Intelligent Agents

Software agents O Monday: O Overview video (Introduction to software agents) O Agents and environments O Rationality O Wednesday: O PEAS (Performance measure, Environment, Actuators, Sensors) O Example of simple and not-so-simple software agents O Friday: O Environment types O Agent types O Group discussion

Very short software agents overview O O What is an agent?: A computer program that is O Embodied (i.e local) O Situated (ie. has knowledge of / responds to its situation / environment) O Autonomous (i.e decides what action to take based on its own situation) O Cooperating (i.e can communicate with other agents to achieve tasks)

Agent and environments

O Agents include humans, robots, softbots, thermostats, etc. O The agent function maps from percept histories to actions: f : P -> A O The agent program runs on the physical architecture to produce f

Vacuum-cleaner world O Percepts: location and contents, e.g., [A;Dirty] O Actions: Left, Right, Suck, NoOp

A vacuum-cleaner agent

Pseudocode What is the right function? Can it be implemented in a small agent program?

Rationality O Fixed performance measure evaluates the environment sequence O one point per square cleaned up in time T? O one point per clean square per time step, minus one per move? O penalize for > k dirty squares? O A rational agent chooses whichever action maximizes the expected value of the performance measure given the percept sequence to date O Rational is different from omniscient O percepts may not supply all relevant information O Rational is different from clairvoyant O action outcomes may not be as expected O Rational is different from successful O Rational exploration, learning, autonomy

PEAS O To design a rational agent, we must specify the task environment. Example: consider, e.g., the task of designing an automated taxi: O Performance measure: O Environment: O Actuators: O Sensors:

Another example O Internet shopping agents: O Performance measure?? O Environment?? O Actuators?? O Sensors??

Environment types O The environment type largely determines the agent design O The real world is (of course) partially observable, stochastic, sequential, dynamic, continuous, multi-agent

Agent types O Four basic types in order of increasing generality: O Simple reflex agents O Reflex agents with state O Goal-based agents O Utility-based agents

Simple reflex agents

Reflex agents with state

Goal-based agents

Utility-based agents

Learning agents

Example O Simple example O Demonstration of software agent on Linux O Complex example:: O mmonsense-aria.html mmonsense-aria.html

Summary O Agents interact with environments through actuators and sensors O The agent function describes what the agent does in all circumstances O The performance measure evaluates the environment sequence O A perfectly rational agent maximizes expected performance O Agent programs implement (some) agent functions O PEAS descriptions define task environments O Environments are categorized along several dimensions: O observable? deterministic? episodic? static? discrete? single- agent? O Several basic agent architectures exist: O Reflex, reflex with state, goal-based, utility-based