Perfect rationality
Assumes the agent knows everything. Always take the action that maximizes the utility. Might not be possible in all scenario.
Bounded rationality
Assumes the agent has limited information. Uses approximate methods to perform tasks. Proposed by Herbert Simon in 1958.
Rational action
Given the percept sequence, the action that maximizes the expected value of the performance measure. It is the best action, but not the optimal action.
Omniscience
An agent is omniscient if it knows the actual outcomes of its actions. An ideal case, and not a practical goal. Rational agent and omniscient agent are independent.
It represents a hypothetical agent with complete knowledge of:
- The environment’s true state
- All possible action outcomes
- The exact utility of each outcome
Real AI systems cannot achieve omniscience due to:
- Inherent uncertainty in complex environments
- Computational limitations
- Incomplete information