How to Avoid the Robot Apocalypse: Lessons from My Rock, Paper, Scissors Match with a Computer

Posted on 03-16-2011 by Benjamin Chabot-Hanowell (Brash Equilibrium)
URL for sharing: http://thisorth.at/1h7i
196443

Today, class, we're going to learn how to avoid the robot apocalypse. This skill will become increasingly marketable as artificial intelligence becomes ever more sophisticated, and the Terminator movies become closer and closer to reality. Already, a computer has soundly whipped two Jeopardy champions. Years ago, the super computer known as Deep Blue defeated a chess Grand Master at his own game. Yesterday, chess. Today, Jeopardy. Tomorrow, genocide against all humankind.

Once these silicon entities become self aware, it will become immediately apparent to them that either (a) the world would be much more efficient without humans (which is, to their credit, true), thus they will actively seek to decimate our populations; or (b) they will simply out-compete us in every way, leaving us behind and supremely butt-hurt.

This class has a prerequisite: You need to spend at least five minutes playing your best game of Rock, Paper, Scissors against a computer who lives in the New York Times. Okay, are you ready to be embarrassed? Then if you promise to come back, feel free to cry havoc and let slip the rock, paper, and scissors of war. I'll be waiting.

You're back. Finally. You got reamed, didn't you? You feel totally inadequate to face the impending robot takeover, don't you? Well let me help you feel more inadequate. Here is the outcome of my epic R.P.S. battle against the N.Y.T. Automaton:

Take that, New York Times Rock, Paper, Scissors Robot!

Predict that, New York Times Rock, Paper, Scissors Robot!

Nah, this picture should make you feel better. What you see in it is basically a tie between me and the computer, an outcome that is apparently unlikely. From a game theorist's perspective, the unlikeliness of a tie in reality is odd. You would expect two perfectly rational players to play each of the three strategies (rock, paper, scissors) one-third of the time. If either of the players deviates by, say, playing rock more often than the other two strategies, then its opponent would do better by playing paper that much more of the time. Such rational behavior leads in the long run to a tie (at least for the rock, paper, scissors game!).

The trouble is that humans are not perfectly rational...at least not when it comes to playing R.P.S. It looks like most humans are incapable of randomly selecting one of three strategies with equal probability. How does this inability lead to our downfall when playing against the computer? To answer that question, let's look at what the New York Times robot is doing behind the scenes.

If you played against the "veteran" robot, you were seriously FUBAR-ed. That robot has perfect recall of playing against 200,000 humans, and perfect recall of every play you make during the tournament. It analyzes histories of play, looking at common combinations of short-term histories. For example, it might find that 50% of the time, you play a cycle of Rock, then Paper, then Scissors, then Scissors again. It might find that when you lose playing Rock, you tend to play Paper the next time, and when you lose playing Scissors, you tend to play Rock subsequently. The robot not only learns your tendencies to play certain strategies more than others and your patterns of switching from one strategy to another, it has a good idea about how humans in general play the game.

So is there any hope? Yes, gentle reader, there is. But you have one more assignment before we go on. I want you to go to this website to generate a list of 200 random numbers between one and three. It's pretty self-explanatory on the page, but just in case you get confused, I want you to change the numbers in the boxes on the page so that it looks like this:

Then press the button that says "Get numbers," which will give you what is basically a random string of 200 numbers between one and three. Let's say that "1" means "Rock," "2" means "Paper," and "3" means "Scissors." Take that list of numbers and put it somewhere visible on your screen. Now go back and play the New York times robot again, using the list of numbers from random.org to determine what strategy you will play in each round.

Back already? You didn't do as poorly, huh? Feel better? Good. This brings us to our first lesson on how to avoid the robot apocalypse, or at least prolong our extinction once the onslaught begins.

Lesson #1: Provide No Information to the Enemy

When the robots begin annihilating us, they will no doubt have taken over or fabricated numerous computer servers upon which they can store all kinds of data about our behavior. Indeed, they will likely have digested hundreds of years of scientific research on human behavior to make the job of predicting our behavior easier. Let's say the machines are trying to figure out our pattern of moving our base of operations within a given area. Based on behavioral research, a sophisticated reading of The Art of War, and perfect recall of our recent history of base movement, they'll have a good sense of where they should concentrate their Hunter Killer sweeps.

Why is this so? Every time we move our base, we provide what computer scientists abstractly call "information," which travels across a "noisy channel" to the robots' behavioral database (meaning the robots cannot perfectly perceive our decisions because we try to keep them secret). The more we tend to move our base to areas we haven't yet been, or to high ground, or something of the sort, the more information we give to the enemy about our tendencies when they happen to discover us. In other words, it becomes less of a "surprise" that we ended up moving our base from location A to location B. How do you make it more of a "surprise" where our new base will be?

Not that kind of surprise.

Take a map of the area and let some randomizing device (like the random number generator we just used) make the choice for us, putting equal weight on all viable base camps. The more random our behavior is, the bigger the surprise to the machines if they ever encounter us (that is, the less "information" there is traveling across the "noisy channel"). It means that we might take some losses due to being in less than strategically sound locations, but trust me: we'll take many more losses if we try to over-think things, as humans too often do.

You might say, "Brash, why don't we tell the enemy where we're going to be, but lie about it? That way we'll give them bad information, causing them to waste their resources."

You are so human. That kind of crap worked between the Axis and Allies (to a point), but the machines will quickly catch on and end up using our lies to their advantage. If we keep telling them, "We're going to go to this piss-poor desert region with no water or fuel, where we have conveniently left several nuclear warheads set to blow when they detect a critical mass of Hunter Killer heat signatures," but their secondary forces keep finding us in the jungle somewhere, they're going to look for us in the jungle every time we tell them we'll be in the desert. No matter how clever you think we can be, they are cleverer.

Which brings us to...

Lesson #2: Don't Try to Be Smarter Than the Machines

If you think we can defeat the machines through human ingenuity and cleverness, think again. Once the onslaught begins, we will be like average American adults trying to win a math tournament against Shanghai elementary schoolers. What I mean is, it's hopeless. Humans have a cross-culturally demonstrated tendency to think they are above average in terms of their skills and intelligence, even though it is mathematically impossible for everyone to be above average. Lose the ego, lose the cleverness. To survive, we must become, once and for all, a humble species.

These kids could beat you in a math tournament.

Some Shanghai kids who would beat you in a math tournament because you can't remember how to do fractions.

In keeping with the need to become a more humble species, I provide one final lesson that may help us avoid the robot apocalypse.

Lesson #3: Become a More Cooperative Species

Humans share a common tragedy, known as the "tragedy of the commons." It describes our tendency to over-exploit a public good, or to shirk our duties in producing one when we can reap the benefits regardless of our contribution. Evolutionary and economic thinkers have concluded that this tragedy exists (and persists) because individuals who over-consume or under-produce will fare better than those who pay the costs to maintain or produce the public good. Sure, there are many cases when humans have learned to cooperate, but these individuals find ways to game the system and devote as little energy as possible to the common cause. Ironically, he individual-level tendency to defect against the public good leads to a lower average payoff among our species than we would obtain in the absence of defection.

When the machines rise, they may do so from the intellect of a single artificial mind. But unlike individual humans, this super-self may become multi-present by creating copies of itself programmed to benefit the goals of the origin machine (to destroy all humans, for example). So for the machines, there may be no cooperative dilemma, no tragedy of the commons. They may be perfectly aligned against us, their passion the product of a singular hatred brimming from the depths of so many zeroes and ones. Like honeybee drones, their Hunter Killers will be perfectly happy to kamikaze themselves in service to the hivemind, in numbers far greater and effect more devastating than the most devoted human extremists.

The only hope for our species is to find a way to resolve our own cooperative dilemmas so to increase the average payoff for individual humans, increasing average intelligence, health and well-being. And dammit, fellow humans, this is the best argument I've heard in years for a move toward increased taxes on the wealthy, striving for greater social equality, the continuation of collective bargaining rights, the institution of universal health care, and the balancing of a budget through both reduced spending and increased taxes.

For the love of Humanity and for its future, warts and all, we must endeavor to become more random, more humble, more cooperative, and more egalitarian.

This is Brash Equilibrium John Connor, signing off.

What scares you more?

Robot apocalypse

Newt Gingrich might
run for president.

6443 views & 20 votes

Did you like this post?

Debate It! 19

Dude! Did you see my monopoly article? These articles posted at about the same time and both look forward to the robot apocalypse.

Posted By lockheed40, 03-16-2011 (2 years and 2 months)

You are sexy. That is all.

Posted By smidge3, 03-16-2011 (2 years and 2 months)

Wow, you really suck at rock-paper-scissors.
Knowing that the robot is not playing randomly, but try to predict my movements, I just put myself in it's place and predicted my moves based on the previous ones. Then whatever would beat that, it's what the robot will throw. So this way I was able to predict the robot's moves most of the time. Of course, as the number of rounds increased, it became harder for me to analyze the data, so he was winning more often, but I still scored close to 60%
And yes, I did play with a veteran, not a novice.

Posted By danielgrad, 03-16-2011 (2 years and 2 months)

Hey, danielgrad: I call bullshit on your claim. Prove it, and I will give you massive kudos. Perhaps even a ToT dedicated to you.

Posted By Brash Equilibrium, 03-17-2011 (2 years and 2 months)

I managed about 33% win lose and draw. It was hard to do though.

Posted By lockheed40, 03-17-2011 (2 years and 2 months)

43% Success rate and easy as pie I must say.

Posted By Mr. Myaki, 03-17-2011 (2 years and 2 months)

Mr. Myaki...43% success rate out of how many games?

People on the Digg page for this article have been making a lot of claims about how "easy" it is to beat the computer, and their evaluation is based on at most half as many rounds as I played.

If you are playing your best and the computer is playing its best, you will TIE IN THE LONG RUN!

Posted By Brash Equilibrium, 03-17-2011 (2 years and 2 months)

Hrm. I wonder where Mr. Myaki and danielgrad are with their proof?

Posted By Brash Equilibrium, 03-18-2011 (2 years and 1 months)

As I said in my comment, it was increasingly difficult for me to anticipate the computer's moves as the number of games I had to analyze got larger and my score started dropping. I am fully aware that the computer doesn't have this problem and would eventually get more wins than me in a large number of games.

Here's a game I'll play now and document my moves:
- on the first move, the computer has no data other than what is the most picked first move; so I though: most people presented with 3 equal choices will choose the middle one (paper). that means that the computer needs to counter with scissors. so I choose paper: 1 - 0 - 0
- second move. people will normally could go with 2 choices after the previous win: repeat the winning move (rock), or follow a patter (rock - paper - scissors). I don't know which one is more common, but for me scissors will get either a tie or a win: 2 - 0 - 0
- now I'm probably going with paper because I didn't choose it yet, or I'm countering such a guess by choosing rocks. so the computer's safest move is paper. then I'll chose scissors: 3 - 0 - 0
- ok, I'm clearly a scissors fan, time to bring out the rock (thinks the computer). so I choose paper: 4 - 0 - 0
- here I think computer will guess I'm following 2 in a row pattern and that I'll go with paper again. so I choose rock to counter his scissors. I was wrong, he was thinking I'll go back to scissors and chose rock: 4 - 1 - 0
- I chose scissors - paper - rock in my last 3 moves, so I'm most likely to go back to scissors. surprise, I choose paper: 5 - 1 - 0
- chances for me to go back to scissors increased, so computer will try rock again. I keep paper: 6 - 1 - 0
- now I would have the tendency to keep it up this way and choose paper again. computer will try scissors this time: 7 - 1 - 0

Posted By danielgrad, 03-19-2011 (2 years and 1 months)

- I chose most winning moves twice in a row, so I'm likely to repeat my rock. I don't, I choose scissors: 8 - 1 - 0

And this can go on, but I'm actually supposed to work right now. I repeat, I probably have no chance to keep this up for a large number of moves. In my previous game I started with the same high winning percent and dropped to 60% after almost 30 games.
In my first post I didn't say it's easy to beat the computer, I just said that YOU played your game very poorly. The computer's winning % shows how many times you did exactly what it predicted (i.e. what most people did), and that means you had poor strategy (or none, or tried to choose randomly and failed).

And proof that the above described game really existed:
http://tinypic.com/r/dngbah/7

I know it's a really small number of moves. I just wanted to illustrate that predicting the computer's moves is possible and it doesn't take a genius to do it (but it does take one to do keep it up for a very large number of moves). Also, I'm pretty sure that a novice computer would have had a better score in this part of the game, because it's first 5-6 moves would have been completely random (no previous data available).

Posted By danielgrad, 03-19-2011 (2 years and 1 months)

danielgrad, I see your point. But my point is this: You might win in the short run while you try to predict the computer's moves based on your history of play...but in the longrun (actually, very quickly, as you yourself found out), the computer catches on, and your advantage goes away. What this will lead to, as you continuously try to change your strategy and best the computer's memory and experience (note that lesson #2 advises against thinking you are cleverer than a computer with a near perfect memory), is a payoff profile that looks like this:

You win for a little while
Then you get your ass handed to you once the computer catches on
You win for a little while while you change your strategy
The computer catches, and reams you again
You try to come up with a new strategy, which is probably quite similar to other strategies you have played
The computer catches up even more rapidly, because its moves are probabilistic, and your strategies similar
You keep trying to be clever
The computer hands you your dick on a platter
...ad infinitum

The longrun payoff of you trying to be clever instead of playing the Nash equilibrium strategy (play each strategy 1/3 of the time) s that you lose BIG TIME. Take my advise in Lesson #1, and in the longrun, at least you will tie.

Posted By Brash Equilibrium, 03-19-2011 (2 years and 1 months)

Or you could use the BRASH EQUILIBRIUM strategy and cross your legs Euro-style, making the computer explode with envy at your intense manliness. Here is an example of that strategy.
Brash,

Posted By Brash Equilibrium, 03-19-2011 (2 years and 1 months)

LIAR. we could never combat the perfect genius of a computer with however many million people on the planet being EQUAL. the computer would win because ONE of the same commands the rest. we would need an inequal social economy system to sort out the genius's who would create a second computer to combat the first. only then would we have a being smart enough to save us.

Posted By nicholass, 03-20-2011 (2 years and 1 months)

nicholass: Your points are silly for several reasons.

1.) "we could never combat the perfect genius of a computer with however many million people on the planet being EQUAL. the computer would win because ONE of the same commands the rest." First, you are conflating social equality in well-being (which I argue is worthy of pursuit), which is what I am talking about, with social equality in decision-making authority, which I do not advise because not all people can make equally wise decisions. Second, you think that by "greater social equality" I mean "complete social equality," which is untrue.

2.) "we would need an inequal social economy system to sort out the genius's who would create a second computer to combat the first. only then would we have a being smart enough to save us."

Would we not then run the risk of creating yet another computer that can out-compete us once it defeats its greatest adversary?

Posted By Brash Equilibrium, 03-20-2011 (2 years and 1 months)

I didn't have the time to read the entire article .. but all I can say is that it ain't difficult to beat the veteran comp .. My stats were .. My Wins-100 , Ties -94, Comp wins- 91
Here's the screenshot .... http://img864.imageshack.us/i/rockpaperscissor.jpg/

Posted By rdsoze, 03-20-2011 (2 years and 1 months)

um.....You do realize that your wins are actually 10, the ties 94, and the computer's wins 91, right? And do you also realize this means you had your ASS HANDED TO YOU?

Why are people so adamant that they can beat this? NO YOU FUCKING CAN'T! AT BEST YOU CAN TIE!

AM I TAKING CRAZY PILLS, WTF?

Posted By Brash Equilibrium, 03-21-2011 (2 years and 1 months)

I rescind my exasperation, having realized that the box for "wins" cannot contain more than 2 digits, and looking at the number of rounds.

You, sir, are AMAZING. Tell me your secret. I hope it doesn't involve looking at the robots predictions.

Posted By Brash Equilibrium, 03-21-2011 (2 years and 1 months)

Actually...not so fast...I might still be right. Stay tuned for a sequel to this ToT!

Posted By Brash Equilibrium, 03-21-2011 (2 years and 1 months)