Earlier this month, in a paper published in Nature, Google DeepMind researchers revealed that theyve attained the holy grail of artificial intelligence game-playing: besting the European champion in the ancient Chinese game of Go.
Go is deceptively simple: players put down black and white stones on points on a 19 x 19 gridded game board (thats 361 squares; smaller game boards are available for shorter games). The objective is to control real estate on the board without getting surrounded and captured by the other players stones. The game gets more complicated when players link stones, which makes them harder but more enticing to capture. For AI, the sheer number of possibilities (2.082 x 10170, according to mathematiciansmore atoms than there are in the universe, in case youre counting) makes Go incredibly difficult to play and win.
But if its one thing we know about AI, its not to count it out for long.
DeepMind, a British AI company that Google acquired two years ago, has a simple mission: to solve intelligence. This means understanding what intelligence is and how it workswhich is easier said than donebut DeepMind has taken another major step toward that end.
The big leap made by DeepMind and other AI creators involves the development of artificial neural networks (ANN), which emulate the neural networks in human brains. In an ANN, nodes function like connected neurons, collectively processing information. The most exciting and game-changing feature of neural networks is the ability to learn by experience and example, which helps gaming AI effectively respond to opponents.
Thats exactly what Googles AI AlphaGo did, employing both value networks and policy networks to understand the layout of the pieces on the board in any given moment and then decide on a play. The AI beat the three-time European Go Champion five games to none, signifying a major triumph in the AI world.
Humans have a history of measuring AIs meddle by throwing down the gaming gauntlet. IBMs Deep Blue supercomputer beat world chess champion Garry Kasparov in a 1996 match, though it ultimately lost the set. IBM developers upgraded Deep Blue and in the 1997 rematch, the AI won the majority of games and a place in history. Deep Blues programming is the precursor of todays artificial neural networks, with 30 nodes each containing microprocessors that allowed it to weigh 200 million moves a second. Kasparov claimed that IBM cheated and that the company dismantled the supercomputer before he could have his revenge. IBM denies those claims and says one of their computer scientists offered to rebuild the Deep Blue program in another machine, but Kasparov refused. Who can blame him?
That same year, the computer program Logistello went 6-0 against the Othello world champion (17 years earlier, a program called Moor went 1-5 against the human champion).
In 2011, IBMs Watson computer beat the top two human Jeopardy contestants of all time, outscoring them by over $50,000 (dont feel too bad for themthey got $300K and $200K for playing). Watson not only has four terabytes of information stored away, including all of Wikipedia, and infallible statistical capabilities, but its also capable of understanding human speech and responding appropriatelya feat unto itself. Watson had another major advantage: it didnt have to use the handheld buzzer.
An AI program called Polaris beat human poker champions in 2008, humbling its human counterparts when it adjusted its approach in the middle of a 500-hand round. Soon, robots with AI will be able to sit at the table with humans and win face-to-robotic face. Algorithms are better than humans at assessing whether people are lyingtheyre right between 77-82 percent of the time, according to one study, compared to 53 percent accuracy for humans. Facial recognition technology, a common component of many social robots, can link facial expressions and emotional states as accurately, if not more, than humans can. Good luck hiding that tell against a robot.
Sowhat games can humans win against AI and robots?
Not ping pong. Or beer pong (unless youre looking to get wasted).
Definitely not rock, paper, scissors.
A robot will break your Rubiks cube-solving record, no problem.
Our chances are pretty good at soccerfor now. Or maybe charades.
Suffice it to say, we should enjoy winning the odd game of hearts against the computeror other humanswhile we can. And be warned: if you invite robots to your game nights, be prepared to lose.
source http://allofbeer.com/2017/07/25/dont-pass-go-ai-will-beat-you-at-pretty-much-everything/
No comments:
Post a Comment