August 9, 2008

Is doing-what-works the most successful social strategy?

Do you know the prisoner's dilemma? In 1979, Robert Axelrod wanted to find out which strategy would be the most effective with repeated prisoner's dilemmas. He organized a computer tournament for which scientist could send in their strategies in the form of a computer program. To his invitation 14 strategies were sent in by scientist from 5 disciplines. During the tournament the programs would play repeated prisoner's dilemmas against each other and against themselves. In total, 225 confrontations took place during the tournament.

The winner was the program Tit-for-tat which was sent in by Anatol Rappoport. Tit-for-tat was the simplest program which had just the following instruction: start positive and then do what the other party did in his previous move. In the nineteen eighties, Axelrod organized another tournament. Now, 62 strategies were sent by people who, of course, knew about tit-for-tat. Some programs were very complex and shrewd but the winner was, again, the simple strategy of Tit-for-that. Axelrod's research got a lot of attention among scientists and among a broader audience. It showed how cooperation could emerge on the basis of reciprocity, even when many individuals followed egoistical strategies.

Axelrod now wondered whether Tit-for-tat was also a stable and resilient strategy that would be able to defend itself against an invasion of egoistical strategies. To find this out, he did a new tournament in which he gave the strategies which had been sent in for his earlier tournaments the capacity to reproduce themselves. The tournament would now take place in multiple rounds. Each round represented a generation of strategies. De degree of success of a strategy in the first round determined how often this strategy would be found in the next round. By doing this, Axelrod simulated the principle of natural selection. By building in this evolutionary principle the strategies were getting stronger by each round. In earlier round there were still many over-naive strategies and many exploiting strategies but in later rounds both disappeared more and more. Axelrod did 1000 rounds and the result was that tit-for-tat was still the most successful and fastest growing strategy of all. If you wanted to describe Tit-for-tat in human psychological terms you could say that it is a positive strategy (because it always starts of with cooperation), that is also prepared to hit back when deceived (because it defects when the other person has done so), but is also forgiving (when the other start cooperation again, it does so too) and transparent/predictable (because of its simplicity and consistency).

Axelrods work has been very important. He wrote the book The Evolution of Cooperation: Revised Edition about it. But Tit-for-tat is not the most successful strategy after all, as turned out several years later. In 1993 a still more successful strategy was identified by Martin Nowak and Karl Sigmund. It was named Pavlov and it had the following instruction: follow the same strategy as in the previous move if it was successful, change if it was not successful. It is a pity this strategy was named Pavlov because Do-what-works would perhaps have been an even more appropriate name (if it works, go on, if not, do something else). Pavlov has one major weakness: it is powerless against the strategy: 'always deceive' (Pavlov keeps on switching when confronted with this strategy). Nowak and Sigmund found that Pavlov can only start to develop really well after Tit-for-tat has terminated the 'always deceive' strategies.

It is interesting to see how the simple and pragmatic Pavlov strategy, which comes down to do-what-works, is perhaps the most successful strategy for repeated social dilemmas.

2 comments:

  1. Coert,

    That's interesting that the Pavlov strategy can be the most successful but fails against an "always deceive strategy." Of course in real life a person following the "Do What Works" strategy would see that their approach of switching strategies was not working and might then switch to Tit For Tat or some other approach. I wonder if it's possible to come up with a combination of "Do What Works" and "Tit-For-Tat." One way is that if Do What Works Fails too many times then it switched to "Tit For Tat." Or it starts out as Tit For Tat and switches to Do What Works after many Deceivers have been eliminated.

    ReplyDelete
  2. Hi Rodney, Could be but I think simplicity has a great advantage. The more complex your strategy becomes the harder it may be to practice.

    ReplyDelete

Enter your email address:

Delivered by FeedBurner