Flashcards
Lecture 2: Intelligent Agents
Cards in this deck (31)
- 1ObjectsEncapsulates some state
Communicates via message passing
Has methods, corresponding to procedures that may be performed on state. - 2Differences between Agents and Objects1. Agents are Autonomous (they decide whether or not to respond to request)
2. Agents are flexible (reactive, proactive, social)
3. Agents are active (each have at least one thread) - 3Objects vs. Agent SloganObjects do it for free, Agents do it because they want to
"Agents do it for money..." - 4Agents vs. Expert Systems1. Expert Systems disembodied
2. Don't directly perceive nor interact with environment/other agents.
3. Expert systems need a human middle man. - 5Intentional SystemsWe often explain the world in terms of intentions. "The computer did not want to boot"
Term coined by D.Dennet. - 6Intentional NotionsWe use folk psychology to explain things using intentional notions. "He brought the umbrella not to get wet"
- 7First order, second order... intentional systems.A first order system has beliefs and desires, but no beliefs and desires about those beliefs and desires. Second order has those (and about those of others) and so forth.
- 8When is it appropriate to attribute beliefs, desires, etc. to computer systems?McCarthy:
1. Legitimate when it expresses the same as about a person.
2. Useful when it helps us understand the machine. - 9When it is useful to assign intentional notions to computer systems?1. Never logically required, but can be useful.
2. Most straightforward about simple systems.
3. Most useful about systems with unknown structure.
McCarthy - 10Light Switch ExampleCan describe it as intentional systems, but it gains us nothing. (Yoav Shoham)
- 11Intentional Stance as AbstractionIn software we make more and more abstractions because it makes programming more efficient, the intentional stance is such an abstraction.
- 12Environment EIs a set of instantanous states
- 13Actions ACAgents are assumed to have a set of actions available to them which transform the state of the environment.
- 14A runIs a sequence of interleaved environment states and actions.
- 15R R(AC) R(E)The set of all possible finite runs.
The subset of all runs that end with an action.
The subset of all actions that end with an environment state. - 16A state transformer functionRepresents the effect on the environment of a run.
- 17Environments are:History dependent
&
Non-Deterministic - 18Ended runWhen no more state transformations can happen, we say the system has ended its run.
- 19A formal definition of an environment as a triple.A triple of environment states, a starting state, and a state transformer function.
- 20Agent functionMaps runs to actions. (An agent makes a decision about what action to perform based on the history of the system)
- 21A systemIs an environment and an agent
- 22Reactive AgentOnly makes actions based on the current state of the system, no past.
Like a thermostat. - 23How to give goals to agents?We want to give them goals and let them figure out how to get there.
- 24Utility functionMake the agent maximize a utility function based on states.
But what to maximize? Average, expected, min?
And how to value long-term goals? - 25Utility function - run viewAs opposed to giving values to individual states, we give values to runs to better include long term goals.
- 26Utility ProblemsWhere do the numbers come from?
Hard to formulate tasks in terms of this framework. - 27Bounded optimal agentThe set of agents that maximize expected utility AND are possibly to implement.
- 28Predicate Task SpecificationsGive value of 1 for completed task, 0 otherwise. Based on runs. So a run can either be successful or not.
- 29Most common types of tasks1. Achievement tasks (achieve state X)
2. Maintenance tasks (Avoid state X)
Think of a game against nature. - 30Agent SynthesisA program that will take a task environment and create an agent that succeeds in this environment
- 31Agent Synthesis Soundness and Completeness1. Sound if whenever it returns an agent, the agent succeeds in the environment (not guaranteed to return anything)
2. Complete if it is guaranteed to return an
Information
- Description
- 6 Multi-Agent Systems Flashcards on Lecture 2: Intelligent Agents, created by askemailer on 20/03/2016.
- Created by
- askemailer
- Views
- 4
- Rating