1.
Intelligent Agents
(more concrete study)
R&N: Chap. 2

2.
Outline
ƒ Agents and environments
ƒ Rationality
ƒ PEAS (Performance measure, Environment,
Actuators, Sensors)
ƒ Environment types
ƒ Agent types
2

3.
Agents
ƒ An agent is anything that can be viewed as
perceiving its environment through sensors
and acting upon that environment through
actuators
ƒ Human agent: eyes, ears, and other organs for
sensors; hands, legs, mouth, and other body
parts for actuators
ƒ Robotic agent: cameras and infrared range
finders for sensors; various motors for
actuators
3

4.
Agents and environments
ƒ The agent function maps from percept
histories to actions: [f: P* Æ A]
ƒ The agent program runs on the physical
architecture to produce f
ƒ agent = architecture + program
4

5.
Vacuum-cleaner world
ƒ Percepts: location and contents, e.g.,
[A, Dirty]
ƒ Actions: Left, Right, Suck, NoOp
5

6.
A vacuum-cleaner agent
6
ƒ What is the right function?
ƒ Can it be implemented in a small agent program?
ƒ Partial tabulation of a simple agent function:
Percept sequence Action
[A;Clean] Right
[A;Dirty] Suck
[B;Clean] Left
[B;Dirty] Suck
[A;Clean], [A;Clean] Right
[A;Clean], [A;Dirty] Suck
…. ….

7.
Rational agents
ƒ An agent should strive to "do the right thing", based
on what it can perceive and the actions it can
perform. The right action is the one that will cause
the agent to be most successful
ƒ Performance measure: An objective criterion for
success of an agent's behavior
ƒ E.g., the performance measure of the vacuum-cleaner
agent example could be amount of dirt cleaned up,
amount of time taken, amount of electricity
consumed, amount of noise generated, etc.
7

8.
Rational agents
ƒ 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.
8

9.
Rational agents
ƒ Rationality is distinct from omniscience (all-knowing
with infinite knowledge)
ƒ Does rationality mean perfection?
ƒ Agents can perform actions in order to modify future
percepts so as to obtain useful information
(information gathering, exploration)
ƒ An agent is autonomous if its behavior is determined
by its own experience (with ability to learn and adapt)
9

10.
Task Environment
ƒ PEAS: Performance measure, Environment, Actuators,
and Sensors
ƒ Must first specify the task environment for
intelligent agent design
ƒ Consider a more complex example: the task of
designing an automated taxi driver:
ƒ Performance measure
ƒ Environment
ƒ Actuators
ƒ Sensors
10

11.
PEAS
ƒ Must first specify the setting for intelligent agent
design
ƒ Performance measure: Safe, fast, legal,
comfortable trip, maximize profits
ƒ Environment: Roads, other traffic, pedestrians,
customers, stray animals, road works, …
ƒ Actuators: Steering wheel, accelerator, brake,
signal, horn, voice synthesizer or display screen, …
ƒ Sensors: Cameras, sonar, speedometer, GPS,
odometer, engine sensors, keyboard or microphone
11

12.
PEAS
Agent: Internet shopping agent
ƒ Performance measure?? price, quality,
appropriateness, efficiency
ƒ Environment?? current and future WWW sites,
vendors, shippers
ƒ Actuators?? display to user, follow URL, fill in
form
ƒ Sensors?? HTML pages (text, graphics, scripts)
12

13.
PEAS
Agent: Internet shopping agent
ƒ Performance measure?? price, quality,
appropriateness, efficiency
ƒ Environment?? current and future WWW sites,
vendors, shippers
ƒ Actuators?? display to user, follow URL, fill in
form
ƒ Sensors?? HTML pages (text, graphics, scripts)
13

14.
Types of Task Environments
ƒ Fully observable (vs. partially observable): An agent's
sensors give it access to the complete state of the
environment at each point in time.
ƒ Deterministic (vs. stochastic): The next state of the
environment is completely determined by the current
state and the action executed by the agent. (If the
environment is deterministic except for the actions
of other agents, then the environment is strategic)
ƒ Episodic (vs. sequential): The agent's experience is
divided into atomic "episodes" (each episode consists
of the agent perceiving and then performing a single
action), and the choice of action in each episode
depends only on the episode itself.
17

15.
Environment types
ƒ Static (vs. dynamic): The environment is
unchanged while an agent is deliberating. (The
environment is semidynamic if the
environment itself does not change with the
passage of time but the agent's performance
score does)
ƒ Discrete (vs. continuous): A limited number of
distinct, clearly defined percepts and actions.
ƒ Single agent (vs. multiagent): An agent
operating by itself in an environment.
18

16.
Environment types: Examples
19
Crossword
puzzle
Chess with
clock
Internet
shopping
Taxi Driver
Observable Yes Yes! No No
Deterministic Yes Strategic Partly No
Episodic No No No No
Static Yes Semi Semi No
Discrete Yes Yes Yes No
Single-agent Yes No Yes!! No
ƒ Are the answers exact?
ƒ The environment type largely determines the agent design
ƒ The real world is (of course) partially observable, stochastic,
sequential, dynamic, continuous, multi-agent

17.
Agent functions and programs
ƒ An agent is completely specified by the agent
function mapping percept sequences to actions
ƒ One agent function (or a small equivalence
class) is rational
ƒ The aim is to design the agent program that
implement the rational agent function concisely
ƒ This program shall run on a computing device
with sensors and actuators: architecture
20

18.
A vacuum-cleaner agent
21

19.
Table-driven to agent construction
ƒ Is it practical?
ƒ Consider the automated taxi
ƒ 30 frames/sec; 640x480 pixels with 24 bits for
color
ƒ The size of the lookup table > 10250B
entries!!!
ƒ Drawbacks:
ƒ Huge table
ƒ Take a long time to build the table
ƒ No autonomy
ƒ Even with learning, need a long time to learn the
table entries
22

20.
Agent types
ƒ Four basic types in order of increasing
generality:
ƒ Simple reflex agents
ƒ Model-based reflex agents
ƒ Goal-based agents
ƒ Utility-based agents
All these can be turned into learning agents
23

21.
Simple reflex agents
24
function Reflex-Vacuum-Agent( [location,status]) re-
turns an action
if status = Dirty then return Suck
else if location = A then return Right
else if location = B then return Left

22.
Model-based reflex agents
25
It keeps track of the current state of the world
using an internal model

23.
Goal-based agents
26
It keeps track of the current state of the world
and a set of goals to achieve.

24.
Utility-based agents
27

25.
Learning agents
28

26.
Summary
ƒ Agents interact with environments through actuators and sensors
ƒ The agent function describes what the agent does in all
circumstances
ƒ The performance measure evaluates the environment sequence
ƒ A perfectly rational agent maximizes expected performance
ƒ Agent programs implement (some) agent functions
ƒ PEAS descriptions define task environments
ƒ Environments are categorized along several dimensions:
ƒ observable? deterministic? episodic? static? discrete? single-agent?
ƒ Several basic agent architectures exist:
ƒ reflex, model-based, goal-based, utility-based
29