Paper Presentation on Artificial Intelligence 1

Chinmayi V., Navya K. Computer Science Engineering, Bhoj Reddy Engineering College for Women. Email ID:[email protected] Email ID: chinmayi.meeragmail.com

ARTIFICIAL INTELLIGENCE Abstract: Artificial intelligence (AI) is the intelligence of machines and the branch of computer science that aims to create it. Textbooks define the field as "the study and design of intelligent agents”. The field was founded on the claim that a central property of humans, intelligence—the sapience of Homo sapiens—can be so precisely described that it can be simulated by a machine. This raises philosophical issues about the nature of the mind and limits of scientific hubris, issues which have been addressed by myth, fiction and philosophy since antiquity. Artificial intelligence has been the subject of optimism, but has also suffered setbacks and, today, has become an essential part of the technology industry, providing the heavy lifting for many of the most difficult problems in computer science. Mechanical or "formal" reasoning has been developed by philosophers and mathematicians since antiquity. The study of logic led directly to the invention of the programmable digital electronic computer, based on the work of mathematician Alan Turing and others. The general problem of simulating (or creating) intelligence has been broken down into a number of specific sub-problems. These consist of particular traits or capabilities that researchers would like an intelligent system to display. The traits have received the most attention, like Deduction, Reasoning, Problem solving, learning, motion capturing and manipulation, etc. Artificial intelligence has been used in a wide range of fields including medical diagnosis, stock trading, robot control, law, scientific discovery and toys. However, many AI applications are not perceived as AI: "A lot of cutting edge AI has filtered into general

applications, often without being called AI because once something becomes useful enough and common enough it's not labeled AI anymore. animated statues were seen in Egypt and

Introduction: Artificial intelligence (AI) is the

Greece and humanoid automatons were

intelligence of machines and the branch

built by Yan Shi, Hero of Alexandria,

of computer science that aims to create

Al-Jazari and Wolfgang von Kempelen.

it. Textbooks define the field as "the study and design of intelligent agents”. The field was founded on the claim that a

central

property

of

humans,

intelligence—the sapience of Homo sapiens—can be so precisely described that it can be simulated by a machine. This raises philosophical issues about the nature of the mind and limits of scientific hubris, issues which have been addressed

by

myth,

fiction

and

philosophy since antiquity. Artificial intelligence has been the subject of optimism, but has also suffered setbacks and, today, has become an essential part of the technology industry, providing the heavy lifting for many of the most difficult problems in computer science. Thinking machines and artificial beings appear in Greek myths, such as Talos of Crete, the golden robots of Hephaestus and

Pygmalion's

Galatea.

Human

likenesses believed to have intelligence were built in every major civilization:

Mechanical or "formal" reasoning has been developed by philosophers and mathematicians since antiquity. The study of logic led directly to the invention of the programmable digital electronic computer, based on the work of mathematician Alan Turing and others. Turing's theory of computation suggested that a machine, by shuffling symbols as simple as "0" and "1", could simulate

any

conceivable

act

of

mathematical deduction. This, along with recent discoveries in neurology, information

theory and

cybernetics,

inspired a small group of researchers to begin

to

seriously

consider

the

possibility of building an electronic brain. The field of AI research was founded at a

conference

on

the

campus

of

Dartmouth College in the summer of 1956. The attendees, including John McCarthy,

Marvin

Minsky,

Allen

Newell and Herbert Simon, became the

leaders of AI research for many decades.

subproblems, the creation of new ties

They and their students wrote programs

between AI and other fields working on

that were, to most people, simply

similar problems, and above all a new

astonishing: computers were solving

commitment by researchers to solid

word problems in algebra, proving

mathematical methods and rigorous

logical theorems and speaking English.

scientific standards.

By the middle of the 1960s, research in the U.S. was heavily funded by the

Problems

Department of Defense and laboratories had been established around the world.

The general problem of simulating (or

AI's

profoundly

creating) intelligence has been broken

optimistic about the future of the new

down into a number of specific sub-

field: Herbert Simon predicted that

problems. These consist of particular

"machines

within

traits or capabilities that researchers

twenty years, of doing any work a man

would like an intelligent system to

can do" and Marvin Minsky agreed,

display. The traits described below have

writing that "within a generation ... the

received the most attention.

founders

were

will be capable,

problem

of

intelligence'

creating will

'artificial

substantially

be

Deduction, reasoning, problem solving

solved". In the 1990s and early 21st century, AI achieved its greatest successes, albeit somewhat behind the scenes. Artificial intelligence is used for logistics, data mining, medical diagnosis and many other areas throughout the technology industry. The success was due to several factors:

the

incredible

power

of

computers today (see Moore's law), a greater emphasis on solving specific

Early

AI

researchers

developed

algorithms that imitated the step-by-step reasoning

that

human

were

often

assumed to use when they solve puzzles, play board games or make logical deductions. By the late 1980s and '90s, AI research had also developed highly successful methods for dealing with uncertain or incomplete information, employing concepts from probability and economics.

For difficult problems, most of these algorithms

can

computational

require

categories and relations between objects;

experience a "combinatorial explosion":

situations, events, states and time; causes

the amount of memory or computer time

and effects; knowledge about knowledge

required becomes astronomical when the

(what we know about what other people

problem goes beyond a certain size. The

know); and many other, less well

search

researched

more

—

to represent are: objects, properties,

most

for

resources

enormous

world. Among the things that AI needs

efficient

problem

domains.

A

complete

solving algorithms is a high priority for

representation of "what exists" is an

AI research.

ontology

(borrowing

a word

from

traditional philosophy), of which the Human beings solve most of their

most general are called upper ontologies.

problems using fast, intuitive judgments rather than the conscious, step-by-step deduction that early AI research was able to model. AI has made some progress at imitating this kind of "subsymbolic" problem solving: embodied agent

approaches

emphasize

the

importance of sensorimotor skills to higher reasoning; neural net research attempts to simulate the structures inside human and animal brains that give rise

Among the most difficult problems in knowledge representation are: 1. Default

reasoning

and

the

qualification problem 2. The breadth of commonsense knowledge 3. The subsymbolic form of some commonsense knowledge

to this skill.

Planning Knowledge representation Intelligent agents must be able to set Knowledge

representation

and

goals and achieve them. They need a

knowledge engineering are central to AI

way to visualize the future (they must

research.

problems

have a representation of the state of the

machines are expected to solve will

world and be able to make predictions

require extensive knowledge about the

about how their actions will change it)

Many

of

the

and be able to make choices that

mathematical

maximize the utility (or "value") of the

learning

available choices.

performance is a branch of theoretical

planning

uses

the

of

algorithms

computer Multi-agent

analysis

science

machine

and

their

known

as

computational learning theory.

cooperation and competition of many agents to achieve a given goal. Emergent

Natural language processing

behavior such as this is used by evolutionary

algorithms

and

swarm

Natural machines

intelligence.

language the

processing

ability

to

read

gives and

understand the languages that humans

Learning

speak. Many researchers hope that a sufficiently powerful natural language

Machine learning has been central to AI research

from

the

beginning.

Unsupervised learning is the ability to find patterns in a stream of input. Supervised

learning

includes

both

classification and numerical regression. Classification is used to determine what

processing system would be able to acquire knowledge on its own, by reading the existing text available over the

internet.

applications

Some of

straightforward

natural

language

processing include information retrieval (or text mining) and machine translation.

category something belongs in, after seeing a number of examples of things from several categories.

Motion and manipulation

Regression

takes a set of numerical input/output

The field of robotics is closely related to

examples and attempts to discover a

AI. Intelligence is required for robots to

continuous function that would generate

be able to handle such tasks as object

the

In

manipulation and navigation, with sub-

reinforcement learning the agent is

problems of localization (knowing where

rewarded

you are), mapping (learning what is

outputs

from

for

the

good

inputs.

responses

and

punished for bad ones. These can be

around

analyzed in terms of decision theory,

(figuring out how to get there).

using

concepts

like

utility.

The

you)

and

motion

planning

Perception

Creativity

Machine perception is the ability to use

A sub-field of AI addresses creativity

input from sensors (such as cameras,

both theoretically (from a philosophical

microphones, sonar and others more

and

exotic) to deduce aspects of the world.

practically (via specific implementations

Computer vision is the ability to analyze

of systems that generate outputs that can

visual input. A few selected subproblems

be considered creative). A related area of

are

computational

speech

recognition,

facial

psychological

perspective)

research

is

and

Artificial

recognition and object recognition.

Intuition and Artificial Imagination.

Social intelligence

General intelligence

Emotion and social skills play two roles

Most researchers hope that their work

for an intelligent agent. First, it must be

will eventually be incorporated into a

able to predict the actions of others, by

machine

understanding

and

(known as strong AI), combining all the

involves

skills above and exceeding human

elements of game theory, decision

abilities at most or all of them. A few

theory, as well as the ability to model

believe that anthropomorphic features

human emotions and the perceptual

like

skills to detect emotions.) Also, for good

artificial brain may be required for such

human-computer

a project.

emotional

their

states.

motives (This

interaction,

an

with

artificial

general

intelligence

consciousness

or

an

intelligent machine also needs to display emotions. At the very least it must appear polite and sensitive to the humans it interacts with. At best, it should have normal emotions itself.

Many of the problems above are considered AI-complete: to solve one problem, you must solve them all. For example, even a straightforward, specific task like machine translation requires that the machine follow the author's argument (reason), know what is being talked about (knowledge), and faithfully

reproduce the author's intention (social intelligence).

Machine

translation,

Cybernetics

and

brain

simulation

therefore, is believed to be AI-complete: it may require strong AI to be done as

In the 1940s and 1950s, a number of

well as humans can do it.

researchers explored the connection between neurology, information theory, and cybernetics. Some of them built machines that used electronic networks to exhibit rudimentary intelligence, such

Approaches

as W. Grey Walter's turtles and the There is no established unifying theory

Johns Hopkins Beast. Many of these

or paradigm that guides AI research.

researchers gathered for meetings of the

Researchers disagree about many issues.

Teleological

A few of the most long standing

University and the Ratio Club in

questions

remained

England. By 1960, this approach was

unanswered are these: should artificial

largely abandoned, although elements of

intelligence simulate natural intelligence,

it would be revived in the 1980s.

that

have

by studying psychology or neurology? Or is human biology as irrelevant to AI research

as

bird

biology

is

Society

at

Princeton

Symbolic

to

aeronautical engineering? Can intelligent

Cognitive simulation

behavior be described using simple,

Economist Herbert Simon and

elegant principles (such as logic or

Allen Newell studied human

optimization)? Or does it necessarily

problem

require solving a large number of

attempted to formalize them, and

completely unrelated problems? Can

their work laid the foundations of

intelligence be reproduced using high-

the field of artificial intelligence,

level symbols, similar to words and

as well as cognitive science,

ideas? Or does it require "sub-symbolic"

operations

research

and

processing?

management

science.

Their

solving

skills

and

research team used the results of

psychological

experiments

to

Researchers at MIT (such as

develop programs that simulated

Marvin Minsky and Seymour

the techniques that people used

Papert)

to solve problems. This tradition,

difficult problems in vision and

centered at Carnegie Mellon

natural

University

eventually

required ad-hoc solutions – they

culminate in the development of

argued that there was no simple

the Soar architecture in the

and general principle (like logic)

middle 80s.

that would capture all the aspects

would

Logic based

found

that

language

solving

processing

of intelligent behavior. Roger

Unlike Newell and Simon, John

Schank described their "anti-

McCarthy felt that machines did

logic" approaches as "scruffy"

not need to simulate human

(as

thought, but should instead try to

paradigms

find the essence of abstract

Stanford).

reasoning and problem solving,

knowledge bases (such as Doug

regardless of whether people

Lenat's Cyc) are an example of

used the same algorithms. His

"scruffy" AI, since they must be

laboratory at Stanford (SAIL)

built by hand, one complicated

focused on using formal logic to

concept at a time.

opposed

to

the

at

"neat"

CMU

and

Commonsense

solve a wide variety of problems, including

knowledge

representation,

planning

and

Knowledge based When

computers

large

learning. Logic was also focus of

memories

the work at the University of

around 1970, researchers from all

Edinburgh

in

three traditions began to build

the

knowledge into AI applications.

development of the programming

This "knowledge revolution" led

language Prolog and the science

to

of logic programming.

deployment of expert systems

Europe

and

which

"Anti-logic" or "scruffy"

elsewhere led

to

the

(introduced

became

with

available

development by

and

Edward

Feigenbaum),

the

first

truly

Statistical

successful form of AI software. The knowledge revolution was

In the 1990s, AI researchers developed

also driven by the realization that

sophisticated mathematical tools to solve

enormous amounts of knowledge

specific subproblems. These tools are

would be required by many

truly scientific, in the sense that their

simple AI applications.

results

are

verifiable,

Sub-symbolic

had achieved great success at simulating demonstration

thinking

in

small

programs. Approaches

based on cybernetics or neural networks were abandoned or pushed into the background. By the 1980s, however, progress in symbolic AI seemed to stall and

many

believed

and

measurable they

have

and been

responsible for many of AI's recent

During the 1960s, symbolic approaches high-level

both

that

symbolic

systems would never be able to imitate

successes. The shared mathematical language has also permitted a high level of collaboration with more established fields (like mathematics, economics or operations research). Stuart Russell and Peter Norvig describe this movement as nothing less than a "revolution" and "the victory of the neats."

Tools

all the processes of human cognition,

In the course of 50 years of research, AI

especially perception, robotics, learning

has developed a large number of tools to

and pattern recognition. A number of

solve the most difficult problems in

researchers began to look into "sub-

computer science. A few of the most

symbolic" approaches to specific AI

general of these methods are discussed

problems.

below.

1. Bottom-up, embodied, situated,

Search and optimization

behavior-based or nouvelle AI 2. Computational Intelligence

Many problems in AI can be solved in theory by intelligently searching through many possible solutions: Reasoning can

be reduced to performing a search. For

Logic

example, logical proof can be viewed as searching for a path that leads from

Logic

premises to conclusions, where each step

representation and problem solving, but

is the application of an inference rule.

it can be applied to other problems as

Planning algorithms search through trees

well. For example, the satplan algorithm

of goals and subgoals, attempting to find

uses logic for planning and inductive

a path to a target goal, a process called

logic programming is a method for

means-ends

learning.

algorithms

analysis. for

moving

Robotics limbs

and

grasping objects use local searches in configuration

space.

Many

learning

algorithms use search algorithms based on optimization.

is

used

for

knowledge

Several different forms of logic are used in

AI

research.

Propositional

or

sentential logic is the logic of statements which can be true or false. First-order logic also allows the use of quantifiers

A very different kind of search came to

and predicates, and can express facts

prominence in the 1990s, based on the

about objects, their properties, and their

mathematical theory of optimization. For

relations with each other. Fuzzy logic, is

many problems, it is possible to begin

a version of first-order logic which

the search with some form of a guess

allows the truth of a statement to be

and then refine the guess incrementally

represented as a value between 0 and 1,

until no more refinements can be made.

rather than simply True (1) or False (0).

These algorithms can be visualized as

Fuzzy systems can be used for uncertain

blind hill climbing: we begin the search

reasoning and have been widely used in

at a random point on the landscape, and

modern industrial and consumer product

then, by jumps or steps, we keep moving

control systems. Subjective logic models

our guess uphill, until we reach the top.

uncertainty in a different and more

Other

are

explicit manner than fuzzy-logic: a given

simulated annealing, beam search and

binomial opinion satisfies belief +

random optimization.

disbelief + uncertainty = 1 within a Beta

optimization

algorithms

distribution. By this method, ignorance can be distinguished from probabilistic

statements that an agent makes with high

decision

confidence.

information value theory.

Default

logics,

non-

theory,

decision

analysis,

monotonic logics and circumscription are forms of logic designed to help with

Classifiers

have been designed to handle specific domains

of

knowledge,

such

as:

description logics; situation calculus, event calculus and fluent calculus (for representing events and time); causal calculus; belief calculus; and modal logics.

The simplest AI applications can be divided into two types: classifiers ("if shiny then diamond") and controllers ("if shiny then pick up"). Controllers do however also classify conditions before inferring

methods

for

actions,

and

therefore

classification forms a central part of many

Probabilistic

statistical

learning methods

default reasoning and the qualification problem. Several extensions of logic

and

AI

systems.

Classifiers

are

functions that use pattern matching to determine a closest match. They can be

uncertain reasoning

tuned according to examples, making Many problems in AI (in reasoning,

them very attractive for use in AI. These

planning,

and

examples are known as observations or

robotics) require the agent to operate

patterns. In supervised learning, each

with

uncertain

pattern belongs to a certain predefined

information. AI researchers have devised

class. A class can be seen as a decision

a number of powerful tools to solve

that has to be made. All the observations

these problems using methods from

combined with their class labels are

probability theory and economics.

known as a data set. When a new

learning,

incomplete

perception or

observation is received, that observation A key concept from the science of economics is "utility": a measure of how

is

classified

based

on

previous

experience.

valuable something is to an intelligent agent. Precise mathematical tools have

A classifier can be trained in various

been developed that analyze how an

ways; there are many statistical and

agent can make choices and plan, using

machine learning approaches. The most

widely used classifiers are the neural

direction) and recurrent neural networks

network, kernel methods such as the

(which allow feedback). Among the

support

k-nearest

most popular feedforward networks are

neighbor algorithm, Gaussian mixture

perceptrons, multi-layer perceptrons and

model, naive Bayes classifier, and

radial basis networks. Among recurrent

decision tree. The performance of these

networks, the most famous is the

classifiers have been compared over a

Hopfield net, a form of attractor

wide

Classifier

network, which was first described by

performance depends greatly on the

John Hopfield in 1982. Neural networks

characteristics

be

can be applied to the problem of

classified. There is no single classifier

intelligent control (for robotics) or

that works best on all given problems;

learning, using such techniques as

this is also referred to as the "no free

Hebbian

lunch" theorem. Determining a suitable

learning.

vector

range

machine,

of

tasks.

of

the

data

to

learning

and

competitive

classifier for a given problem is still

Control theory

more an art than science.

Control

Neural networks

theory,

cybernetics, The study of artificial neural networks

has

the

grandchild

many

of

important

applications, especially in robotics.

began in the decade before the field AI research was founded, in the work of Walter Pitts and Warren McCullough. Other important early researchers were Frank Rosenblatt, who invented the perceptron developed

and the

Paul

Werbos

who

backpropagation

algorithm. The main categories of networks are acyclic or feedforward neural networks (where the signal passes in only one

Languages AI researchers have developed several specialized languages for AI research, including Lisp and Prolog.

Evaluating progress In 1950, Alan Turing proposed a general procedure to test the intelligence of an agent now known as the Turing test.

This procedure allows almost all the

diagnosis, robot control, law, scientific

major problems of artificial intelligence

discovery and toys. However, many AI

to be tested. However, it is a very

applications are not perceived as AI: "A

difficult challenge and at present all

lot of cutting edge AI has filtered into

agents fail.

general applications, often without being called AI because once something

Artificial

intelligence

can

also

be

evaluated on specific problems such as small problems in chemistry, handwriting recognition and game-playing. Such tests have been termed subject matter expert Turing tests. Smaller problems provide more achievable goals and there are an ever-increasing number of positive results.

becomes useful enough and common enough it's not labeled AI anymore." "Many thousands of AI applications are deeply embedded in the infrastructure of every industry." In the late 90s and early 21st century, AI technology became widely used as elements of larger systems, but the field is rarely credited for these successes.

The broad classes of outcome for an AI test are: •

Optimal: it is not possible to perform better

•

Strong super-human: performs better than all humans

•

Super-human: performs better than most humans

•

Sub-human: performs worse than most humans

Applications

CONCLUSION: In

order

to

Artificial intelligence has been used in a

competitiveness,

wide range of fields including medical

compelled

to

maintain

their

companies

feel

adopt

productivity

increasing measures. Yet, they cannot

other, their joint contribution can be of

relinquish the flexibility their production

unquestionable

cycles need in order to improve their

understand a little better the importance

response, and thus, their positioning in

of ES within the production system

value

in

order

to

the market. To achieve this, companies must combine these two seemingly opposed principles. Thanks to new technological advances, this combination is already a working reality in some companies. It is made possible today by the

implementation

of

computer

integrated manufacturing (CIM) and artificial intelligence (AI) techniques,

References •

fundamentally by means of expert systems (ES) and robotics. Depending on how these (AI/CIM) techniques contribute

to

automation,

their

immediate effects are an increase in productivity and cost reductions. Yet also, the system's flexibility allows for easier adaptation and, as a result, an increased ability to generate value, in other

words,

competitiveness

is

improved. The authors have analyzed three studies to identify the possible benefits or advantages, as well as the inconveniences,

that

this

type

of

technique may bring to companies, specifically in the production field. Although the scope of the studies and their approach differ from one to the

•

Russell, Stuart J.; Norvig, Peter (2003), Artificial Intelligence: A Modern Approach (2nd ed.), Upper Saddle River, New Jersey: Prentice Hall, ISBN 0-13790395-2, http://aima.cs.berkeley.edu/ Kurtzweil, Ray (2005), The singularity is near : when humans transcend biology, New York: Viking, ISBN 9780670033843