Perfect / Imperfect Information
Not to be confused with Complete Information game.
In an imperfect-information game, an agent does not fully know the state of the world or the agents act simultaneously. In a game like poker, the agent can only make partial observations.
We use Information Set.
A game of imperfect information is one in which you lack knowledge of any of the following:
- The state of the game (e.g. current market prices).
- The rewards you will receive from various states (i.e. utility and cost functions).
Difference with POMDP?
In partially observable process (specifically, a POMDP), the requirement is that you must not know which state you are in.
Imperfect information games are hard because an optimal strategy for a subgame cannot be determined from that subgame alone.
Our entire policy is common knowledge, but the outcome of random processes are not common knowledge.
https://web.stanford.edu/~jdlevin/Econ%20203/Bayesian.pdf
Examples
Perfect Information Games (each player can see all the pieces on the board at all time)
- Chess
- Tic-tac-toe
- Reversi
- Checkers
- Go
- backgammon (has chance events but still perfect information)
- Monopoly (has chance events but still perfect information)
Imperfect Information Games
- Ridge
- Poker
- Battleship
- Negotiation
- Multi-stage auctions
- Sequential Auctions of multiple items
- Political Campaigns (TV Spending)
- Military (allocating troops; spending on space vs. ocean)
- Medical Treatment
https://www.youtube.com/watch?v=2dX0lwaQRX0&ab_channel=NoamBrown