Вы находитесь на странице: 1из 5

BLG521E - Lecture 2

BLG521E Advanced Artificial Intelligence


Lecture 2: Intelligent Agents

Outline
Agents and Environments Rationality PEAS Environment Types Agent Types

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

Agents

Agents
Agents perceive their own actions

Effects?

Percept: the agent perceptual input Percept sequence: the complete history Action choices depend on the pecept sequence Agent function, abstract mathematical description Agent program implements the function
Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007 3 Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007 4

Vacuum-Cleaner World

A Vacuum-Cleaner Agent Function

What is the right way to fill out the table? What makes an agent good, bad or stupid?
Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007 5 Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007 6

BLG521E - Lecture 2

A Vacuum-Cleaner Agent Function

Rationality
A rational agent does the right thing What is rational at any given time depends on:

The performance measure that defines the p The agents prior knowledge of the The actions that the agent can perform The agents percept sequence to date
environment criterion of success

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

Rational Agent
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. With a rational agent, what you ask is what you get
Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007 9

Further Dimensions in Rationality


Information gathering

exploration

Learning Autonomy With or without initial knowledge


Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007 10

Agents <--> Environments


Task environment forms the problem

The Nature of Environments


PEAS for task environments:

Rational agents are the solutions

The task environment affects the appropriate design of the agent

Performance measure Environment Actuators Sensors

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

11

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

12

BLG521E - Lecture 2

Properties of Task Environments


Fully observable vs. Partially observable Deterministic vs. Stochastic

Properties of Task Environments


Discrete vs. Continous Single agent vs. Multiagent

strategic

Episodic vs. Sequential vs Static vs. Dynamic

competitive cooperative p

semidynamic

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

13

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

14

Examples of Task Environments


Task Environment Crossword puzzle Chess with a clock Poker Backgammon Taxi driving Observable Deterministic Episodic Static Discrete Agents

Examples of Task Environments


Task Environment Crossword puzzle Chess with a clock Poker Backgammon Taxi driving Observable Deterministic Episodic Static Discrete Agents

Fully

Deterministic

Sequential

Static

Discrete

Single

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

15

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

16

Examples of Task Environments


Task Environment Crossword puzzle Chess with a clock Poker Backgammon Taxi driving Observable Deterministic Episodic Static Discrete Agents

Examples of Task Environments


Task Environment Crossword puzzle Chess with a clock Poker Backgammon Taxi driving Observable Deterministic Episodic Static Discrete Agents

Fully

Deterministic

Sequential

Static

Discrete

Single

Fully

Deterministic

Sequential

Static

Discrete

Single

Fully

Strategic

Sequential

Semi

Discrete

Multi

Fully

Strategic

Sequential

Semi

Discrete

Multi

Partially

Stochastic

Sequential

Static

Discrete

Multi

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

17

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

18

BLG521E - Lecture 2

Examples of Task Environments


Task Environment Crossword puzzle Chess with a clock Poker Backgammon Taxi driving Observable Deterministic Episodic Static Discrete Agents

Examples of Task Environments


Task Environment Crossword puzzle Chess with a clock Poker Backgammon Taxi driving Observable Deterministic Episodic Static Discrete Agents

Fully

Deterministic

Sequential

Static

Discrete

Single

Fully

Deterministic

Sequential

Static

Discrete

Single

Fully

Strategic

Sequential

Semi

Discrete

Multi

Fully

Strategic

Sequential

Semi

Discrete

Multi

Partially Fully

Stochastic Stochastic

Sequential Sequential

Static Static

Discrete Discrete

Multi Multi

Partially Fully Partially

Stochastic Stochastic Stochastic

Sequential Sequential Sequential

Static Static Dynamic

Discrete Discrete Continuous

Multi Multi Multi

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

19

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

20

The Structure of Agents


The job of AI is to design the agent program Agent architecture Agent = Architecture + Program

Agent Types
Simple reflex agents Model-based reflex agents Goal-based reflex agents Utility-based Utility based agents All these agents can be converted into learning agents

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

21

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

22

Simple Reflex Agents

Reflex Vacuum Agent Program

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

23

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

24

BLG521E - Lecture 2

Simple-Reflex Agent Program

Model-based Reflex Agents

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

25

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

26

Model-based Agent Program

Goal-Based Agents

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

27

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

28

Utility-based Agents

Learning Agents

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

29

Advanced Artificial Intelligence (BLG521E) @ Istanbul Technical University, Sanem Sariel, Fall 2007

30