Robert Axelrod
Hit back once, then come straight back
The winner was not the meanest, it was the quickest to resume
The Evolution of Cooperation, 1984, chapters 2-3: the two computer tournaments
Someone has crossed you and you are deciding whether to hit back, and how often. A tournament worked the answer out.
What actually happened?
In 1980 Axelrod wrote to game theorists around the world: send your best strategy as a program and I will play them all against each other over two hundred rounds. Fourteen arrived, some hundreds of lines long, modelling the opponent and probing for weakness. The winner was the shortest program submitted - TIT FOR TAT, four lines, from Anatol Rapoport in Toronto. Cooperate on the first move, then copy whatever the other side just did. The second tournament drew sixty-two programs whose authors all knew it would win, and not one of them could beat it.
None of the leaders moved first
Rank the first tournament and the clean dividing line is not sophistication, it is whether a program ever defected first. The top eight never did. They passed up every chance to take advantage and still scored highest.
The colleague who wins every small negotiation is usually the one with the fewest people left to call.
Retaliate at once, once only, then return
It answers a defection on the very next move, and the moment the other side cooperates again it cooperates again too - no arrears, no ledger. Programs that punished forever scored badly. They turned one accident into a permanent deadlock.
He missed one commitment, you went cold for six months, and the six months cost more than the commitment did.
It never won a single match
TIT FOR TAT cannot outscore anyone head to head. It always defects a move later and takes one hit more, so a draw is its ceiling. Its total was still the highest, because of what it drew out of the other side.
Keep score against your colleague and the thing the two of you could have built goes unbuilt.
Don't be too clever
The elaborate programs that modelled their opponent and hunted for a quiet advantage all lost. The problem was not accuracy. A player whose behaviour shows no pattern is indistinguishable from noise, and nobody cooperates with noise.
When your standard changes every time, people stop reading you and default to the most cautious thing available.
Its real weakness is misunderstanding
Add noise and the rule breaks: one misread move sends two TIT FOR TAT players into endless recrimination. Axelrod and Dion returned to this in Science in 1988 - where errors are certain, the rule needs room to let one go.
An unanswered message is not always a snub, and strict reciprocity will run you both off the road.
How do I use it today?
Where you are: someone has taken a piece out of you, and you are choosing between letting it go and never dealing with them again.
Ask first: after I answer this, is there a clear route back to working together - because without one I have not answered once, I have set the relationship permanently to retaliation.
Where it goes wrong: reading it as a warrant for grudges. The same rule forbids you to move first and requires you to stop the instant they do; neither half stands without the other.
Lines to keep
None of the leading programs was ever the first to defect.
Hit back once, then come straight back. No ledger.
It never outscored a single opponent and still finished first.
Don't be envious, and don't be the first to defect.
Same situation, other people are asking
He says all the right things and I'm not sure I believe him.He talks well. I've never seen him deliver.She is the strongest candidate and the way she operates worries me.Three months of doing everything right. Is that enough to go on?Can I trust this person · all 9 questions →If this one named what you are going through,
send it to someone who needs it, or keep it somewhere you will find it again.