This week, Google’s AlphaGo defeat a grandmaster in the complex game Go — an artificial intelligence landmark (see”How success for Google’s Go AI is stoking panic in South Korea”,”Machines are teaching themselves to grapple with the real universe” and”People hit back: How Lee Sedol won a match against AlphaGo”). Here’s what the experts say AI’s next big challenge ought to be.
No-limit pokerGo represents the ultimate in games where all of the information is currently available to players. But AI still fights with games where information is incomplete — like poker, where a participant doesn’t understand what card is coming .
“Computers have beaten the best individuals at heads-up limit Texas Hold’em, but not yet at no-limit, a far more complex game,” states Peter Stone at the University of Texas at Austin.
Diplomacy:”AlphaGo doesn’t know the significance of some of these symbols it is really adroitly manipulating: it doesn’t even know that it is playing Go,” states Mark Bishop at Goldsmiths, University of London. So he suggests the strategy board game Diplomacy, where gamers pose as European forces competing for land and resources.
Diplomacy embodies lots of the obstacles between current and true AI. “Interestingly, it is a game which in theory a computer could play quite well, as moves are communicated in writing,” says Bishop. However, it would have to pass the Turing test — people could team up against the AI when they figured out which player it had been.
“These spins on gambling go past the mathematical challenges being breached by present AI”
StarCraft: In Go, there could be approximately 300 potential moves at any moment. In StarCraft, a plan video game with countless pieces, there could be 10300. “You can’t even examine all possible moves in the current condition, let alone all of potential future transfer sequences,” says Stuart Russell in the University of California, Berkeley.
Instead, the AI might have to think about its actions and goals on a higher level, then work out a plan to get there — requiring reasoning methods applicable to a wider range of real issues.
Dungeons & Dragons:”What we’re seeing AlphaGo isn’t attempting to establish or disprove a humanlike sense of reality or believability, but instead is purely goal-centred — to win the match,” says Julie Carpenter at the California Polytechnic State University in San Luis Obispo. She says it’d be interesting to throw AI at some thing like a role-playing game. There, the machine’s goals wouldn’t be as obvious. It would have to rely on skills like social communication and higher-level situational awareness so as to succeed.
Cheating: Individual players may read their competitor’s faces and body language for clues about what to do next. They’re also able to get ahead using deceptive strategies, like misdirection. Could a robotic hustler ever spot those false behaviors — or even cheat without being discovered? “These spins on gambling go beyond the mainly mathematical challenges which are currently being breached by present AI,” says Ronald Arkin of the Georgia Institute of Technology in Atlanta.
The real world:”I am not particularly interested in seeing AI pitted against other matches,” says Murray Shanahan at Imperial College London. That’s helpful for analyzing an algorithm or new learning procedures, he says, but the true frontier is the real world. “When machine learning is as good at comprehending the everyday world because it is at Go, we will be well on the way to human-level artificial general intelligence.”