motivation: we want to model and design agents that act rationally. A rational agent is one that can come up with a PLAN OF ACTIONS consistent with its BELIEFS in order to achieve the GOALS it has been designed for.
"If I have a goal of staying dry, and I believe it is raining, then it is rational of me to take an umbrella when I leave my house" (Wooldridge in "Reasoning about Rational Agents")
Note that for the agent to behave rationally, it does not matter whether it is indeed raining or not.
See also Dennett's intentional stance and the Mc Carthy quote in AgentDefinition.
3 components to BeliefDesireIntention:
- a philosophical theory of practical reasoning (Bratman)
- a logical foundation (Shoham, Cohen & Levesque, Wooldridge...)
- an implementation framework (Shoham, Rao & Georgeff)
mental constructs: Practical reasoning vs theoretical reasoning
- Beliefs are a set of statements that an agent has about the world and itself.
- Desires are a set of beliefs that an agent would like to reach
- Intentions are a subset of desires that the agent selects, through a deliberation process, and commits to achieving.
properties:
- consistency (between intentions, between intentions and beliefs, between beliefs...)
- persistence
- Intention is choice with commitment. (Cohen & Levesque) [7]
- beliefs persist by default, and their absence as well, until the belief is learnt (Shoham) [4]
- good faith: only commit to what you believe you are capable of
- introspection
- intentions drive means-ends reasoning
formal background: Epistemic logic, based on ModalLogic.
implementation: BdiArchitecture
issues:
- mapping of modal logic to the proposed architecture?
- omniscience and other theorem proving issues
- resulting performance
- finding a good balance between overcommitment to an intention and being overly cautious
refs:
Stanford's encyclopedia of philosophy: [3]
Wooldridge's slides available here [5]
Andre Vellino's lecture slides available here [2].
(last edited November 16, 2023)
Find Page by browsing or searching