Iterated PD

by Chris Bertram on October 20, 2004

Tyler Cowen “had a discussion of this”: a few days ago, but I think it worth a mention here: tit-for-tat was beaten in a recent iterated PD computer tournament. The winners entered a large number of different strategies programmed to communicate with one another. By signalling their existence to their confederates and adopting master and slave roles, some strategies were able to gain full exploiter’s advantage over many rounds and thereby build up huge scores. Non-confederates were systematically punished by strategies from this stable, thus damaging the scores even of conditionally co-operative rivals. Full details “here”:,1284,65317,00.html?tw=wn_tophead_6 .



John Quiggin 10.20.04 at 11:52 pm

The claim that the confederate entrants “won” is dependent on using a tournament-style payoff structure. The average score for the confederates was apparently low, but the masters got the top places thanks to the sacrifices of the slaves. (A nice metaphor for the actual Confederate economy, which produced much the same outcomes).


Motoko 10.21.04 at 12:22 am

I dunno, but isn’t the whole point of PD that every program tries to optimize its own result? What’s in it for the slaves?


dsquared 10.21.04 at 12:54 am

Well, what’s in it for religious martyrs, soldiers who die for the glory of empire, etc, etc? Life has lots of situations with tournament-style payoffs and (unsurprisingly) for this reason offers a fair few examples of Southampton-type strategies.


Ethesis 10.21.04 at 2:29 am

The fact that these games are so sensitive to the rules sets is important, as is the unavailability of threats.

But I think we learn from watching them too.


reinstein 10.21.04 at 3:49 am

tit-for-tat was beat in spatially distributed SPD games by tit-for-tat with occasional generosity or forgiveness. the result has been know for 15 years. see article by Lezek & Nowick in Nature in 1991 or so.


Patri Friedman 10.21.04 at 7:26 am

This result basically just exploited the fact that it was free to enter multiple teams. That plus the tournament payoff structure. Its not particularly applicable to real life – which is the interesting reason to study these games.

I don’t think DD’s examples disprove this assertion. Note that religious martyrs and soldiers both cost resources, significant resources, before they die. The extra Southampton entries cost exactly *nothing*. Just by creating extra entries, they were able to create extra points for their main teams, at no cost. Real life is all about tradeoffs – even sacrificial lambs and soldiers cost resources.

I still think its great that they exploited this design flaw, so we know it exists. But I don’t think it tells us anything about human behavior.


Jamie McCarthy 10.21.04 at 4:57 pm

Think you can beat TFT too? Come submit an entry in my tournament.

Comments on this entry are closed.