• What are intelligent agents?

• What are the features of an intelligent agent?

• How to design an intelligent agent?
• An example
• Demo systems

1
 What is an agent?
 An agent is anything that can viewed as perceiving its environment
through sensors and acting upon that environment through effectors.

Sensors: eyes, camera, Examples:

… a human driver
percepts
a robot driver
a driver based on programs
environment
agent
actions

Effectors: hands, motors, …

A generic agent diagram

2
What are intelligent agents
 An agent always requires a certain amount of intelligence to perform its tasks.
Consequently, one refers to intelligent agent.

Intelligent
Processing
Input Output
(percepts) (actions)
Agent as black box

– At the highest level, three major categories of agents can be distinguished: human
agents, hardware agents, and software agents.
– For example, a human travel agent, a robot, an automated taxi

3
Software intelligent agents
 Intelligent software agents are defined as being a software program that can
perform specific tasks for a user and possesses a degree of intelligence that
permits it to perform parts of its tasks autonomous in a useful manner.
 What is intelligence that refers to software agents?
– A software program that think like humans
– A software program that act like humans
– A software program that think rationally
– A software program that act rationally
a very wide variation in the area of intelligence can be envisaged that ranges
from simple agents with limited intelligence through to complex, highly-
intelligent systems.
 The field of AI (artificial intelligence), attempts to understand intelligent
entities.

4
The characteristics of intelligent agents

 Internal characteristics are

– Learning/reasoning:
an agent has the ability to learn from previous experience and to
successively adapt its own behavior to the environment.
– reactivity:
an agent must be capable of reacting appropriately to influences or
information from its environment.
– autonomy:
an agent must have both control over its actions and internal states. The
degree of the agent’s autonomy can be specified. There may need
intervention from the user only for important decisions.
– Goal-oriented:
an agent has well-defined goals and gradually influence its environment
and so achieve its own goals.

5
 continue …

 External characteristics are

– communication:
an agent often requires an interaction with its environment to fulfill its tasks, such as
human, other agents, and arbitrary information sources.
– cooperation:
cooperation of several agents permits faster and better solutions for complex tasks
that exceed the capabilities of a single agent.
– mobility:
an agent may navigate within electronic communication networks.
– Character:
like human, an agent may demonstrate an external behavior with many human
characters as possible.
 What are software agents different from traditional software programs?
– Perception: Software agents are aware of environment changes
– Autonomy: Software agents perform tasks largely autonomously
– Learning: Software agents are accumulating knowledge (learning through
experience)
– Communication: Software agents communicate with users and/or other agents

6
Areas of influence

Decision theory

Artificial Intelligence (AI)

Autonomy
Network communication

Learning reasoning
Mobility

Characteristics
Reactivity
Communication
Character
Cooperation

Psychology

Distributed AI

7
 Applications
 information retrieval and filtering
 news watcher

 friend-making agents

 shopping agents

 scheduling agents

 Web document maintenance agents

……
8
 How to design the agent
program
 agent = architecture + agent program
– The architecture, in general,
 makes the percepts from the sensors available to the program,
 runs the program,
 feeds the program action’s choices to the effectors
– architecture may be
 a plain computer
 a special-purpose hardware
 some software
– The agent program is a function that implements agent mapping from
percepts to actions. It is run on the architecture.

percepts in

agent program
actions out

9
An example: designing an automated
taxi driver
Percepts cameras, speedometer, GPS, sonar

Actions steer, accelerate, brake

Goals Safely to destination
Environment traffic light, other traffic, pedestrians, in Japan

The taxi driver agent and its PAGE description

Four types of agent program:

-Simple reflex agents
-Agents that keep track of the world
-Goal-based agents
-Utility-based agents
10
 Simple reflex agents
It works by finding a rule whose condition matches the current situation
and then doing the action associated with that rule.

function SRAgent(percept)
return action
Sensors static: rules, a set of condition-action rules

Environment
state of the environment
state <- INTERPRET-INPUT(percept)
rule <- RULE-MATCH(state, rules)
action <- RULE-ACTION[rule]

condition-action rules action return action

Effectors
Agent

11
 Agents with awareness of
the world
It works by finding a rule whose condition matches the current situation (as defined by
the percept and the stored internal state) and then doing the action associated with
that rule. (car model evolving)

function SRSAgent(percept) return

action
previous states
static: current world state
Sensors rules, a set of condition-action rules

Environment
evolution of the world
update state of the world
state <- UPDATE-STATE(state, percept)
history of action
rule <- RULE-MATCH(state, rules)
action <- RULE-ACTION[rule]
condition-action rules action
state <- UPDATE-STATE(state, action)
return action

Agent Effectors

12
 Goal-based agents
It works by decision making towards to the final goals.

function Goal-Agent(percept) return

action

Sensors static: current world state

Environment
previous state evolution
of the world history of update state of the world rules, a set of condition-action rules
action

action state <- UPDATE-STATE(state, percept)

condition-action rules
? do { rule <- RULE-MATCH(state, rules)
action <- RULE-ACTION[rule]
goals?
state <- UPDATE-STATE(state, action)
Effectors }while (not goals? <- state)
Agent
return action

13
 Utility-based agents
It works by evaluating each decision in quantity. Utility is a fuction that map a
state onto a real number, which describes the associated quality.

function Utility-Agent(percept) return

action

Sensors static: current world state

Environment
previous state evolution
of the world history of update state of the world rules, a set of condition-action rules
action

action
state <- UPDATE-STATE(state, percept)
condition-action rules
? do { rule <- RULE-MATCH(state, rules)
action <- RULE-ACTION[rule]
utility goals?
state <- UPDATE-STATE(state, action)
Effectors }while ((not goals? <- state)||(not good quality?))
Agent
return action

14
Properties of environment
 Accessible vs. inaccessible
 Deterministic vs. non-deterministic

 Episodic vs. non-episodic

 Static vs. dynamic

 Discrete vs. continuous

15
 Multi-agent system
 Tamura-san’s smart eco-home

16
 Simulated home Data collection
Space log
Space Person Device Time
Agent
Agent Agent living Bob 19:52
living Bob light 19:52
Refrigerator living Bob light, air-conditioner 19:53

Agent is opened!!! data light, air-conditioner,

living Bob 19:55
refrigerator
Agent light, air-conditioner,
living Bob 19:58
refrigerator, TV

Person log
Person Device Time
Hand Hip Sofa logs
Structured Cup Air TV Remote
What is Bob doing? Bob light 19:52
conditioner Bob controller
air-conditioner 19:53
Hand 1 0 0 0 0 Bob 0 refrigerator
1 19:55
Bob TV 19:58
make advice to close Hip - 1 1 0 0 0 0
Recommendation engine Sofa - -
Hand Hip

1
Sofa Cup Air

0
conditioner
TV

0
Remote
controller
0 Device0 log
Device State Detail Time
Hand 1 0 0 0 0 0 1

Cup - - - 1 0 light 0 0
Hip - 1 1 0 0 0 0

make suggestions on - 19:52

Sofa - - 1 0 0 0 0
Cup - - - 1 0 0 0

Air - - - - 1 air-conditioner
0 on 1 25℃ 19:53
Air - - - - 1 0 1
conditioner

refrigerator on open 19:55

TV - - - - - 1 1

conditioner Remote
controller
- - - - - - 1

TV on ch1 19:58
Intelligent Entity Pool TV - Generating
- - activity
- - 1 1
matrix
Remote - - - - - - 1
controller
Activity matrices based data mining
Refrigerator is Hand Hip Sofa Cup Air TV Remote

People’s profile
conditioner controller

opened.
Hand 1 0 0 1 0 0 1
Hip - 1 1 0 0 0 0

Agent Bob is Sofa

Cup
-
-
-
-
1
-
0
1
0
0
0
0
0
0

watching TV. Air

conditioner
TV
-

-
-

-
-

-
-

-
1

-
0

1
1

0
Remote - - - - - - 1
controller

Drinking
Hand Hip Sofa Cup Air TV Remote
conditioner controller
Hand 1 0 0 0 0 0 1
Hip - 1 1 0 0 0 0
Sofa - - 1 0 0 0 0
Cup - - - 1 0 0 0

Analyzed human Extracting Air

conditioner
- - - - 1 0 1

Recommendation people’s activity

TV

Remote
-
-
-
-
-
-
-
-
-
-
1
-
1
1

activities log
controller

engine
Watching TV
Real world applications
 Demo
– Searching for a solution
– Game: computer as player
– Smart garden
– Treasure finding agent
– Maze Robot: find a way out
– Automatic conversation

18
 Work in class/home work
 Run 1-6 demo applets from my course
home page https://cis.k.hosei.ac.jp/~rhuang/
 if you are an indoor garden designer,
how will you design a smart indoor garden?
Please describe your smart indoor garden system.

19