Now! After centuries of mystery!
Get the low-down on the throw-down showdown!
Artificial intelligence to the rescue!
The ancient Japanese game of janken, known as rock-paper-scissors in English, has been mastered by a robot powered by artificial intelligence. Don’t remember rock-paper-scissors?
Then you haven’t been in a school yard, a terminally boring car ride or a Florida courtroom lately.
Here’s how it works:
On the count of three, each player extends her hand shaped to form either a rock (clenched fist), a leaf of paper (flat hand) or a pair of scissors (two fingers spread in a scissors-like V.) Rock crushes scissors, scissors cuts paper and paper wraps rock. To win, you form your hand shape in anticipation of your opponent. If she throws down a “rock,” you want to throw down “paper.” If you think she will throw down paper, you would opt for scissors and so on.
It’s a guessing game, the kind of game that makes Tic-Tac-Toe seem intellectually challenging. However, the unpredictability and the 50-50 odds of rock-paper-scissors make it useful to settle unresolvable conflicts: which team bats first, who is better—Superman or Batman, or where a legal deposition should be held.
In 2006, litigants in a civil case in Federal court, Avista Management v. Wausau Underwriters, contested every step in the proceedings even down to the location for a deposition. After multiple motions and counter motions, the judge ruled in favor of rock-paper-scissors justice:
…counsel [both lawyers] shall engage in one (1) game of “rock, paper, scissors.” The winner of this engagement shall be entitled to select the location for the 30(b)(6) deposition to be held somewhere in Hillsborough County during the period 11–12 July 2006.
They don’t teach rock-paper-scissors strategy in law school. However, strategies are studied in Japan where humans compete in janken tournaments. Strategies led to metastrategies and metastrategies led to algorithms which led to the First International RoShamBo Programming Competition in 1999. Like pokemons, the algorithms compete on behalf of their masters to win cash prizes.
The algorithms select play strategy by anticipating opponents’ moves. These predictions are calculated from an opponent’s history, correlating past sequences with current patterns; frequency of specific moves, favoring rock over scissors, for example; and, when all else fails, the algorithm guesses. Using these strategies, the algorithms are adept at second-guessing and triple guessing, but not more than that. There are only three options. Unless they second-guess their triple guess of the first guess.
Confused? I’d guess so. And, you can see right away that this a problem crying out for more sophisticated artificial intelligence. Enter the robot.
In 2012, engineers at the University of Tokyo developed a robot hand that swept the janken field. The robot hand wins every time, never even ties. Was it a single, brilliant algorithm? Was it a self-learning machine like DeepMind’s AlphaZero and AlphaGo? Was it a major artificial intelligence breakthrough?
No, the robot hand does what any aggressive, sneaky human would do. It peaks.
Embedded in the robot hand is a high-speed camera that instantly records the slightest movement of a human hand. Within a millisecond, the hand identifies the opponent’s hand shape, selects a winning shape and wins.
The rock-paper-scissors algorithm tournaments may seem inconsequential—prizes of $1,000 have been awarded. (Human tournaments in Las Vegas offer up to $50,000 in prizes featuring players with nicknames like “Wicked Fingers” and “the Bulldog.”) The practice of designing an algorithm for three random options teaches better algorithmic understanding. According to scientists, useful algorithms for other fields have emerged from the process.
However, the robot hand conquers all with good, old-fashioned peaking.
If only the lawyers in Florida had known about the robot hand.
