The Jeopardy problem, put simply, is this: Watson could pack more information into its databases than even Ken Jennings could remember, but computers struggle to understand what a clue in plain English is asking them—something kids can do without a second thought—and have even more trouble with the wordplay involved with many Jeopardy clues.
Thus, question-and-answer machines are a profound challenge for artificial intelligence. We all fight this every day when we struggle to tell Google what we're asking for. Just imagine the results if a machine connected to the knowledge of the world could understand your request in plain English: That's why IBM's commercials in the run-up to Watson's Jeopardy appearance have praised its potential to revolutionize medicine, research, and fields far beyond fun and games. Still, games have spurred machine language understanding, even before Jeopardy.
Ten to 15 years ago, Massey says, the hot challenge for programmers was crossword puzzles. After all, he says, there are myriad ways to fill in the board with letters, but only one that fits the clues. And like Jeopardy clue writers, crossword-clue writers are given to puns and bad jokes.
But perhaps the most intriguing word game computers play is Scrabble. It's intriguing because they don't destroy all humans in the competition.
The Scrabble board may seem to provide a perfect arena for computer perfection. After all, it's simple to give an AI a dictionary. Based on the board in front of it and its tiles in hand, you can imagine it choosing the maximum-point move in seconds while its human competitors mentally labor over the proper plays. However, Massey explains, the game of Scrabble is much more than a vocabulary contest.
A Scrabble endgame—the last few turns, when contests at the highest level are frequently decided—is the dream battleground of a tactician, not a linguist. The best Scrabble competitors play not only to score points for themselves, but to block off letters that would prove opportune to their opponents. They can do this because they've gotten a pretty good idea about what letters their opponents have by watching them play, and thinking to themselves: "If she had the 'X' and the 'Q,' she would've played 'quixotic' on that triple word score." That's a leap of logical reasoning beyond knowing all the words, Massey notes. And besides, top 1 percent Scrabble champs know all the words they need to. In any given game they may only see one or two words they didn't know before.
Make no mistake; computers are great at Scrabble. But their skill is not (yet) enough to overcome the best.
Massively Multiplayer Meltdown
Deception, intuition, and mastery of language give us gaming advantages that computers can't touch. But there's something else we have over them: Computers weaken when the number of players expands.
Scrabble can be played with several players, and Jeopardy is a game for three (though it's organized differently). But most of the games computer programmers have attempted to crack have one thing in common: They are heads-up, two-player, zero-sum games, like chess. Truly multiplayer games add many more variables that computers struggle with. Schaeffer's poker systems, for instance, don't play well against more than one opponent.
Massey says there are many more examples: "Risk is another game that computers don't play well at all, because it's a significantly multiplayer game. In Jeopardy you don't really worry much about strategic stuff, you just try to get as many points as you can. You can't play Risk like that. You can't just try to capture as many countries as you can. You have to think not only about how you're going to beat your opponents, but how your opponents are going to interact with each other."
Their technology may have conquered the world, but computers haven't proven capable of conquering the Risk map quite yet.
Minds like us?
Nobody knows the power of the human mind the way programmers and AI experts like Schaeffer, Massey, and Satz do—they've invested countless hours of their lives trying to match and then surpass the human abilities to play games, and they've come up against tactics like deception that machines struggle with and humans pull off with ease. In games like Stratego that play to human strengths, the obvious solution might be to make machines that resemble the human mind as much as possible.
But Satz doesn't see it that way.
"I have a completely different view: That you should never make an assumption that a computer program should operate on the same principles as a human," he says. "I don't think it's valid to make the starting assumption that the way humans have always played a game is necessarily the best way."
That's the power of a gaming computer. It's not just that computers can execute so many more calculations per second than we can, memorize more openings and endgames than we can, and evaluate many more possible options than we can. Because of those superhuman mathematical abilities, a machine plays the game in ways human players might have rejected as wrong, or never considered at all.
As a result, gaming computers aren't just trying to beat us. They're teaching us.
"Computers aren't bound by human preconceptions," Schaeffer says. "We've seen this in chess, and checkers, and bridge, and backgammon. Computers have revolutionized how humans play, because the computer doesn't have all the human baggage—the biases that we're taught. It comes to these games with fresh eyes."