Past and Upcoming Tournaments
|2017-09-08||Tournament One||Game One||Bertrand-Edgeworth|
|Pending||Tournament Two||Game One||Bertrand-Edgeworth|
These are general issues that are applicable to all the games that we play.
The objective of firms is profit maximization. We cannot exactly mimic that objective function in a game. Indeed, if there are only two players, then winning a game means doing better than the opponent. This is a different objective function than profit maximization. According to my back-of-the-envelope checks, Bertrand-Edgeworth paradox still holds but the mixed strategy solution will be different. This issue can be alleviated with a round robin tournament. If there are many players and the winner is the player who scores most total profits from playing tet-a-tet against all his opponents, then the objective function should be closer to profit maximization than to maximizing the probability of scoring better than the opponent in each single game. This questions is in itself an open research question.
We can either play many isolated rounds where the programs are not allowed to learn from previous rounds, or we could play many consecutive rounds where the programs can learn from past rounds. Both games are of interest, and the current idea is to play spot games (no learning), short games (30 rounds), and long games (1000 round).
I do not think there is an easy way to prevent collusion between programs. In real life, there are antitrust authorities, but there is no unambiguous way to translate the antitrust laws into restrictions on programs. Also, seeing if collusion emergence is of interest in its own right. Anyways, given that our first game is most simple and thus most susceptible to collusion, additional restrictions apply to prevent it (see the description of the game).