Introduction

Artificial intellegence(Al) is U1e study of how to make computer do things which at

the moment people do better.
T he Al problems
Some of tbe task domains of Al are
• Mundane Tasks
■ Perception
♦ V.ision
♦ Speech
■ Natural Language
♦ Understanding
♦ Generation
♦ Translation
■ Commonsense reasoning

•
■
TRACE KTU
Robot control
Fonnal Taks
■ Games
♦ Chess
♦ Backgammon
♦ Checkers
■ Mathematics
♦ Geometry
♦ Logic
♦ Integral calculus
♦ P roving properties of programs
■ Expert Tasks
♦ Engineering
• Design
 • Fault Finding
• Manufacturing plannjng
■ Scientific Analys is
■ Medical Diagnosis
■ Financial Analysis
Perception of the world around us is crucial to our survival. Animals with much Jess
intelligence tban p eople are capable of more sophisticated visual perception than are
current machines. Perpetual rufficult because they involve analog signals, the signals
are typically very noisy and usually a large number of things must be perceived at
once.
The abiLi ty to use language to communicate a wide variety of ideas is perhaps tbe
most important thing that seperates humans from the other animals. The problem of
understanding spoken language is a p erceptual problem. This problem, usually
referred to as natural language understandimg. In order to understand sentences about
topic, it is necessary to know not only the language itself but also a good deal about
the topic so that unstated assumptions can be recognized.
AI focused on the sort of problem so lving that we do every day when we decide how
to get to work in tbe morning, often called commonsense reasorung.
Game Playing and theorm proving show the property that people who do them well
are considered to be displaying intelligence. Compters could perform well at those
tas ks simply by being fast at exploring a large number of solution paths and then

TRACE KTU
selecting the best one.
Now thousands of programs called expert systems in day to day operation throughout
all areas of industry and goverment. Each o f these system attempts to solve part of a
practical, sign ificant problem that previously requ ired scarce expertise.
Al is a system that acts like human bein gs
For this, a computer would need to possess the fo llowing capabi lities.
• Natural language processing To enable it to communicate successfully m
English.
• Knowledge representation To store wlhat it knows or hears.
• Automated reasoning To use the stored information to answer questions and to
draw new conclusions.
• Machine learning To adapt to new circumstances and to detect and extrapolate
patterns.
• Computer vision To perceive objects.
• Robotics To manipulate objects and move about.
Al is a system th at thin ks like human bei11 gs.
First we must have some way of determining how humans thjnk. We need to get
ins ide the workings of the human minds. Once we have a sufficiently precise theory
of the mind, it becomes possible to express that theory us ing a computer program.
The field of cognitive science brings together computer models from AI and
experimental techniques from psychology to try to construct precise and testable
theories of the workings of the human mind.
AJ is a system th at thinks rationally
For a given set of correct premises, it is possible to yield new conclusions.
For eg.
"Socrates is a man; all men are mortal; therefore, Socrates is mortal. "
These laws of thought were supposed to govern the operation of the mind. This
resulted in a field called logic. A precise notation for the statem en ts about all kinds of
things in the world and about relations among them are developed. Programs exist
that could in principle solve any so lvable problem described in logical notation.
There are 2 main obstacles to this approach.
• First it is not easy to take informal knowledge and state it in the formal terms
required by logical notation.
• Second, there is a big difference between being able to so lve a problem " in
principle" and doing so in practice.
AI is a system that acts rationally
An agent is something that acts. A rational agent is one that acts so as to achieve the
best outcome or, when there is uncertainty, the best expected outcome. We need the
abi lity to represent knowledge and reason with it because this enables us to reach

TRACE KTU
good decisions in a wide variety of situations . We need to be able to generate
comprehensive sentences in natural language because saying those sentences helps us
get by in a complex society. We need learning because having a better idea ofhow the
world works enables us to generate more effective strategies for dealing with it. We
need visual perception to get a better idea of what an action might achieve.
Al Application Areas
The 2 most fundamental concerns ofAl researchers are knowledge representation and
search.
• Knowledge representation
lt addresses the problem of capturing the ful I range of knowledge required for
inte lligent behavior in a formal language,
i.e. One suitable for computer manipulation.
Eg. predicate calculus, LISP, Prolog
• Search
It is a problem solving technique that systematica lly explores a space of problem
states, ie, successive and alternative s tages in the problem solv ing process.

The fo llow ing explains the major application areas of Al.

1. Ga me Playing
 Much of the early research in AI was done using common board g?
4 / 30 ••
••
♦ checkers, ••
♦ chess, and
♦ 15 puzzle.
Board games have certain properties that made them idea l for AI research. Most
games are played using a weU defined set of rules. This makes it easy to generate
the search space. The board configuration used in playing these games can be
easily represented on a computer. As games can be easil y played, testi ng a game
playing program presents no fi nancial or ethical burden.
2, Heuristics
Games can generate extremely large search spaces. So we use powerful
techniques called heuristics to explore the problem space.
A heuristic is a useful but potentially fa llible problem strategy, such as checki ng
to make sure that an unresponsive appliance is plugged in before assuming that it
is broken. Since most of us have some experience with these simple games, we do
not need to find and consult an expert. For these reasons games provide a rich
domain fo r the study of heuristic search.
3. Automated Reasoning and Theorem Proving
Examples for automatic theorem provers are
NewelJ and Simon's Logic Theorist,

TRACE KTU
General Problem Solver (GPS).
Theorem proving research is responsible fo r the development of languages such
as predicate calculus and prolog. The attraction of automated theorem proving lies
in the rigor and generality of logic. A wide variety of problems can be attacked by
representing the problem description as logical axioms and treating problem
instances as theorems to be proved. Reasoning based on formal mathematical
logic is also attractive. Many important prob lems such as design and verification
of logic circuits, verification of the correctness of computer programs and control
of complex systems come in this category.
4. Expert Systems
Here comes ilie importance of domain specific knowledge. A doctor, for example,
is effective at diagnosing illness because she possesses some innate general
problem solving skill; she is effective because she knows a lot about medicine.
A geologist ms effective at discovering mineral deposits. Expert know ledge is a
combination of theoretical understanding of the problem and a co llection of
heuristic problem solvi ng ru les that experience has shown to be effective in the
domain. Expert systems are constructed by obtaining this knowledge from a
human expert and coding it into a form that a computer may apply to similar
problems. To develop such a system, we must obtain knowledge from a human
domain expert. Examples for domain experts are doctor, chemist, geo logist,
engineer etc..
The domain expert provides the necessary knowledge of the problem domain. The
Al specialist is responsible for implementing this knowledge in a program.
 Once such a progra m has been written, it is necessary to refine its expert
ise
tltroug h a process of giving it examp le problems to solve and makin
g any
required chang es or modifications to the program's knowledge.
Dendral is an expert system designed to infer the structure of organi c molec
ules
from thei r chemical formu las and
mass spectrographic information about the chemi cal bonds presen t
in the
molecu les.
Mycin is an expert system which uses expert medical knowl edge to diagnose
and
prescribe treatment for spinal meningitis
and bacterial infections of the blood.
Prospector is an expert system for determ ining the probab le location and
type of
ore deposits based on geolog ical
information about a site.
Internist is an expert system for perfor ming diagnos is in the area of internal
medic ine.
The clipme ter advisor is an expert sys tem for interpreting the results of
oil well
drilling logs.
Xcon is an expert system for config uring VAX compu ters.

5. Natural Language Under standi ng and Seman tic Mode ling

TRACE KTU
One goa l of Al is the creation of progra ms that are capabl e of unders tandin
genera ti ng human language. System s that can use natura l langua ge with
flex ibi lity and generality that charac terize human speech are beyon d
g and
the
curren t
metho dologies. Understanding natural language involves much more than
parsin g
senten ces into their individual parts of speech and lookin g those words
up in a
diction ary. Real unders tanding depend s on extens ive backg round knowledge.
Consid er for examp le, the difficulties in carryin g out a conve rsation
about
baseball with an individual who unders tands Englis h but knows nothin g about
the
rules of the game. This person will not be able to understand the meani ng
of the
sentence.
"With none down in the top of the ninth and the go ahead run at second
, the
manag er called his relief from the bull pen".
Even though hall of the words in the sentence may be individ ually unders
tood ,
this sentence would be difficu lt to even the most intelligent non base bail fan.
The
task of collecting and organi zing this backg round knowl edge in such a way
that it
may be applied to language compr ehensi on forms the major proble
m in
autom ating natural language unders tandin g.
6. Modeling Human Performance
We saw that hu man intelligence is a reference point in consid ering artific
ial
intellig ence. It does not mean that progra ms should pattern thems elves after
the
organi zation of the human mind. Programs that take non human approa
ches to
solving proble ms arc often more succes sful than their human counte rparts.
Still,
 the design of systems that expl icitly model some aspect of human performance
has been a fertile area of research in both AI and psychology.
7. Planning and Robotics
Research in plan ning began as an effort to design robots that could perform their tasks
with some degree of flexibi lity and responsiveness to outside world. Planning
assumes a robot that is capable of perfo rming certain atomic actions.
Planni ng is a difficul t problem because of the s ize of the space of possible sequences
of moves. Even an extremely simple robot is capable of generating a vast number of
potential move sequences. One method that human beings use in planning is
hierarcb_ical probl em decomposition. If we plan a trip to London, we wi ll general ly
treat the problems of arranging a flight, getting to the air port, making airl ine
connections and fi nding ground transportation in London separately. Each of these
may be further decomposed into smaller sub problems. Creating a computer program
that can do the same is a difficult challenge.
A robot that blindl y performs a sequence of actions without responding to changes in
its environment cannot be considered intell igent. Often, a robot will have to formulate
a plan based on the incomplete information and correcr its behavior. A robot may not
have adequate sensors to locate all obstacles in the way of a projected path .
Organizing plans in a fas hion that allows response to changing environmental
conditions is a major problem for plann ing.
8. Machine Learning

TRACE KTU
An expert system m ay perform extensive and costly computa tions to solve a problem.
But if it is given the same or s imilar problem a second time, it us ually does not
remember the solution. lt performs the sam e sequence of computations again.
This is not the behavior o f an intell igent problem so lver. The programs m ust learn on
their own. Learning is a difficult area. But there are several programs that suggest that
it is poss ible. One program is AM, the automated mathematician which was designed
to discover mathematical laws. Initially given the concepts and axioms of set theory,
AM was abl e to induce important mathematical concepts such as cardinality, integer
arithmetic and many of the results o f number theory. AM conjectured new theorems
by modifying its current knowledge base.
Early work includes Wi nston's research on the induction o f structural concepts such
as "arch" fro m a set of examples in the blocks world. The ID3 algorithm bas proved
successful in learning general patterns from examples. Meta dendral learns rules for
interpreting mass spectrograpruc data in organic chemistry from examples of data on
compounds of known structure.Teiresias, an intelligent front end for expert systems,
converts high level advice into new rules for its knowledge base. There are also now
many important biological and sociological models of learning.
9. Neural Nets and Genetic Algorithms
An approach to build intelligent programs is to use models that para llel the structure
of neurons in the human brain. A neuron consists of a cell body that bas a number of
branched protrusions caUed dendrites and a single branch called the axon. Dendrites
receive signals from other neurons. When these combined impulses exceed a certain
threshold, the neuron fires and an impulse or spike passes down the axon.
CS464: ARTIFICIAL INTELLIGENCE

TRACE KTU
Dendrites

Thi s description of the neuron captures featu res that are relevant to neural models of
computation. Each computational unit computes some fu nction of its inputs and
passes the result along to connected units in the network; the fina l results are
produced by the parallel and distributed processing of thjs network of neural
connection and threshold weigh ts.
Languages and Environments for Al
Programm ing environments include knowledge structuring techniques such as object
oriented program1n ing and expert systems fram eworks.
High level languages such as Lisp, and Prolog support modu lar development.
Agents in Artificial Intelligence
An Al system can be defined as the study of the
rational agent and its environment. The agents
sense the environment through sensors and act on
their environment through actuators. An Al agent
can have mental properties such as knowledge,
belief, intention, etc.

What is an Agent?
An agent can be anything that perceiveits
environment through sensors and act upon that
environment through actuators. An Agent runs in

TRACE KTU
the cycle of perceiving, thinking, and acting. An
agent can be:
0 Human-Agent: A human agent has eyes, ears,
and other organs which work for sensors and
hand, legs, vocal tract work for actuators.
o Robotic Agent: A robotic agent can have
cameras, infrared range finder, NLP for sensors
and various motors for actuators.
0 Software Agent: Software agent can have
keystrokes, file contents as sensory input and
act on those inputs and display output on the
screen.
Hence the world around us is full of agents such
as thermostat, cellphone, camera, and even we are
also agents.
Sensor: Sensor is a device which detects the
change in the environment and sends the
information to other electronic devices. An agent
observes its environment through sensors.

Actuators: Actuators are the component of

machines that converts energy into motion. The
actuators are only responsible for moving and
controlling a system. An actuator can be an
TRACE KTU
electric motor, gears, rails, etc.

Effectors: Effectors are the devices which affect

the environment. Effectors can be legs, wheels,
arms, fingers, wings, fins, and display screen.

Percepts

Environment , • Effectors

J✓
Actions
Inte llige nt Agents:
An intelligent agent is an auto nom ous entit y which
act upon an environment using sensors and
actu ators for achieving goals. An intel ligen t agent
may learn from the environment to achieve their
goals. A therm osta t is an example of an intel ligen t
agent.

Following are the main four rules for an Al agen t

o Rule 1: An Al agent mus t have the abili ty to
perceive the environment.
o Rule 2: The observation mus t be used to make
decisions.
o TRACE KTU
Rule 3: Decision should result in an action.
0 Rule 4: The action taken by an Al agent mus t be
a rational action.
Rational Agent:
A rational agent 1s an agent which has clear
preference, mod els uncertainty, and acts in a way
to max imize its performance measure with all
possible actions.

A rational agent is said to perform the right things.

Al is abou t creating rational agents to use for
game theory and decision theo ry for various real-
world scenarios. X
For an Al agent, the rational action is most
important because in Al reinforcement learning
algorithm, for each best possible action, agent
gets the positive reward and for each wrong
action, an agent gets a negative reward.

r/ Note: Rational agents in Al are very sirrnlar to

I
intelligent agents.
TRACE KTU
Rationality:
The rationality of an agent is measured by its
performance measure. Rationality can be judged
on the basis of following points:
o Performance measure which defines the
success criterion.
o Agent prior knowledge of its environment.
o Best possible actions that an agent can
perform.
o The sequence of percepts.
Structure of an Al Agent
The task of Al is to design an agent program which
implements the agent function. The structure of
an intelligent agent 1s a combination of
architecture and agent program. It can be viewed
as:
Agent = Architecture+ Agent program

Following are the main three terms involved in the

structure of an Al agent:

Architecture: Architecture is machinery that an Al

agent executes on.
TRACE KTU
Agent Function: Agent function is used to map a
percept to an action.

r t.P* - A

.
Agent program: Agent program 1s an
implementation of agent function . An agent
program executes on the physical architecture to
produce function f . X
PEAS Representation
PEAS is a type of model on which an Al agent
works upon. When we define an Al agent or
rational agent, then we can group its properties
under PEAS representation model. It is made up of
four words:

o
TRACE KTU
P: Performance measure
o E: Environment
o A: Actuators
o S: Sensors
Here performance measure is the objective for the
success of an agent's behavior.
Let's suppose a self-driving car then PEAS
representation will be:
TRACE KTU
Performance: Safety, time, legal drive, comfort
Environment: Roads, other vehicles, road signs,
pedestrian

Actuators: Steering, accelerator, brake, signal, horn

Sensors: Camera, GPS, speedometer, odometer,
accelerometer, sonar.
The Nature of Environments
Some programs operate in the entirely artificial environment confined to keyboard
input, database, computer file systems and character output on a screen.

In contrast, some software agents (software robots or softbots) exist in rich, unlimited
softbots domains. The simulator has a very detailed, complex environment. The
software agent needs to choose from a long array of actions in real t ime. A softbot
designed to scan t he online preferences of t he customer and show interesting items to
the customer works in t he real as well as an artificial environment.

The most famous artificial environment is t he Turing Test environment, in which one
real and other artificial agents are tested on equal ground. This is a very cha llenging
environment as it is highly diffic ult for a software agent to perform as we ll as a human.

Turing Test
The success of an intelligent behavior of a system can be measured wit h Turing Test.

Two persons and a machine to be evaluated participate in t he test. Out of the two
persons, one plays the role of the tester. Each of them sits in different rooms. The
tester is unaware of who is machine and who is a human. He interrogates the
questions by typing and sending t hem to both intelligences, to which he receives typed
responses.

This test aims at fooling the tester. If the t ester fails to determine machine's response
from the human response, t hen the machine is said to be intelligent.

Properties of Environment

TRACE KTU
The environment has multifold properties -

a Discrete / Continuous - If t here are a limited number of distinct, clearly defined,

st ates of the environment, t he environment is discrete (For example, chess);
otherwise it is continuous (For example, driving).

a Observable / Partially Observable - If it is possible to determine t he complete

state of the environment at each t ime point from the percepts it is observable;
otherwise it is only partially observable.

a Static I Dynamic - If the environment does not change while an agent is acting,
then it is static; otherwise it is dynam ic.

a Single agent/ Multiple agents - The environment may contain other agents which
may be of the same or different kind as that of t he agent.

a Accessible / Inaccessible - If the agent's sensory apparatus can have access to

t he complet e stat e of the environment, t hen the environment is accessible to that
agent.

a Deterministic / Non-deterministic - If the next state of the environme nt is

completely determined by the current state and t he actions of t he agent, then t he
environment is deterministic; otherwise it is non-deterministic.
u Episodic / Non-episodic - In an episodic environment, each episode consists of
the agent perceiving and then acting. The quality of its action depends just on the
episode itself. Subsequent episodes do not depend on the act ions in t he previous
episodes. Episodic environments are much simpler because the agent does not
need to t hink ahead.