Given that there are likely to be many rounds, one strategy might be to write down a list of several strategies, randomly try each of them for awhile, and then choose strategies in the future based on which (strategy, my reputation, my opponent’s reputation) combinations were most successful in the past. This assumes that most other strategies will be based largely on reputation rather than on other variables that people track (and also that your own reputation will vary sufficiently to notice what effect it might have, which may be difficult to ensure). I don’t have a good grasp on what the bulk of the opponents will look like; what kind of people do brilliant.org competitions?
Edit: A related idea is to cooperate for awhile, defect for awhile, learn a function (my reputation, my opponent’s reputation) → (my opponent’s move) using your favorite machine learning technique, and then use this function to predict future moves. This strategy seems to at least have the benefit that it should detect the most obvious patterns, e.g. almost everyone else defecting.
That sounds to me like a great tactic. Brilliant.org, as far as I know, is largely frequented by high-IQ kids and teens… so they’ll be smart, but not necessarily skilled at game theory already.
Given that there are likely to be many rounds, one strategy might be to write down a list of several strategies, randomly try each of them for awhile, and then choose strategies in the future based on which (strategy, my reputation, my opponent’s reputation) combinations were most successful in the past. This assumes that most other strategies will be based largely on reputation rather than on other variables that people track (and also that your own reputation will vary sufficiently to notice what effect it might have, which may be difficult to ensure). I don’t have a good grasp on what the bulk of the opponents will look like; what kind of people do brilliant.org competitions?
Edit: A related idea is to cooperate for awhile, defect for awhile, learn a function (my reputation, my opponent’s reputation) → (my opponent’s move) using your favorite machine learning technique, and then use this function to predict future moves. This strategy seems to at least have the benefit that it should detect the most obvious patterns, e.g. almost everyone else defecting.
That sounds to me like a great tactic. Brilliant.org, as far as I know, is largely frequented by high-IQ kids and teens… so they’ll be smart, but not necessarily skilled at game theory already.