Can You Beat A Computer At One Of The World's Oldest Games?

As the world continues its love affair with Watson. the Jeopardy champion computer, The New York Times creates a simple online game that demonstrates how an artificial intelligence can outsmart a human at a slightly simpler game.

Rock, Paper, Scissors has been played in some form or another since China's Han Dynasty (206 BCE – 220 CE), and that's only recorded history. I'd guess that ever since early man figured out how to make symbols with their hands he's been arbitrarily deciding one is better than the other and using them to solve disputes. It may have just been Rock Vs. Other Rock, but it got the job done.

While the element of chance factors into the game's outcome, there is a key strategy successfully playing Rock, Paper, Scissors: Know thy opponent.

Studying how an opponent plays is an integral part of the game, allowing a player to recognise patterns of play that give them an edge. Is the opponent simply mimicking the player's choices? Do they tend to select the opposing choice to their previous selection? Is there such a thing as playing Rock, Paper, Scissors in a truly random fashion?

That's now what The New York Times Rock, Paper, Scissors computer seems to believe. It studies your moves as you play, identifying patters in your selection and using them to predict what you'll throw at it next.

The game presents two different AI computers to play against. One is a novice computer, learning the game for the first time. The second uses a database of more than 200,000 moves to do battle.

I played both. Let's see how I did.

Versus the Novice Computer

The novice computer was only just learning the relationships between rock, paper, and scissors. Once it figured out the relationship between the three, it started looking for patterns in my play. It couldn't find any series of four throws that matched my last four, nor three throws. I had kept things random enough that the only pattern that occurred was one of two throws, and it only happened once.

Still, the computer came pretty close to beating me, only one win short of a tie. That's pretty smart.

Versus the Expert Computer

The veteran computer beat me soundly, drawing on a database of more than 200,000 moves to help predict what I would throw. It went deep into the sequences I was putting down, calculating the odds that the fifth throw of five would fall in line with the fifth throws of five from all of those previous rounds.

In short, it kicked my arse.

So what does this Rock, Paper, Scissors robot teach us? Even a game that weighs heavily on human intuition can be broken down into patterns even a simple computer can understand. That, or I just suck at the game. Hit up the link below to try your hand.

Rock-Paper-Scissors: You vs. the Computer [The New York Times]


Comments

    so ... this is skynet huh

    WOW a computer can remember more and calculate faster than a human.
    Don't write the program and see how good it is

    So... is the only way to win, not to play?

      Would you like a game of Global Thermonuclear War?

    I won 12-5-4 in a 21 round showdown.
    The machines have nothing on me :)

    7/2/1 Novice, 5/4/1 Expert.

    *flex*

      PS - Does this make me the new John Connor?

      XD

    I can't even beat a real person let alone a computer.

    Expert: 13/14/10. Started off getting beaten badly, but my dice rolls were too random for the computer. It'll even out with a few more.

    That was pretty fun, and impressive although after playing it for a while you could start to see how the logic worked and come up with counters, although being a human with a limited memory span I started forgetting the patterns I had "trained" it on. That is to say, I trained it up thinking I followed one pattern and then deliberately inverted or changed it to attempt to confuse the computer. It worked for a while but then I started confusing myself as to what I had been using. You really need to play for 20 or so rounds before the results mean anything.

    On novice it seemed to learn pretty well or follow the law of averages, creating an even distribution of 33-28-33 over time. On expert, with 200,000 samples to choose from, I wasn't as good with 37-40-52. Still, if I had access to that many patterns and given that the average chance of winning would be closer to 50-50 than 1 in 3 I'm sure I could figure it out too.

    We should all just play Rock-Paper-Scissors-Lizard-Spock.

    Once again, (?) I got 7/7/7 on expert.

Join the discussion!

Trending Stories Right Now