Artificial Intelligence

(IT3160E)

Than Quang Khoat

[email protected]

School of Information and Communication Technology

Hanoi University of Science and Technology

2025
Content:
n Introduction of Artificial Intelligence
n Intelligent agent
q Definition of agent
q Work environment
q Environment types
q Agent types
n Problem solving: Search, Constraint satisfaction
n Logic and reasoning
n Knowledge representation
n Machine learning

Artificial intelligence
2
Definition of Agent
n An agent (tác tử) is anything (e.g., humans, robots,
thermostats, etc.) which can perceive (cảm nhận) its
surrounding environment through sensors and act (hành động)
accordingly to that environment through actuators
n Human agent
q Sensors: eyes, ears and other body parts
q Actuators: hands, legs, mouth and other body parts
n Robot agent
q Sensors: cameras, infrared signal detectors
q Actuators: motors

Artificial intelligence
3
Agent and Environment

n Agent function: maps the history of perception to actions

f: P* ® A
n Agent program: operates based on the actual
architecture of the function f
n Agent = Architecture + Program

Artificial intelligence
4
Example: Vacuum cleaner agent

n Perceptions
q Vacuum cleaner’s location and cleanliness level
q Example: [A, Dirty], [B, Dirty]

n Actions
q The vacuum cleaner moves left, right, or sucks

Artificial intelligence
5
Vacuum cleaner agent
Table of actions of vacuum cleaner agent
Sequence of perceptions Action
[A, Clean] Move right
[A, Dirty] Suck
[B, Clean] Move left
[B, Dirty] Suck
[A, Clean], [A, Clean] Move right
[A, Clean], [A, Dirty] Suck
...

function Reflex-Vacuum-Agent( [location, status]) returns an action

if status = Dirty then return Suck
else if location = A then return Right
else if location = B then return Left

Artificial intelligence
6
Rational agent (1)
n The agent should strive to "do the right thing to do",
based on what it perceives (i.e., knows) and the
actions it can perform
n A right (rational) action is the one that helps the agent
achieve the highest success to the given target
n Performance evaluation: The criteria to evaluate the
level of success in the performance of an agent
q Example: Criteria to evaluate the performance of a vacuum
cleaner agent can be: cleanness level, vacuuming time, power
consumption, noise levels, etc.

Artificial intelligence
7
Rational agent (2)

n Rational agent
q Given a sequence of perceptions,
q A rational agent needs to choose an action that
maximizes that agent's performance evaluation
criteria,
q Based on the information provided by the sequence of
perceptions and the knowledge possessed by that
agent

Artificial intelligence
8
Rational agent (3)
n Rationale ¹ The understanding of everything
q The understanding of everything = Know everything, with infinite
knowledge
q Perceptions may not provide all the relevant information

n Agents can perform actions to change perceptions in the

future, for the purpose of obtaining useful information
(e.g., information gathering, knowledge discovery)
n Autonomous agent is one whose actions are determined
by its own experience (along with the ability to learn and
adapt)

Artificial intelligence
9
Work environment – PEAS (1)

n In order to design an intelligent (i.e., rational) agent, it is

first necessary to define the values of the PEAS
components
n PEAS
q Performance measure: Performance evaluation criteria
q Environment: Surrounding environment
q Actuators: Those parts that allow the agent to do the actions
q Sensors: Those parts that allow the agent to perceive the
surrounding environment

Artificial intelligence
10
Work environment – PEAS (2)
n Example: Design a taxi driving agent
q Performance measure (P): safe, fast, in compliance
with traffic laws, customer satisfaction, optimal profit,
etc.
q Environment (E): roads (streets), other vehicles in
traffic, pedestrians, customers, etc.
q Actuators (A): steering wheel, accelerator, brake,
signal lights, horn, etc.
q Sensors (S): cameras, speedometer, GPS, distance
meter, motor sensors, etc.

Artificial intelligence
11
Work environment – PEAS (3)
n Example: Design a medical diagnostic agent
q Performance measure (P): the patient's health level,

minimizing costs, lawsuits, etc.

q Environment (E): patients, the hospital, medical staffs,

etc.
q Actuators (A): screen to display the questions, tests,
diagnoses, treatments, instructions, etc.
q Sensors (S): keyboard to enter the symptom

information, patient responses to questions, etc.

Artificial intelligence
12
Work environment – PEAS (4)

n Example: Design an object pick-up agent

q Performance measure (P): percentage of the items
placed in the correct boxes (i.e., containers)
q Environment (E): Conveyor on that there are objects,
boxes (i.e., containers)
q Actuators (A): arms and connected hands
q Sensors (S): camera, angle/direction sensors

Artificial intelligence
13
Work environment – PEAS (5)

n Example: Design an interactive English-teaching agent

q Performance measure (P): maximizing students'
English test scores
q Environment (E): a group of students
q Actuators (A): screen to display exercises,
suggestions, assignments’ corrections
q Sensors (S): keyboard

Artificial intelligence
14
Work environment – PEAS (6)

n Example: Design a spam email filtering agent

q Performance measure (P): the number of errors
(e.g., false positives, false negatives)
q Environment (E): email server and clients
q Actuators (A): spam email marker, notification sender
q Sensors (S): the module that receives and analyzes
the emails’ content

Artificial intelligence
15
Environment types (1)
n Fully observable (vs. partially observable)?
q The agent's sensors give it access to the full state of the
environment at a time

n Deterministic (vs. stochastic)?

q The next state of the environment is determined exactly by the
current state and the agent's action (at this current state)
q If an environment is deterministic, except for the actions of other
agents, it is called the strategic environment

Artificial intelligence
16
Environment types (2)
n Episodic (vs. sequential)?
q The agent's experience is divided into atomic "episodes"
q Each episode consists of the agent’s perceiving and then
performing a single action
q The choice of action in each episode depends only on the episode
itself (i.e., not on the other ones)

n Static (vs. dynamic)?

q The environment is unchanged while the agent is deliberating
q The environment is semi-dynamic if the environment itself does
not change with the passage of time but the agent's performance
score does
n Example: Timed game programs

Artificial intelligence
17
Environment types (3)
n Discrete (vs. continuous)?
q A limited number of distinct, clearly defined percepts and actions

n Single agent (vs. multi-agent)?

q An agent operating by itself (i.e., not dependent on/relating to any
others) in an environment

Artificial intelligence
18
Environment types: Examples
Chess Chess Taxi driving
with a clock without a clock
Fully observable? Yes Yes No
Deterministic? Strategic Strategic No
Episodic? No No No
Static? Semi-dyna. Yes No
Discrete? Yes Yes No
Single agent? No No No

n The environment type largely determines the agent

design
n A real-world environment is often: partially observable,
stochastic, sequential, dynamic, continuous, multi-agent

Artificial intelligence
19
Agent types
n Four basic agent types:

q Simple reflex agents

q Model-based reflex agents

q Goal-based agents

q Utility-based agents

Artificial intelligence
20
Simple reflex agents (1)
® Act according to a rule that has its conditions consistent
with the current state of the environment

function SIMPLE-REFLEX-AGENT(percept)
static: rules (a set of rules in format of <conditions> - <action>)

state ¬ INTERPRET-INPUT(percept)
rule ¬ RULE-MATCH(state, rules)
action ¬ RULE-ACTION[rule]
return action

Artificial intelligence
21
Simple reflex agents (2)

Artificial intelligence
22
Model-based reflex agents (1)
n Use an internal model to monitor the current state of the
environment
n Choose the action: The same as for simple reflex agents

function REFLEX-AGENT-WITH-STATE(percept)
static: state (representation of the current state of the environment)
rules (a set of rules in format of <conditions> - <action>)
action (the previous/latest action)

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

rule ¬ RULE-MATCH(state, rules)
action ¬ RULE-ACTION[rule]
return action
Artificial intelligence
23
Model-based reflex agents (2)

Artificial intelligence
24
Goal-based agents (1)
n Know the current state of the environment: Not enough
® Need information of the goal
q The current state of the environment: At an intersection, a taxi can
turn left, turn right, or go straight
q Goal information: The taxi needs to reach the passenger's
destination

n Goal-based agent
q Keep track of the current state of the environment
q Keep a set of goals (to be achieved)
q Choose the action that allows to (finally) achieve the goals

Artificial intelligence
25
Goal-based agents (2)

Artificial intelligence
26
Utility-based agents (1)
n In many environments, the information of the goals is not
sufficient to assess the effectiveness of actions
q There are several (or many) sequences of actions to allow a taxi
to reach its destination (i.e., achieve the goal)
q But: Which sequence of actions is faster, safer, more reliable,
lower cost?

n Need an assessment of the utility (i.e., benefit) to the

agent
n Utility function
q Mapping the sequence of environmental states to a real number
(i.e., the level of utility/benefit to the agent)

Artificial intelligence
27
Utility-based agents (2)

Artificial intelligence
28
Learning agents (1)
n The ability to learn allows the agent to improve its
performance

n The 4 elements make up a learning agent:

q Performance: undertakes the choice of action
q Critic: evaluates the performance
q Learning: helps to improve the performance - based on critics, to
change (improve) the Performance element
q Problem generator: helps to generate new experiences

Artificial intelligence
29
Learning agents (2)

Artificial intelligence
30
Multi-agent (1)
n Work environment: Collaborative or Competitive?
n In many practical problems, the work environment is
always changing ® the agent needs to get updated
n Need a model to represent the plans of other agents
n Collaborative agents
q Share goals or plans together
q Example: Planning (for group activities) in a double's tennis game
q Collaboration mechanisms: Separate and distribute tasks for each
agent

Artificial intelligence
31
Multi-agent (2)
n Competitive agents
q Example: Chess game
q Each agent must be aware of the existence (and activity) of the
other agents
q Each agent computes (i.e., predicts) the plans of (some) other
actors
q Each agent computes (i.e., predicts) the effect of the others' plans
on its own
q Each agent determines the optimal action against this predicted
effect

Artificial intelligence
32
Agent: Summary
n An agent interacts with the environment through its sensors
and actuators
n A rational agent maximizes its performance
n The agent function determines the actions an agent performs
in situations
n Agent programs implement (i.e., execute) the agent functions
n PEAS descriptions define the work environment
n The environments are classified according to the criteria: Fully
observable? Deterministic? Episodic? Statistic? Discrete?
Single agent?
n Basic agent types: Simple reflex, Model-based reflex, Goal-
based, Utility-based
Artificial intelligence
33