Prisoner's Dilemma and Mechanical Turk

[The Good Reverend Roger]

Douglas Hofstadter wrote a great deal on the Prisoner's Dilemma in Metamagical Themas.

The question he was exploring was, "Is cooperation rational?"

On a related note, a paleontologist recently put forward the notion that altruism is a survival trait.  I can't remember who it was, but I will try to dig it up.  He makes a fairly good case, and we've seen what self-absorbed greed brings.

[Bebek Sincap Ratatosk]

Douglas Hofstadter wrote a great deal on the Prisoner's Dilemma in Metamagical Themas.

The question he was exploring was, "Is cooperation rational?"

On a related note, a paleontologist recently put forward the notion that altruism is a survival trait.  I can't remember who it was, but I will try to dig it up.  He makes a fairly good case, and we've seen what self-absorbed greed brings.

RAW argued for it in Prometheus Rising... tribal systems = bio-survival and tribal systems only work if the members are working for the good of all.

I think that cooperation can be the rational choice, but I also think that the "rationality" of the choice depends heavily on the environment that the participants grew up in.

JW's cooperate fantastically with each other, I spent several years on their "Quick Build" teams where we would build an entire Kingdom Hall, including plumbing, electric, masonry, flooring etc etc in 3 days. The altruistic cooperation was amazing to watch... and all volunteer.

Yet, those same JW's individually, with a group of non-JW's don't necessarily display those qualities at all.

So in Leary's labels... If you're 1st and 2nd circuit are programmed "positively" with the "I'm OK, You're OK" script, then altruism would likely be a natural option, whereas if you imprint "I'm OK, You're Not OK" or "I'm not OK, You're OK" or "I'm not OK, You're Not OK"  then altruism seems much less likely.

Maybe?

[Cain]

Evolutionary psychologists (stop sniggering at the back) have suggested altruism, contrary to Ayn Rand et al, is indeed rational, for much the same reasons that Ratatosk and Roger have mentioned.  30,000 years ago, your survival was much enhanced if you were part of a group.  Groups could lay ambushes for animals, pick berries, share child-care duties etc etc all of which enhanced survival compared to those who, for whatever reason, went it alone.  Even only about 2500 years ago, banishment was still a favoured punishment, probably because of the greater risk an individual took by not being allowed to interact with his group.  Of course, there are environmental and cultural inputs as to what is considered "rational", and as always, it doesn't seem to apply often to those outside the "tribe".

[P3nT4gR4m]

Evolutionary psychologists (stop sniggering at the back) have suggested altruism, contrary to Ayn Rand et al, is indeed rational, for much the same reasons that Ratatosk and Roger have mentioned.  30,000 years ago, your survival was much enhanced if you were part of a group.  Groups could lay ambushes for animals, pick berries, share child-care duties etc etc all of which enhanced survival compared to those who, for whatever reason, went it alone.  Even only about 2500 years ago, banishment was still a favoured punishment, probably because of the greater risk an individual took by not being allowed to interact with his group.  Of course, there are environmental and cultural inputs as to what is considered "rational", and as always, it doesn't seem to apply often to those outside the "tribe".

You don't need to look as far back as 30,000 years. Take a look at group dynamics in any species which operates a pack nowadays. Lions, monkeys, caribou ... etc. A degree of altruism is hardwired into them or Darwin steps in to sort things out. Humans are, as far as I'm aware, the only species who's intellect has evolved to such a degree that they can break the coding and thus fail hilariously to pull off the kind of balanced society that even fucking ants don't find hard. 

[Golden Applesauce]

Douglas Hofstadter wrote a great deal on the Prisoner's Dilemma in Metamagical Themas.

The question he was exploring was, "Is cooperation rational?"


he pointed to a contest in 1980, where college kids competed to design the best computer program to play the prisoner's dilemma. It was basically a ten or twenty round game of the exact same scenario.

One robot's programming was to cooperate every round until his opponent defected, then defect for the rest of the game. Another robot randomly flipped back and forth, being totally unpredictable.

Hofstadter thought the most interesting program was one called Tit for Tat. It's programming was to do the same thing its opponent did in the previous round.

This means that Tit for Tat almost always loses (although it can tie if both parties cooperate for every round), but only loses by a very narrow margin.

For a little extra background:
The idea motivating the contest was that a repeated P.D. game is a different beast than a single instance of it, since the idea of establishing yourself as "trustworthy" over the long term came into play, and an opponent you screwed over could take revenge later, etc.  So they (I think it might have been someone at RAND or something) just organized a contest to see which strategies were good.

Each contestant submitted a program that played the Prisoner's Dilemma repeatedly.  I don't remember if every program played every other program, or just a sampling, but basically the scores (5-0 for defect/trust, 3-3 for trust-trust, and 0-0 for defect/defect) were cumulative across the entire tournament, each round of which was a 100 consecutive games of the Prisoner's dilemma.  So if in one round you managed to get trust-trust pairs every time, you were +300 points towards winning the tournament.

A lot of the submitted programs were pretty complex, since they were trying to figure out the opponents program over the 100 games.  The way the scoring was set up, it was better to cooperate twice (6 points) than to outright win or lose every other game (5 + 0 = 5 points.)  So the winning strategy would cooperate as little as possible while still managing to convince the opposing program to cooperate, while not being fooled in the same way itself.

The overall winning strategy turned out to be "Tit-for-Tat," which is conceptually very simple:
1.  Game one, play "cooperate."
2.  Every game, after the first, play what your opponent played the previous game.

So if your opponent defects, he's guaranteed to get nothing in the next round, since you defect just to spite him.  He might get 5 points for defecting when you cooperated, but the next round he gets 0, no matter what.  He'd have gotten 6 points if he had just played "cooperate" twice, since you would have cooperated with him.  Tit-for-tat "loses" or ties every round, in that the opposing program might get a few more points than Tit-for-tat did, but wins the overall tournament, since its matches delivered more points.  Tit-for-tat might "lose" 297-302 in a round, but when all the other programs had rounds something like 150-50, it still gets more points overall.

After Tit-for-Tac won hands down in the tournament, they ran the tournament again, which the explicit goal of finding a strategy that would do better overall than Tit-for-Tat.  No entrant sent in a program significantly better than Tit-for-tat.

[Triple Zero]

Actually it has even been mathematically proven you cannot do better than Tit-for-tat. It may seem obvious, but the proof is rather complex.

[LMNO]

Here's an interesting version, and one that may be better suited for real-life mapping.

You play for a small amount of money ($0.50 or so), but your personal information is given to the person you play with, and vice versa; that is to say, it's not anonymous.  There could be ramifications to acting like a dick, like getting phone calls or emails from the person you fucked over. 

The intention being that your actions are not limited to a single, anonymous event; they are usually determined by judging future outcomes.  If you can be a dick to an anonymous person on the internet, there's a greater chance you will be.  If you know that your actions now can affect you in the future, you might not be as much of a dick.

[The Good Reverend Roger]

Actually it has even been mathematically proven you cannot do better than Tit-for-tat. It may seem obvious, but the proof is rather complex.

Then my lifestyle is vindicated.

[P3nT4gR4m]

Here's an interesting version, and one that may be better suited for real-life mapping.

You play for a small amount of money ($0.50 or so), but your personal information is given to the person you play with, and vice versa; that is to say, it's not anonymous.  There could be ramifications to acting like a dick, like getting phone calls or emails from the person you fucked over. 

The intention being that your actions are not limited to a single, anonymous event; they are usually determined by judging future outcomes.  If you can be a dick to an anonymous person on the internet, there's a greater chance you will be.  If you know that your actions now can affect you in the future, you might not be as much of a dick.

LMNO invents game theory karma ITT