Week 3: Agents

Agent definition

Agent is anything which:
- Lives in an environment
- Can perceive the environment via its sensors
- Can act upon the environment via its actuators
Agents also have memory of the current and past perceptions of the environment
- Percept – the current state
- Percept Sequence – the complete history of Percepts

Important For Each possible percept sequence, a rational agent should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge the agent has.

The action a rational agent takes depends on:

the preferences of the agent
the agent's information of its environment, which may come from past experiences
the actions, duties and obligations available to the agent
the estimated or actual benefits and the chances of success of the actions.

Important PEAS – Performance Measure, Environment, Actuators, Sensors.

Properties of the Environment

Fully Observable v. Partially Observable
Single v. Multiple Agent
Deterministic v. Stochastic
Episodic v. Sequential
Static v. Dynamic
Discrete v. Continuous
Known v. Unknown

Worst Case:

An environment which is partially observable, multiagent, stochastic, sequential, dynamic, continuous, and unknown.

Types of Agents

Simple reflex agents
- Only current percept matters
- Work only in fully observable environment
- Too many condition-action rules may be required
Model-based reflex agents
- Work better in partially observable environment because of keeping track of the world
- Internal state exists
Goal-based agents
- More flexible than previous agent types
- Goal replacement does not entail rewriting of all the rules
- May be hard to choose between multiple goals
Utility-based agents
- Allow to maximize the “happiness” of the agent
- Allows to find the tradeoff between the conflicting goals
Learning agents