DailyDirt: AlphaGo Plays Better Go Than Puny Humans...

from the urls-we-dig-up dept

In case you missed it, humanity has been dealt a decisive intellectual blow by a go-playing computer program called AlphaGo. We mentioned AlphaGo back in January when Google announced that it had defeated European Go champion Fan Hui and was challenging Lee Sedol next. So now that the results are in, AlphaGo has shown the world that artificial intelligence can best the best of humanity at our most difficult games. We've seen this already with chess, and if you don't remember, people tried to make a variant of chess called Arimaa that humans could hold up as a game people could win over computers (ahem, that didn't work). We still have Calvinball, Diplomacy and certain forms of poker....

• AlphaGo won the match 4-1, beating the 18-time world champion soundly and winning a $1 million prize. This is a major milestone for AI, and this win and its coverage will spur more AI research (and counteract the AI winter of the 1970s). [url]
• In the second game of the five-game match, AlphaGo made a very non-human move. This unorthodox move was described as "beautiful" -- and could lead to more non-intuitive strategies for human go players. [url]
• AlphaGo was defeated in game 4 by Lee Sedol, as Lee found a weakness in the algorithm to exploit. The commentary on the game will probably be discussed and continued by expert Go players for years to come, but it's likely that go-playing software will improve -- whereas humans won't be able to upgrade their skills so readily. [url]
• Plenty of naysayers predicted that computers would never reach this level of play and beat a top go champion. However, there's at least one person (computer scientist Feng-Hsiung Hsu) with written evidence who predicted in 2007 that go software would de-throne top human players before 2017. [url]

After you've finished checking out those links, take a look at our Daily Deals for cool gadgets and other awesome stuff.

Filed Under: ai, algorithms, alphago, arimaa, artificial intelligence, chess, fan hui, feng-hsiung hsu, game algorithms, games, go, lee sedol
Companies: deepmind, google

DailyDirt: Winning Isn't Everything

from the urls-we-dig-up dept

Not that long ago, we mentioned that progress towards an algorithm that could play the game of Go better than humans was on the horizon. It looks like our wetware shouldn't be too smug about being able to play Go now, but we can still have fun playing, right? And it'll still take a while before robots are any good at (non-contact) sports. Ping pong, FTW!

• Google has announced that its machine learning system AlphaGo has beaten a human expert in 5 games. That's impressive, considering that all other attempts haven't been able to reliably defeat amateur players. In March, AlphaGo is scheduled to play a top player in the world, Lee Sedol, to really test its abilities. And hopefully, this time the human player will lose more gracefully without accusing the computer's side of cheating. [url]
• There are some games that you can play to win, but it's not a very fun experience. Try playing Monopoly ruthlessly. If you played Monopoly as a kid, you probably didn't learn the rules correctly. And you probably up-ended the board to end the game. [url]
• Ideally, we humans should take advances in machine learning as an opportunity for more human-machine cooperation to solve the world's most dire problems. Together, machines and people should be able to solve the "wicked" hard problems of poverty, pollution, diseases and more. Maybe we can convince the robots to help us. [url]
• Oh. And Facebook's effort to get a computer playing Go that can beat humans is "getting close" -- but maybe it can get exponentially better by March? The guy sitting a few feet from Mark has a bit of time to really get his algorithms working. Or maybe it should challenge Alphago to a match to see how the two compare. [url]

After you've finished checking out those links, take a look at our Daily Deals for cool gadgets and other awesome stuff.

Filed Under: algorithms, alphago, artificial intelligence, deep learning, game algorithms, games, go, machine learning, monopoly
Companies: google

DailyDirt: A Minute To Learn, A Lifetime To... No, Go Is Much Harder To Learn, Actually

from the urls-we-dig-up dept

For many years, the ancient game Go has been held up as the game that artificial intelligence can't win. AI has beaten humans at several games quite handily, even games like Poker and Jeopardy! that should give humans a bit of an edge. Still, Go hasn't been cracked... yet. Any bets on when humans won't be so smug about Go?

• Apparently, Google is working on a Go algorithm -- and it might have a "surprising" announcement in a few months. Well, it won't be so surprising now, will it? And maybe if you've been playing Go online and you've noticed a player get really, really, REALLY good in a short amount of time... that's actually not a person playing. [url]
• Facebook AI researchers are also working on a Go project called Darkforest. This AI already plays decently well against many human players, but it's not beating the best humans. Combining neural networks and other algorithms might crack Go, as well as help improve Facebook addictiveness. Yay? Perhaps Elon Musk is right to be worried about advanced AI. [url]
• If you'd like to get an overview of Go and why it's hard from a software engineer's perspective, check out this talk. The history of Go is fascinating, and the challenges for AI to play this game are impressive -- so predictions for when AI will beat the best humans at Go vary widely. [url]

After you've finished checking out those links, take 10% off any $50+ order from our Daily Deals using the promo code DAILYDIRT.

Filed Under: ai, algorithms, artificial intelligence, darkforest, deep learning, game algorithms, games, go, machine learning, neural networks
Companies: facebook, fog creek, google

DailyDirt: Computers Like To Sit In Front Of Computers And Play Games All Day, Too

from the urls-we-dig-up dept

Artificial intelligence software has been getting better and better over the years at beating humans at their own games. Games like Connect Four and Checkers are already solved, and while we humans might like to point out that there are games like Othello, Go, Diplomacy and Calvinball that still favor human players, it may only be a matter of time before computers outwit us at those games, too. Check out a few more games that algorithms are learning to play better than human brains.

• A specific version of Texas Hold 'em (heads-up limit Texas Hold ’em) will likely be dominated by computer players now that an algorithm has minimized a "regret" function for playing it. Poker hasn't been solved, but humans better watch out when playing online to make sure their opponents are actually other humans (if it's even possible to tell). [url]
• A computer simulating ant behavior has found almost half a million novel solutions to the "knight's tour" problem in chess. This isn't really a game, but it shows how AI can use some pretty wild strategies to solve game-related challenges. [url]
• The game of Go (aka wei qi) isn't going to be solved by a computer any time soon, but Go-playing software is getting better against human opponents. To make move decisions, advanced Go AI programs play randomized simulations of entire games to try to pick between move options. That's not quite how humans play the game, but apparently it's a somewhat winning strategy to use against human players. [url]

If you'd like to read more awesome and interesting stuff, check out this unrelated (but not entirely random!) Techdirt post via StumbleUpon.

Filed Under: ai, algorithms, artificial intelligence, chess, game algorithms, games, go, knight's tour, poker

DailyDirt: Computers Are Beating Us At Our Own Games

from the urls-we-dig-up dept

Artificial intelligence is a fun topic -- especially when it's applied to playing games with humans. The classic "man vs. machine" battles are always entertaining... until we hit the Singularity, and computer AI just consistently trounces humans in everything. Here are a few more examples of cool AI projects.

• Microsoft Research is working on AI that can play the ancient game of Go -- and incorporating it into an XBox Live game called The Path of Go. This is a pretty cool project to try to get more people playing go, but adding avatars and some storyline about finding your missing twin doesn't sound like a more fun way to play Go to me. [url]
• Starcraft is another AI challenge that requires more than fast reflexes and a pulse. Instead of creating supercomputers that play MMORPGs perfectly, I'd settle for AI that just happily mines gold for me all day. [url]
• Computers can also beat us at really simple games that you wouldn't think need any intelligence to play. So don't play Rock-Paper-Scissors to the death, and never wage a land war in Asia. [url]
• Maybe football coaches should be replaced by computers, too. When your fantasy league is short human players, try a few bots as competitive players. [url]
• If only the neural nets that play 20 questions were a bit more useful... These kinds of programs were supposed to help diagnose medical diseases, but now they just play trivial games and advertise for Whoppers. [url]
• To discover more interesting AI-related content, check out what's currently floating around the StumbleUpon universe. [url]  

By the way, StumbleUpon can recommend some good Techdirt articles, too.

Filed Under: 20 questions, ai, game algorithms, go, rock-paper-scissors, starcraft
Companies: microsoft