Transcript An `artificial intelligence`

I - Quick look
1
1 - Artificial Intelligence ? (a few definitions)
1.1 - An ‘artificial intelligence’ ?

Artificial + Intelligence
• Intelligence : ability to understand and reason (like a human being does)
• Artificial : man made
Thus, Artificial Intelligence or “A.I.” is the study of cognitive phenomenon (dealing with
knowledge) made by a machine (real or not) in order to bring it as near as possible to
those used by man .
2
V 3.19
1.2 - ‘Larousse’ dictionary

Set of theories and techniques used in order to build machines able to simulate
human intelligence in a definite domain.
3
V 3.19
1.3 - Laurière (Jean-Louis)

Every problem for which no algorithmic solution is known, a priori comes under AI.
An algorithm is :
“an ordered sequence of operations, that can run on a computer and gives a
solution in a finite time”.
Examples :
•
•
•
•
•
•

playing chess
sumarising a text, translating it.
diagnose an illness
recognising someone’s face
proving a theorem
learning…
The object of AI is to rebuild with artificial means, for the most part computers,
reasonings and intelligent actions.
4
V 3.19
1.4 - As a conclusion…


One of the the aims of AI is to describe quite precisely human reasoning to make
it work on a computer.
But replacing human reasoning is not enough, the reasoning has to be ‘complex’
(i.e. not evident).
n
i
Add the n
first
integers
i =1
n x (n-1)
2
begin
S := 0;
for i := 1 to
n do
S := S +
i;
S := (n * succ(n))/2
Figure 1.1 - A trivial problem
5
V 3.19
1.5 - Kind of summary

From a ‘problem’ point of view : AI deals with ‘intelligent’, complex problems
difficult or impossible to solve with current means.

From a ‘tool’ point of view : AI wants to make more human a computer (in its
input/output, reasoning…).
6
V 3.19
1.6 - By the way,
what is ‘intelligence’ ?
Chimpanzee
Diagnose
Expert System
Ability to live
together
Ability to find the
reason of the
break-down
Figure 1.2 - Intelligence ?
7
V 3.19
2 - History
2.1 - Birth

August 1956, Darmouth College conference (Hanover, New Hampshire), John MacCarthy
gives the science its name : ‘Artificial Intelligence’ and propose, in presence of Minsky,
Newell, Simon, Shanon...
“to study if all that is behind ‘intelligence’
could be describe precisely enough
to be fulfilled by a machine”.
8
V 3.19
2.2 - AI & Logic

First programs : “Logic Theorist” (theorem proving) and a chess game both from
Newell & Simon. (Lisp too !)

Logic, Mathematics and AI are deeply linked (paradoxes). Since antiquity
(Aristotle’s syllogism) man has been interested in “clever” machines.
9
V 3.19
2.3 - Thinking machines
Charles Babbage and
Adelaïde Augusta Baroness of Lovelace

Babbage’s analytic machine, 1842 (fictive machine) was able, according to Ada
Baroness of Lovelace, to alter itself ; this machine had a warehouse (memory)
and a factory (processor)

Pascal or Leibniz machines (XVII th C.) only computed fixed operations :
additions, multiplications.
10
V 3.19
2.4 - Alan Turing
(Universal machine 1936)

English mathematician dead in 1954. During the second world war, he achieved
the machine he had first conceived in 1936, a machine which decodes, in less
than a day, German messages coded by Enigma.

A TUM can execute, hopelessly slowly, any of our present programs.
11
V 3.19
2.5 - Four decades of AI

The 50s : Birth of AI.

• Scientists’ exaggerated optimism.
• Under estimation of problems leading to failure especially in games (chess)
and speech recognition fields.
The 60s : Real start.
• Heuristic research algorithms (introduction of an intelligent function in an
algorithm)
• GPS
• MACSYMA (formal calculation in mathematics)
12
V 3.19
2.5 - Four decades of AI (2)

The 70s : Explosions of studies
• Foundations of AI in :
•
•
•
•
Representation of knowledge
Expert systems
Natural language
Advanced robotics

The 80s : Entry of AI in economy

• AI goes from research towards industry. Research goes on and even
increases.
The 90s : Diversification
• AI techniques, such as fuzzy logic, object representation, natural language or
forward chaining, are integrated in classical tools (data bases, automaton,
controllers, help systems…)
13
V 3.19
3 - Applications fields
We are now going now to try to enumerate, rather schematically, applications fields
of AI. We will give, for every scope, application examples, and the names of
artificial intelligence tool used.
We will first deal with points what won’t be dealt more deeply later in this course.
14
V 3.19
3.1 - To be seen later

Knowledge representation (in II-A)

Problem resolution (in IV)

Reasoning (in II-B)

Problem where the methodology of resolution is impossible to get (in III)
15
V 3.19
3.2 -Human reasoning vs computer reasoning

Psychologists
Paradox : Complex tasks are easy to explain, whereas innate tasks can be
impossible to explain.
Play chess
Talk with someone
Invert a matrix
Recognize a face
Make a diagnosis Be conscious of oneself being…

Help device in Santa-Anna

Loan in a bank
A machine can tell if a loan is authorized after having asked questions to a
customer.

Turing test (slow and with errors ?)

Problem of the limit of human intelligence, what does the machine have to do
when it can do better than man ?
16
V 3.19
3.3 - Natural language

Kind of problems
• Sentence interpretation and generation in a man-computer dialog context.
• In natural language, word for word translation never works.
• To understand (and translate) a sentence we need to take different language
dimensions :
•
•
•
•
The lexicon (dictionary)
The syntax (language grammar)
The semantics (meaning)
The pragmatic (social context)
17
V 3.19
Total Recall
18
V 3.19
3.3 - Natural language (2)

Does a unique representation of something in the brain exist whatever the tongue
and the form of the sentence may be ?

Technical tools

• Prolog language, semantic networks, algorithmic languages…
Example
• www.pagesjaunes.fr (very few AI)
A human being knows that he can buy bread at the baker’s, a computer
doesn’t, pagejaunes does. The machine proceeds toward human.
The phone data base has a phonetic index MEUZIN —> MESSIN,
MORROIT -> ...
• Since 93 : Natural language : “I want to hire a car”, “I want to get rid of my
mother-in-law”.
19
V 3.19
3.4 - Speech processing

Purpose
• This point covers two aspects : speech recognition (you speak to a machine
and are understood) and speech synthesis (the machine talks).
• Problems
• Noise in the analogic voice signal
• Various speakers
• Continuous speech
• There are tools (via voice, kid games…) but they do not usually go further
than half the semantic level.
• Technical tools
• Neural networks, algorithms, morphological models
20
V 3.19
3.4 - Speech processing (2)
Figure 1.3 - The whole process of a natural dialog between man and machine
21
V 3.19
3.5 - Robotics

“Robot” : Czech word meaning “Forced work”.

The robot is closer to human because we physically see a result of its reasoning.

It frightens us a little : What if this intelligence would dominate our ? (cf. Asimov,
Huxley, Wells, Orwell…).

In industry, robots can for example look at their environment and take a decision
in terms of task scheduling.
22
V 3.19
3.6 - Vision, pattern recognition

Kind of problem
• Nowadays, a whole scene (a photo for example) is quite impossible to
understand, on the contrary, things are done in
•
•
•
•
•
Printed or hand made type recognition,
Cell recognition in a microscopic picture,
Specific areas in a satellite picture,
Numbers or defaults on parts that comes before a camera,
…
23
V 3.19
3.6 - Vision, pattern recognition (2)
Figure 1.4 - A french zip code

Technical tools
• Algorithms or neural networks.
24
V 3.19
3.7 - Other : Logic problems

Prolog (PROgrammation LOGique) created by Alain Colmerauer in 1972 in
Marseille-Luminy university (France).

Prolog is a declarative language, it uses Robinson’s resolution principal on Horn
clauses.

Example
• father (john, peter).
• father (andrew, john).
• father (john, michael).

Use
• father (X, peter) ?
• father (john, X) ?
• father (john, andrew) ?
Andrew
John
Michael
Peter
25
V 3.19
3.7 - Other : Logic problems (2)

Then new pieces of information are added (they are methods but they use the
same syntax as data) :
grandfather (X, Y) :father (X, A),
father (A, Y).
brother (X, Y) :-

father (A, X),
father (A, Y).
Second use
• grandfather (andrew, peter) ?
• brother (peter, michael) ?
• grandfather (X, michael) ?
• grandfather (X, Y) ?
26
V 3.19
3.8 - Other : Symbolic problems

Symbols manipulation. We do not study a number for what it is but rather for what
it stands for.

A unique syntax to represent both data and program : lists

• Code = Data = Symbols (represented by lists)
Language : Lisp (1958, John MacCarthy)

Example : derived function in Lisp
(de derived (expr)
(cond ((= expr constante) 0)
((= expr var) 1)
((= (car expr) +) (+(derived (cadr expr))
(derived (caddr expr))))
((= (car expr) x) ( ... u’v + uv’ ...))
...))
27
V 3.19
3.9 - Other : Mixed problems

When there is a part of things that are known, and another part that is not. In a
process for example, it can be useful to get help from other tools :
•
•
•
•
“neuroAgent” technology
Case based reasoning (CBR)
Induction
…
28
V 3.19
Two ways...

For a computer to deal with intelligent functions, two different ways exist
• A connexionist approach
• A cognitive approach
29
V 3.19
Cognitive or symbolic approach
30
V 3.19
Connexionist or neural approach
31
V 3.19