I was reading this paper about incentive mechanisms in peer-to-peer networks as a part of my midterm exam. It's kind of old; was written back in 05. There was this part of it that reminded me of "Industrial Engineering Concepts"! I mean like those "D'uh!! who doesn't know that?!" guidelines that we miserably fail to follow in practice.

Anyhoo..., apparently there was a competition based on the iterated prisoner's dilemma in which 62 teams of computer scientists and game theoreticians submitted entries to. Each entry was put up against others and the ones with the highest total score won the tournament.

If you're too busy to read the Wikipedia page, it's enough to know that it's a game between two players trying to collect pay-offs. They trade stuff through a magical non-transparent channel simultaneously. If they both send items they'll benefit (say 3 each). If one gets clever and receives an item without sending one, she'll benefit even more (say 5) and her opponent is screwed with zero payoff for being the stupid one who share without receiving! (life's unfair, I know!) (I sounded like joey preparing a toast for a wedding! lol ref pps). And finally in the case in which both ends play the "clever" [air quotes] role of not sending they'll earn a very small benefit (1 each).

Based on those payoff values it's easy to see what the dilemma is about. The expected value for defecting is (5+1)/2 = 3 and for cooperating (3+0)/2 = 1.5! In fact you'll never gain less that your opponent, i.e. 5>0 and 1=1. So it may look like a good idea to defect all the time. But is it?

Just like me, by now you're probably sure you've played a game with similar rules somewhere before! Hmmm... what was the name? ["It's called life you idiot!", god's shouting!] Here's the part I liked the most in the article. It's R. Axelrod's analysis of the winning entries, i.e. those who earned highest payoffs after playing against all other entries:
"

1. They are nice, which means they cooperate as long as the opponents do. In other words, they never defect first.
2. They are retaliatory, which means that they stop cooperating if the opponents defect. This property prevents them from being exploited.
3. They are forgiving, which means that they cooperate again if the opponents resume to cooperate after mutual defection. This property helps reconstruct trust.
4. Their behavior is clear, which means that they signal to the opponents that they act reciprocally. This property leads the opponents, particularly those who "probe" others, into cooperation.

"

How many of these "winning entry properties" do we have?

ps. You AI people are probably thinking about modifying number 3 and adding a profile list for keeping track of people's previous responses to your forgiveness. After all not everyone is going to turn out to be Jean Valjean.

pps.

[Joey just got ordained via the internet so that he could perform Monica and Chandler's wedding]
Joey: Hey, I started working on what I'm going to say at the ceremony. Wanna hear it?
Monica, Chandler: Yeah.
Joey: We are gathered here today on this joyous occasion to celebrate the special love that Monica and Chandler share.
[Monica and Chandler look impressed]
Joey: It is a love based of giving and receiving as well as having and sharing. And the love that they give and have is shared and received. And through this having and giving and sharing and receiving, we too can share and love and have... and receive.
[later]
Joey: Okay, you guys, I've got a little more written... are you ready?
Chandler: Yeah, yeah. Okay.
Joey: When I think of the love that these two givers and receivers share, I cannot help but envy the lifetime ahead of having and loving and giving... and then I can't think of a good word for right here.
Monica: How bout receiving?
Joey: Yes!

Robert Axelrod. The Evolution of Cooperation. Basic Books, 1984.
Seung Jun, Mustaque Ahamad. Incentives in BitTorrent Induce Free Riding, 2005.