View
244
Download
2
Category
Preview:
Citation preview
Chapter 2
Intelligent Agents
1
Chapter 2 Intelligent Agents
What is an agent ?An agent is anything that
perceiving its environment through sensors( أجهزة(االستشعار
acting upon that environment through actuators(المحركات)
Example: Human is an agentA robot is also an agent with cameras and motorsA thermostat detecting room temperature.
2
Intelligent Agents
3
Diagram of an agent
What AI should fill 4
Simple Terms
PerceptAgent’s perceptual inputs at any given instant
Percept sequenceComplete history of everything that the agent
has ever perceived.
5
Agent function & programAgent’s behavior (سلوك ) is mathematically described byAgent functionA function mapping any given percept
sequence to an action
Practically it is described by An agent programThe real implementation
6
Vacuum-cleaner world
Perception: Clean or Dirty? where it is in?
Actions: Move left, Move right, suck, do nothing
7
Vacuum-cleaner world
8
Concept of Rationality
Rational agentOne that does the right thing= every entry in the table for the agent
function is correct (rational).
What is correct?The actions that cause the agent
to be most successfulSo we need ways to measure success.
9
Performance measurePerformance measureAn objective function that determines
How the agent does successfullyE.g., 90% or 30% ?
An agent, based on its percepts action sequence if desirable, it is said to be performing well.No universal performance measure for all
agents10
Performance measure
A general rule:Design performance measures according to
What one actually wants in the environmentRather than how one thinks the agent should
behave ( بدال من أن يفكر المرء كيف ينبغي أن يتصرف(الوكيل
E.g., in vacuum-cleaner worldWe want the floor clean, no matter how the
agent behaveWe don’t restrict how the agent behaves
11
Rationality (عقالنية)
What is rational at any given time depends on four things:The performance measure defining the criterion
of successThe agent’s prior knowledge of the environmentThe actions that the agent can performThe agents’s percept (المدرك) sequence up to
now
12
Rational agent For each possible percept sequence, an rational agent should select
an action expected to maximize its performance measure
given the evidence provided by the percept sequence whatever built-in knowledge the agent has
E.g.(latin exempli gratia : for example), an examMaximize marks, based on the questions on the paper & your knowledge13
Example of a rational agent
Performance measureAwards one point for each clean square
at each time step, over 10000 time steps
Prior knowledge about the environmentThe geography of the environmentOnly two squaresThe effect of the actions
14
Actions that can performLeft, Right, Suck and NoOp
Percept sequencesWhere is the agent?Whether the location contains dirt?
Under this, circumstance(ظرف) the agent is rational.
Example of a rational agent
15
An omniscient agentKnows the actual outcome of its actions in
advanceNo other possible outcomesHowever, impossible in real world
An example
crossing a street but died of the fallen cargo(حمولة) door from 33,000ft(33 000 feet =10.0584 kilometers) irrational?
Omniscience( معرفة غير(محدودة
16
Based on the circumstance, it is rational.
As rationality maximizesExpected performance
Perfection maximizesActual performance
Hence rational agents are not omniscient.
Omniscience
17
Learning
Does a rational agent depend on only current percept?No, the past percept sequence should also
be usedThis is called learningAfter experiencing an episode, the agent
should adjust its behaviors to perform better for the same job next time.
18
Autonomy (الحكم الذاتي)
If an agent just relies on Its own prior knowledge of its designerRather than its own percepts then the agent lacks(تفتقر) autonomy
E.g., a clockNo input (percepts)Run only but its own algorithm (prior
knowledge)No learning, no experience, etc.
19
Sometimes, the environment may not be the real worldE.g., flight simulator, video games, InternetThey are all artificial but very complex
environmentsThose agents working in these environments
are calledSoftware agent (softbots : Software robots)Because all parts of the agent are software
Software Agents
20
Nature of environmentsTask environments are the problemsproblemsWhile the rational agents are the solutionssolutions
Specifying the task environmentPEAS description as fully as possible
PerformanceEnvironmentActuatorsSensors
Use automated taxi driver as an example
21
Nature of environments
Performance measureHow can we judge the automated driver?Which factors are considered?
getting to the correct destinationminimizing fuel consumptionminimizing the trip time and/or costminimizing the violations of traffic lawsmaximizing the safety and comfort, etc.
22
EnvironmentA taxi must deal with a variety of roadsTraffic lights, other vehicles, pedestrians,
stray animals, road works, police cars, etc. Interact with the customer
Nature of environments
23
Actuators (for outputs)Control over the accelerator, steering(توجيه),
gear shifting (نقل العتاد) and braking(فرملة)A display to communicate with the customers
Sensors (for inputs)Detect other vehicles, road(طريق) situationsGPS (Global Positioning System) to know
where the taxi is Many more devices are necessary
Nature of environments
24
A sketch of automated taxi driver
Nature of environments
25
Properties of EnvironmentsFully observable vs. Partially observableagent’s sensors access to the complete
state of the worldThe environment is effectively and fully
observable if the sensors detect all aspectsThat are relevant to the choice of action
Partially observableA local dirt sensor of the cleaner cannot tellWhether other squares are clean or not
26
Deterministic (القطعية )vs. stochastic next state of the environment
Completely determined by the current stateand the agent’s actions
then the environment is deterministic Strategic: deterministic except the actions
from other agentsCleaner and taxi driver
Stochastic because of some unobservable aspects noise or unknown
Properties of Environments
27
Episodic (عرضي) vs. sequential(تسلسلي)An episode = agent’s single pair of
perception(اإلدراك) & action The quality of the agent’s action does not
depend on other episodes Every episode is independent of each other
Episodic environment is simplerThe agent does not need to think ahead(قدما)
SequentialCurrent action may affect all future decisions
Properties of Environments
28
Static vs. dynamic A dynamic environment is always changing
over time E.g., the number of people in the street
While static environment E.g., the destination
Semidynamicenvironment is not changed over timebut the performance score does
Properties of Environments
29
Discrete vs. continuous If there are a limited number of distinct,
clearly defined percepts and actions, the environment is discrete
E.g., the gear phase of a car: 1, 2, 3, 4, 5, RContinuous: 0~100, in between there may
exist real numbers
Properties of Environments
30
Single agent VS. multiagentPlaying a crossword puzzle – single agentChess playing (لعب الشطرنج) – two agentsCompetitive multiagent environment
Chess playingCooperative multiagent environment
Automated taxi driverAvoiding collision(تجنب االصطدام)
Properties of Environments
31
Examples of environments
32
Structure of agents
Agent = architecture + programArchitecture = some sort of computing
device (sensors + actuators) (Agent) Program = some function that
implements the agent mapping = “?”Agent Program = Job of AI
33
Agent programs
Input for Agent ProgramOnly the current percept
Input for Agent FunctionThe entire percept sequenceThe agent must remember all of them
Implement the agent program asA look up table (agent function)The next figure but it is no good
34
Agent programs
Skeleton (هيكل عظمي )design of an agent program
Append = ألحق ; look up = بحث عن35
Agent Programs
P = the set of possible percepts
T= lifetime of the agentThe total number of percepts it receives
Size of the look up table
Consider playing chessP =10, T=150Will require a table of at least 10150 entries
T
t
tP
1
36
Agent programsDespite of huge size, look up table does what we want.
The key challenge of AIFind out how to write programs that, to the
extent possible, produce rational behaviorFrom a small amount of codeRather than a large amount of table entries
E.g., a five-line program of Newton’s MethodV.s. huge tables of square roots, sine(جيب),
cosine, …37
Types of agent programs
Four typesSimple reflex agentsModel-based reflex agentsGoal-based agentsUtility-based agents
38
Simple reflex agentsIt uses just condition-action rulesThe rules are like the form “if … then …” efficient but have narrow range of applicabilityBecause knowledge sometimes cannot be
stated explicitly Work only
if the environment is fully observable
39
Simple reflex agents
40
Simple reflex agents (2)
41
Model-based Reflex Agents For the world that is partially observable the agent has to keep track of an internal state
That depends on the percept historyReflecting some of the unobserved aspectsE.g., driving a car and changing lane
Requiring two types of knowledgeHow the world evolves independently of the
agentHow the agent’s actions affect the world
42
Model-based Reflex Agents
The agent is with memory43
Model-based Reflex Agents
44
Goal-based agents
Current state of the environment is always not enough
The goal is another issue to achieve Judgment of rationality / correctness
Actions chosen goals, based on the current state the current percept
45
Goal-based agents
ConclusionGoal-based agents are less efficientbut more flexible
Agent Different goals different tasksSearch and planning
two other sub-fields in AI to find out the action sequences to achieve its goal
46
Goal-based agents
47
Utility-based agents
Goals alone are not enough to generate high-qualityhigh-quality behavior E.g. meals in Canteen( وجبات الطعام في
? good or not ,(مقصف
Many action sequences the goals some are better and some worse If goal means success, then utility means the degree of success
(how successful it is) 48
Utility-based agents (4)
49
Utility-based agents
it is said state A has higher utility If state A is more preferred than others
Utility is therefore a function that maps a state onto a real number the degree of success
50
Utility-based agents (3)
Utility has several advantages: When there are conflicting goals,
Only some of the goals but not all can be achieved
utility describes the appropriate trade-off(مفاضله)
When there are several goals None of them are achieved certainlyutility provides a way for the decision-making
51
Learning AgentsAfter an agent is programmed, can it work immediately?No, it still need teaching
In AI,Once an agent is doneWe teach it by giving it a set of examplesTest it by using another set of examples
We then say the agent learnsA learning agent
52
Learning Agents
Four conceptual componentsLearning element
Making improvementPerformance element
Selecting external actionsCritic
Feedback from user or examples, good or not?Problem generator
New derived situation, try new solution or method53
Learning Agents
54
Recommended