cap 492 course dr. souham meshoul 1 artificial intelligence cap492 dr. souham meshoul information...

CAP 492 Course Dr. Souham Meshoul

1

Artificial Intelligence CAP492

Dr. Souham Meshoul

Information Technology DepartmentCCIS – King Saud University

Riyadh, Saudi Arabia

[email protected]


2

INTRODUCTION TO ARTIFICIAL INTELLIGENCE

Chapter 1


3


Goal of Artificial Intelligence: Not only to understand how does mind work? but also how to build intelligent entities?.

Engineering point of view: -Solve real-world problems using knowledge and reasoning -Develop concepts, theory and practice of building intelligent entities - Emphasis on system building

Scientific point of view: - Use computers as a platform for studying intelligence itself - Emphasis on understanding intelligent behavior.

Artificial Intelligence is one of the newest sciences which emerged after the world war II. AI represents a big and open field.

The name Artificial Intelligence was adopted for the first time in 1956. (Computational Intelligence)

Artificial Intelligence can be viewed as a universal field: Ho to automate intellectual tasks?


4


What is artificial Intelligence?

Several definitions are available in the literature. Thinking vs Behavior Model humans vs Work from an ideal standard

Two points of views:

1. Thinking/Acting humanly: success is measured in term of fidelity to human performance.

2. Thinking/Acting rationally: success is measured using an ideal concept of intelligence called Rationality.

Rational System = system which does the « right thing » given what it knows.


5


Some AI Definitions:

According to thought processes and reasoning

Thinking like humans:

« The exciting new effort to make computers think…machines with minds, in the full and literal sense. » (Haugeland, 1985).

« The automation of activities that we associate with human thinking, activities such as decision-making, problem solving, learning… » (bellman, 1978).

Thinking rationally:

« The study of mental faculties through the use of computational models. » (Charniak and Mcdermott, 1985).

« The study of the computations that make it possible to perceive, reason and act. » (Winston 1992).


6


Some AI Definitions: According to behavior

Acting Humanly

« The art of creating machines that perform functions that require intelligence when performed by people. » (Kurzweil, 1990)

« The study of how to make computers do things at which, at the moment, people are better. » (Rich and Knight, 1991)

Acting rationally

« Computational intelligence is the study of the design of intelligent agents. » (Poole et al.,1998).

« AI…is concerned with intelligent behavior in artifacts. » (Nilsson, 1998).


7


Some other definitions

• Weak AI: (Searle) AI develops useful, powerful applications

• Strong AI: claims machines have cognitive minds comparable to humans

debates on strong AI bring up some difficult conceptual problems in philosophy..

Keep this one : Science of making machines do things that would require

intelligence if done by humans Marvin Minsky


8


Acting Humanly

Turing Test: Alan Turing 1950 « Computing Machinery and Intelligence »


9


Test: The computer passes the “test of intelligence” if a human, after posing some written questions, cannot tell whether the responses were from a person or not.

Provides an operational definition of AI. Still being used today within the AI community in annual competitions – the Loebner Prize

($100,000 and a true 18K gold medal)

To give an answer, the computer would need to posses some capabilities:

Natural language processing: To communicate successfully. Knowledge representation: To store what it knows or hears. Automated reasoning: to answer questions and draw conclusions using stored information. Machine learning: To adapt to new circumstances and to detect and extrapolate patterns. Computer vision: To perceive objects. Robotics to manipulate objects and move.


10


Example: ELIZA program, some extracts from the test: Patient: Men are all alike. ELIZA: In what way? Patient: They are always bugging about something or the other. ELIZA: Can you think of a specific example? Patient: Well, my boyfriend made me come here. ELIZA: Your boyfriend made you come here? Patient: He says I’m depressed much of the time. ELIZA: I am sorry you are depressed. Problems: ELIZA and similar programs stressed simple syntactic analysis and generation of sentences.

They used pattern matching with known sentences. Turing’s test is not reproducible or amenable to mathematical analysis (it suggests duplicate

an examplar). An alternative: Study the underlying principles of intelligence (Wright brothers)


11


Thinking Humanly:

Program think like human → How humans think? Requires Scientific theories of internal activities of the brain (cognitive science and cognitive

neuroscience).

Example: The General Problem Solver (GPS designed by Newell and Simon In 1963) was meant to be a

program that simulated human thought.

GPS: used means-end analysis in its search for solutions, computing the difference between the goal and current, and then attempting to minimize the difference.

Newell and Simon by comparing GPS traces with those of human subjects discovered that the behavior of GPS was largely a subset of human behavior


12


Thinking rationally:

The Laws of Thought approach is based on pattern for argument structure arising from Aristostle’s syllogisms.

Example, “Socrates is a man; all men are mortal, therefore Socrates is mortal.” The laws of thought initiated the field of logic.

The formal logic movement was advanced by Peano, Boole, Frege,, Godell and others (late 1800’s and early 1900’s)

Inspired perhaps by early progress, Hibert became a proponent of a school of thought known as logicism or formalism. The goal of this was to devise a logic, or formal system, capable of deriving all mathematical theorems.


13


Acting rationally:

Modern AI can be characterized as the engineering of rational agents.

An agent is simply an entity that perceives and acts. A rational agent is an entity that perceives, reasons and acts rationally (correctly).


14


Foundations:

An interdisciplinary subject found on:

Philosophy, mathematics, economics, neuroscience, psychology, computer engineering, linguistics, and so on


15


History of Artificial Intelligence

Big dream

Ultimately, we are dealing with the question: “What are we (human beings) doing when we are thinking?”

Thought processes in the human mind are computational in nature. There are mechanistic procedures for generating these thoughts.

Such computations can be simulated and implemented by a Turing machine. Therefore, it can be programmed.


16



Early days (1943-1955)

1943: first piece of AI work: Warren McCulloch and Walter Pitts Model of artificial neurons Mathematical learnable functions that generate “on/off” depending on inputs (logic gates) Any computable function can be computed by a network of connected neurons. Suitably defined networks can learn.

1949: Hebbian learning A mechanism for updating the connection strength of a neuron. Today, neurologists have confirmed that something similar to Hebbian learning indeed is going on in our brain

when we are learning.

1950: Turing test, complete vision of AI in “computing machinery and Intelligence” 1951: first neural network computer

Implemented by M. Minsky and D. Edmonds


17



Early days (1943-1955): Mcculloch and pitts artificial neuron

0.3

-1

0.5

1

0 1


18


History of Artificial IntelligenceBirth of AI 1956

1956: Dartmouth Conference Organized by John McCarthy and colleagues for starting a new area in studying

computation and intelligence. John McCarthy introduced the term “artificial intelligence” in the conference.

The next 20 years witnessed steady growth of the field led by the pioneers appeared in the Dartmouth conference.


19



Expectations and Initial enthusiasm (1952 – 1969)

1956: Samuel’s checkers program: First game playing program achieving human-competitive performance.

1957: Simon’s general problem solver (GPS): Imitates the way a human would solve planning problems.

1958: Invention of LISP by J. McCarthy. The first AI programming language.1958: Minsky’s microworlds

The concept of creating a “controlled environment” in which problem solving appears to require intelligence was born. The study of computation and intelligence can become more manageable in these micro-worlds


20


Expectations and Initial enthusiasm (1952 – 1969)

1963: Thomas Evan’s program ANALOGSolved analogy problems in an IQ test.

1965: ELIZASimulates a dialog with a computer in English on any topic.Became popular when programmed to simulate a psychotherapist (Fedora’s Emacs).

1967: Dendral program (developed at Stanford)First successful program for scientific reasoning – one of the earlier rule based expert systems. A program that can infer molecular structures given the information provided by a mass spectrometer (that gives the masses of the various fragments of a molecule). The program relies on expert knowledge (encoded as rules) to constraint the generation of possible molecular structures that are

consistent with the information from the mass spectrometer


21



Reality Check (1966 – 1973): series of disappointments and frustrations AI was poured little buckets of “reality cold water”Problems Most early systems contain little or no knowledge of their subject matter

Knowledge acquisition bottleneck. Example: Poor performance of earlier machine translation system (Russian English): “the spirit is willing

but the flesh is weak” was translated to “the vodka is good but the meat is rotten”.

Computational Intractability of AI problems Theory of computational complexity was not developed. Polynomial solvable problems, NP-completeness, etc People thought a faster machine could solve any hard problem. Initial frustration with theorem proving led to a disappointment in AI. Theorem proving is exponential in

complexity


22



Resurgence (1969 – 1979)

1971: T. Winograd’s Ph.D. thesis (MIT) demonstrated a system that can understand English in a micro-domain (the block world).

1972: PROLOG was developed by a group of Europeans and became alternative to LISP as an AI programming language.

1974: MYCIN was developed by Ted Shortliffe. Expert system for medical diagnosis. Sometimes called the first expert system.

1978: The Version Space algorithm was developed by Tom Mitchell at Stanford. First symbolic machine learning algorithm. “Father of Machine Learning”.

1979: Non-monotonic logic. Began to be formalized by John McCarthy and his colleagues.


23



Resurgence (1969 – 1979): Winograd 1972


24


History of Artificial IntelligenceAI becomes an industry (1980 – present) AI started to become industrially and commercially beneficial

1982: R1 was deployed at DEC – an expert system that saved the company around $40M / year

Du Pont had 100 in use and an estimated 500 in development at late 90’s to early 21st century

At an international level, AI was considered a part of a country’s technological developments

Japan: “First Generation” project (10 year plan to build intelligence machines running in Prolog)

USA: Microelectronics and Computer Technology Corporation (MCC) was formed in response

Britain: Funding for AI was reinstated


25


History of Artificial IntelligenceRenewing with connectionism and AI becomes a science (1986 – present)

Work of the physicist John Hopfield (1982) on using techniques from statistical mechanics.

Connectionist models of intelligent systems competitor to the symbolic models (Newell and Simon) and logicist approach (McCarthy). (complementary approaches in fact).

Several revolutions in many fields: pattern recognition, computer vision, robotics…

Emergence of intelligent agents.


26


Examples of AI applications: Game Playing TDGammon, the world champion backgammon player, built by Gerry Tesauro of IBM research.

Perception: keyboard input.Reason: reinforcement learning.Actuation: graphical output shows dice and

movement of piece.

Deep Blue chess program beat world champion Gary KasparovPerception: input symptoms and test results.

Reason: Bayesian networks, Monte-Carlo simulations.

Actuation: output diagnoses and further test suggestions.


27


Examples of AI applications: Natural Language Understanding

Natural language understanding (spell checkers, grammar checkers)

AI translators – spoken to and prints what one wants in foreign languages : Alta Vista’s translation of web pages.

Advanced systems can answer questions based on the information in the text and produce useful summaries.

PROVERB (Littman 1999) crossword puzzles

Examples of successes : English conversationSTART system: accesses raw data tables, and then can carry on a

dialogue


28

INTRODUCTION TO ARTIFICIAL INTELLIGENCE Examples of AI applications: Expert systems

In geologyprospector expert system carries evaluation of mineral potential of geological site

or region Diagnostic Systems

Pathfinder, a medical diagnosis system (suggests tests and makes diagnosis) developed by Heckerman and other Microsoft research

Microsoft Office Assistant in Office provides customized help by decision-theoretic reasoning by an individual user.

MYCIN system for diagnosing bacterial infections of the blood and suggesting treatments

System Configuration"XCON" (for custom hardware configuration) configures computers doing work

of 300 people using 10,000 rules


29


Examples of AI applications: Robotics

Robotics becoming increasing important in various areas like: games, to handle hazardous conditions and to do tedious jobs among other things.

Examples: automated cars, ping pong player, mining, construction, robot assistant in microsurgery,…


30


Main issues in AI Representation Search: many tasks can be viewed as searching a very large problem space for

solution space

Inference: related to search, inferring other facts from some given facts. e.g., knowing all “elephants have trunks” and “Jo is an elephant,” can we answer does Jo have a trunk?

Learning: inductive inference, neural networks, artificial life, genetic algorithms, evolutionary strategies

Planning: starting with general facts about the world, facts about the effects of basic actions, facts about a particular situation, and a statement of a goal, generate a strategy for achieving that goal in terms of a sequence of primitive steps or actions


31


Summary Intelligence is studied from many perspectives: Are you concerned with thinking or

behavior?

AI can help us solve difficult, real-world problems, creating new opportunities in business, engineering, and many other application areas.

The history of AI has had cycles of success, misplaced optimism, and resulting cutbacks in enthusiasm and funding. There have also been cycles of introducing new creative approaches and systematically refining the best ones.

AI has advanced more rapidly in the past decade because of greater use of the scientific method in experimenting with and comparing approaches.

cap 492 course dr. souham meshoul 1 artificial intelligence cap492 dr. souham meshoul information...

Documents

artificial intelligence

ideal concept of intelligence

ai definitions

intelligent behavior

searle ai

definitions weak ai

reasoning thinking

human thinking