Choices by one agent in which background conditions are independent of what other agents are doing. We usually represent these decisions with a decision matrix or decision table. Sometimes called evidential decision theory.
See also: causal decision theory
e.g. Pascal’s Wager
- Rows are possible acts
- Acts are functions that map states to outcomes
- Columns are possible states of the world
- Probabilities are sometimes included for decisions under risk.
- Should not depend on agent action
- States should be
- Mutually exclusive
- Exhaustive: no possibility is left out
- Relevant partition: distinctions that actually have impact on probability or utility of outcomes
- Independence: (optional) each state should be causally and probabilistically independent of the acts
- Dominance principle only holds if independent holds
- Cells are outcomes.
- Art: providing a good formalization of a decision into a table
- providing a justified recommendation based off of the formalization
They can also be represented using decision trees
We can transform decision tables between each other using “reasonable transformations”
- PIR: assign equal probabilities to all states (if we have no knowledge of probabilities, aka DUI)
- Merger: if two states yield identical columns, then we can merge them into one state and add probabilities if we know them