Then we have a special “class” of DefectBots (a name stolen from Orthonormal’s excellent article) consisting of only one single strategy. But this is by far the most popular strategy — I have received it in four copies. Unfortunately for DefectBots, according to the rules they have to be included in the pool only once.
A shame the defectors didn’t cooperate a little better! A few more of us and some noise and we’d have been unstoppable. ;)
A shame the defectors didn’t cooperate a little better!
Meta-defection: Co-operate to drown a tournament in DefectBots.
Meta-meta-defection: ‘co-operate’ to discover how other defectors are planning to drown the tournament and trick as many of them into submitting copies of a single variant on DefectBot.
In the round robin the gains from scoring a D—C 30 times (the beginning play where most strategies went C to start, and I bumped the number to 30 assuming you snuck plenty of trivial variations on DefectBots) is 210 points; tit-for-tat analogues make that many points halfway through one game with another tit-for-tat. Eyeballing the field, there aren’t any places where DefectBot would make a serious profit except Random, so I can’t see even a supermajority of DefectBots winning the round-robin.
In the evolutionary tournament… DefectBots will kill off anyone they’re paired with, but all it takes is two strategies that can cooperate for 90 or so turns and the next generation will be comprised of something like one-third those two and two-thirds DefectBot, which will rapidly favour cooperating bots.
In both cases, the gains from being able to cooperate for ~90 turns is just so huge; a DefectBot needs to defect against a cooperate 50 times to match that score (and no strategy was more defection-tolerant than Random, so they wouldn’t get their 50 D—Cs).
(If the noise was overpowering enough to break even robust tit-for-tats, it wouldn’t be much of a tournament.)
In the evolutionary tournament… DefectBots will kill off anyone they’re paired with, but all it takes is two strategies that can cooperate for 90 or so turns and the next generation will be comprised of something like one-third those two and two-thirds DefectBot, which will rapidly favour cooperating bots.
‘Few’ was intentionally understated. Enough DefectBots just win.
A shame the defectors didn’t cooperate a little better! A few more of us and some noise and we’d have been unstoppable. ;)
Meta-defection: Co-operate to drown a tournament in DefectBots.
Meta-meta-defection: ‘co-operate’ to discover how other defectors are planning to drown the tournament and trick as many of them into submitting copies of a single variant on DefectBot.
It’s straightforward cooperation at the game level. Cooperating with others to force the Nash equilibrium in which you all win.
I don’t think so. Just an intuition, but...
In the round robin the gains from scoring a D—C 30 times (the beginning play where most strategies went C to start, and I bumped the number to 30 assuming you snuck plenty of trivial variations on DefectBots) is 210 points; tit-for-tat analogues make that many points halfway through one game with another tit-for-tat. Eyeballing the field, there aren’t any places where DefectBot would make a serious profit except Random, so I can’t see even a supermajority of DefectBots winning the round-robin.
In the evolutionary tournament… DefectBots will kill off anyone they’re paired with, but all it takes is two strategies that can cooperate for 90 or so turns and the next generation will be comprised of something like one-third those two and two-thirds DefectBot, which will rapidly favour cooperating bots.
In both cases, the gains from being able to cooperate for ~90 turns is just so huge; a DefectBot needs to defect against a cooperate 50 times to match that score (and no strategy was more defection-tolerant than Random, so they wouldn’t get their 50 D—Cs).
(If the noise was overpowering enough to break even robust tit-for-tats, it wouldn’t be much of a tournament.)
‘Few’ was intentionally understated. Enough DefectBots just win.