results." In the first game, he made a non-standard, sub-optimal early move designed to confuse the AI, but it didn't get confused and made him pay for it (he started a bit of a comeback later in the game, but it didn't work out). From what I've read in the media, he played well in the next two games, but the AI played better. The third game just ended and the score is now 3-0 in favor of AlphaGo so it has already won despite there still being 2 games left to play. Looks like the xkcd difficulty chart needs to be updated. Link to comment Share on other sites More sharing options...
Gears of the Beast Posted March 12, 2016 Share Posted March 12, 2016 Google's a pretty scary company; their self driving cars drive better than humans, their AI can beat our best humans in some game I've never heard of and they have an insane amount of data on hundreds of millions of people. When we were in New Zealand a few years ago we almost missed our flight home because for some reason all 3 of us thought our flight left later than it actually was going to. The thing that saved us was that the flight info had been forwarded to me (my Gmail account) like a month earlier and since I have an Android (Google) phone it sent me a notification about a flight departing at x time. My phone warned us that our flight was going to leave without any human actually prompting the notification. I guess it wouldn't be that difficult to program Android detect emails from airline companies and automatically scan them for departure times but it's still pretty crazy. We're kinda living in the future. Link to comment Share on other sites More sharing options...
Jo498 Posted March 12, 2016 Share Posted March 12, 2016 I'll get scared when one and the same wobot can beat a chess or go champion, climb a tree and tie its own shoelaces. Link to comment Share on other sites More sharing options...
The Marquis de Leech Posted March 12, 2016 Share Posted March 12, 2016 3 minutes ago, Jo498 said: I'll get scared when one and the same wobot can beat a chess or go champion, climb a tree and tie its own shoelaces. People have this idea that the goal of AI is to pass the Turing test/imitate a human. It isn't. The goal of an AI is to perform specific tasks - in this case, be really good at a board game. And if you want to make an AI pass the Turing test, your biggest obstacle is getting the damn thing to understand humour and sarcasm. No-one's been able to pull that off, ever. We're not in danger of a SkyNet coming along any time soon. Link to comment Share on other sites More sharing options...
Jo498 Posted March 12, 2016 Share Posted March 12, 2016 I am very well aware of that. But one of the main point of the newer approaches (learning with "deep" neural networks) seems to be to get computers closer to general learning and problem solving. I had the impression that this was partly because of the paradox that computers/AI excel at what's hard for humans (huge calculations) but until very recently did rather poorly at all kinds of stuff that come "natural" to humans and are often mastered by 3-year-old children. Therefore I go with the combination of tying shoelaces and playing masterly chess, one "hard" and one "easy" tasks combined. I do not know anything about Go and believe that this is an important step, but at least on the surface, it is again one very specialized "theoretical" task, i.e. something computers are supposed to be good at anyway. Link to comment Share on other sites More sharing options...
maarsen Posted March 12, 2016 Share Posted March 12, 2016 2 hours ago, Jo498 said: I'll get scared when one and the same wobot can beat a chess or go champion, climb a tree and tie its own shoelaces. I 'll start to worry when it falls in love with a lathe. Link to comment Share on other sites More sharing options...
Horza Posted March 12, 2016 Share Posted March 12, 2016 We're putting a brave face on it, fellow humans, but a learning computer but smashing the world Go champ is game over, man. Game over. Link to comment Share on other sites More sharing options...
A Horse Named Stranger Posted March 12, 2016 Share Posted March 12, 2016 I am not an actual Go player. But as far as I know they don't have an official World Championship. What I have read, Lee is atm arguably the strongest Go player in the world, but he is not the World Go Champ or something. On somewhat related news anybody following the candidates (Chess, the game computer engines captured a good while ago). Aronian versus Anand should be a real treat. Link to comment Share on other sites More sharing options...
Altherion Posted March 12, 2016 Author Share Posted March 12, 2016 It will be interesting to see if Lee can win at least one game. The commentary on the internet is divided: the games so far have looked close, but it might be simply because AlphaGo does not try to rack up the score. That is, it could be that the AI is just a bit stronger than Lee, but it could also be that it is in fact much, much stronger and the games only look close to humans because there's nobody smart enough to appreciate that it has already won halfway through the game. Link to comment Share on other sites More sharing options...
Altherion Posted March 13, 2016 Author Share Posted March 13, 2016 Lee Sedol won the fourth game so the score is now 3-1. According to the commentators, the AI was doing well, but then decided to make some moves off to the side for some reason which didn't work out. I guess there are still some things they need to work out (well, it's either that or it's internal thought process is "See, little humans, I'm only a little bit better than you, nothing to be afraid of..." ). Link to comment Share on other sites More sharing options...
Humble AK Posted March 13, 2016 Share Posted March 13, 2016 I'm incredibly curious how the final match will play out. Sedol previously remarked about how the psychological pressure had a part to play in his performance against the Deep Mind, and now that the game has been lost, the pressure isn't as present. Apparently his victory in the fourth game was pretty decisive. I wonder if he'll repeat that in the final game? This is an interesting article on that fourth game:https://gogameguru.com/lee-sedol-defeats-alphago-masterful-comeback-game-4/ Link to comment Share on other sites More sharing options...
all swedes are racist Posted March 14, 2016 Share Posted March 14, 2016 On March 11, 2016 at 7:31 AM, Horza said: We're putting a brave face on it, fellow humans, but a learning computer but smashing the world Go champ is game over, man. Game over. but seriously, someone wake me up when one of these technological terrors beats someone at benjukka Link to comment Share on other sites More sharing options...
Horza Posted March 14, 2016 Share Posted March 14, 2016 Observation: it is way past time @HK-47 gave us its thoughts on this topic. Link to comment Share on other sites More sharing options...
White Walker Texas Ranger Posted March 14, 2016 Share Posted March 14, 2016 When a computer can consistently beat me in Rock Paper Scissors, then I'll be scared. Link to comment Share on other sites More sharing options...
Sir Thursday Posted March 14, 2016 Share Posted March 14, 2016 6 minutes ago, White Walker Texas Ranger said: When a computer can consistently beat me in Rock Paper Scissors, then I'll be scared. Be scared: http://www.bbc.co.uk/news/technology-24803751 Link to comment Share on other sites More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.