0% found this document useful (0 votes)
54 views70 pages

IA01 en

This document provides an introduction to artificial intelligence. It discusses intelligence and different definitions of artificial intelligence. It also outlines the history and subfields of AI as well as provides an overview of the topic. The document is presented by Florin Leon and discusses intelligence, both natural and artificial intelligence. It examines the various types of intelligence and what capabilities are required for intelligence.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
54 views70 pages

IA01 en

This document provides an introduction to artificial intelligence. It discusses intelligence and different definitions of artificial intelligence. It also outlines the history and subfields of AI as well as provides an overview of the topic. The document is presented by Florin Leon and discusses intelligence, both natural and artificial intelligence. It examines the various types of intelligence and what capabilities are required for intelligence.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 70

Artificial Intelligence

1. Introduction to Artificial Intelligence


Florin Leon

“Gheorghe Asachi” Technical University of Iași


Faculty of Automatic Control and Computer Engineering

“Al. I. Cuza” University of Iași


Faculty of Computer Science
Florin Leon - Inteligenta artificiala
https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Introduction to Artificial Intelligence
1. Overview
2. Intelligence and artificial intelligence
3. Definitions of artificial intelligence
4. History of artificial intelligence
5. Subfields and applications
6. Conclusions

Florin Leon - Inteligenta artificiala 2


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Introduction to Artificial Intelligence
1. Overview
2. Intelligence and artificial intelligence
3. Definitions of artificial intelligence
4. History of artificial intelligence
5. Subfields and applications
6. Conclusions

Florin Leon - Inteligenta artificiala 3


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Overview
 Artificial Intelligence (AI) tries to build
intelligent entities
 It is possible, we only need to understand
ourselves
 We need to find the right goals
 Computers allow the implementation of ideas
 AI has shown significant progress in the past
few years

Florin Leon - Inteligenta artificiala 4


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Introduction to Artificial Intelligence
1. Overview
2. Intelligence and artificial intelligence
3. Definitions of artificial intelligence
4. History of artificial intelligence
5. Subfields and applications
6. Conclusions

Florin Leon - Inteligenta artificiala 5


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Artificial Intelligence
 Artificial (easy to define)
 Man-made, non-natural, manufactured
 AI is the science of making machines
do things that would require intelligence
if done by people

Florin Leon - Inteligenta artificiala 6


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Intelligence: what it is
 Intelligence is much harder to define
 It is actually an umbrella term for many related abilities
 The ability to reason, plan, solve problems, think
abstractly, comprehend complex ideas, learn quickly
and learn from experience (Mainstream Science of
Intelligence, 1994)
 Understanding
 Learning, perceiving relationships
 Reasoning

Florin Leon - Inteligenta artificiala 7


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Multiple intelligences
 Psychologist Howard Gardner identified 7 distinct
types of intelligence (Frames of Mind, 1983), and
more recently another 2 (1995)
 Linguistic
 The ability to communicate well, both orally and in writing,
perhaps in several languages

 Which word does not belong to the group?


 Tree, Mouse, Man, Paper

Florin Leon - Inteligenta artificiala 8


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Multiple intelligences
 Logical-mathematical
 The ability to learn higher mathematics, to handle complex
logical arguments
 Manifested in solving arithmetic problems or in strategy
games

 What is the next number in the series?


 3, 4, 6, 8, 12, 14, 18, 20, 24, 30, 32, 38, ?

Florin Leon - Inteligenta artificiala 9


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Multiple intelligences
 Spatial
 The ability to accomplish tasks requiring three-dimensional
visualization
 Manifested in solving spatial geometry problems, finding
the path in a maze, the talent for drawing or building Lego
shapes

 What image is next? ?

Florin Leon - Inteligenta artificiala 10


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Multiple intelligences
 Bodily-kinesthetic
 The ability to use one’s physical body well, to identify and process
physical information, e.g. to “feel” the rhythm of a dance
 Musical
 The ability to learn, perform, and compose music, e.g. to “have an
ear” for music, to perceive and distinguish sounds that seem to be
the same for other people
 Interpersonal
 The ability to sense the others’ feelings and be in tune with others,
e.g. leadership, ease of communication, empathy (to understand
the others’ feelings)
 Intrapersonal
 Self-awareness, the ability to know one’s own feelings and
possibilities
Florin Leon - Inteligenta artificiala 11
https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Multiple intelligences
 Naturalistic
 The ability to recognize patterns in nature, understand

different species, and classify natural objects: biologists,


naturalists
 Existential
 The ability to handle philosophical issues of life. People who

learn best by thinking about analytical questions: readers,


writers, religious speakers, philosophers
 2016: in an interview, Gardner mentioned that he thinks about
including the pedagogical intelligence

 Every cognitive area may have its own type of intelligence?...

Florin Leon - Inteligenta artificiala 12


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Types of intelligence
1. Linguistic 6. Interpersonal
2. Logical-mathematical 7. Intrapersonal
3. Spatial 8. Naturalistic
4. Bodily-kinesthetic 9. Existential
5. Musical 10. Pedagogic?

 Classical IQ tests usually address the first 3 types

Florin Leon - Inteligenta artificiala 13


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Intelligence: what it needs
 Perception
 The ability to make complex observations about the environment
 Information processing
 The ability to handle and transform information, all forms of reasoning
 Memory
 The ability to store and reuse information
 Learning
 The ability to develop new knowledge and capabilities and use the
accumulated experience
 Adaptability
 The ability to flexibly adapt one’s behaviour to current situations
 Input-output process: get – filter – store – generalize – act

Florin Leon - Inteligenta artificiala 14


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Intelligence: what it does
 The ability to use limited resources, including time,
in an optimal way in order to achieve goals
 Greater intelligence means reaching objectives more
often or quicker
 The ability to solve new problems or to solve
problems in new, more efficient ways
 The ability to adapt to the environment by changing
oneself or the environment

Florin Leon - Inteligenta artificiala 15


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Discussion
 Humans are less physically adapted; they had to
compensate
 Intelligence is a potential

 Difference types of intelligence: octopuses


 Evolutionary intelligence of the species: cockroaches,
tardigrades

Florin Leon - Inteligenta artificiala 16


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Intelligence
 Intelligence is a measure of the capability
to achieve goals (to solve problems) in a
complex, dynamic environment

Florin Leon - Inteligenta artificiala 17


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Introduction to Artificial
Intelligence
1. Overview
2. Intelligence and artificial intelligence
3. Definitions of artificial intelligence
4. History of artificial intelligence
5. Subfields and applications
6. Conclusions

Florin Leon - Inteligenta artificiala 18


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Approach #1
Acting humanly
 AI is: “The art of creating machines that
perform functions that require intelligence
when performed by people” (Kurzweil)
 It can be tested by the Turing test (1950)

Florin Leon - Inteligenta artificiala 19


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
The Turing test
 A human judge engages in a natural
language conversation with two other
parties, one a human and the other a
machine
 If the judge cannot reliably tell which
is which, then the machine is said to
pass the test

Florin Leon - Inteligenta artificiala 20


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
In practice
 It needs:
 Natural language processing
 Knowledge representation
 Automated reasoning
 Machine learning

Florin Leon - Inteligenta artificiala 21


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
The Chinese room
 Searle (1980), counter-argument against defining intelligence by
the results of the Turing test
 A person uses the set of rules of a computer that passes the
Turing test to answer questions in Chinese
 The person doesn’t know Chinese
 A program that manipulates symbols, even if it behaves
intelligently, does not understand and has no mental states

Florin Leon - Inteligenta artificiala 22


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Approach #2
Thinking humanly
 AI is: “[The automation of] activities that we
associate with human thinking, activities such
as decision-making, problem solving,
learning…” (Bellman)
 The goal is to build systems that function
internally in some way similar to the human
mind

Florin Leon - Inteligenta artificiala 23


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Cognitive modelling
 Cognitive science tries to model the human mind based on
experimentation
 The cognitive modelling approach to AI assumes intelligent action
while internally mimicking the human mind
 We should know how the human mind works
 Introspection
 Psychological experiments
 Programs should not only solve problems, but solve them in the
same way as people do
 Minsky, The Society of Mind (1988)
 The mind is a collection of agents representing different, possibly
concurrent processes

Florin Leon - Inteligenta artificiala 24


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Approach #3
Thinking rationally
 AI is: “The study of the computations that make it possible
to perceive, reason and act” (Winston)
 Using logic to make complex decisions
 Aristotle: syllogisms (4th century B.C.)
 Rigorous formalization, power of representation and
reasoning
 We can find the truth by manipulating symbols
 How can knowledge be represented logically, and how
can a system make deductions?
 Difficulties: uncertain knowledge, informal knowledge
(e.g., “I think I love you”), learning from noisy data

Florin Leon - Inteligenta artificiala 25


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Approach #4
Acting rationally
 AI is: “...concerned with the automation of intelligent
behaviour” (Luger, Stubblefield)
 Acting rationally means doing the right thing
 Emphasis is on behavior
 Inference is useful, but not always necessary
 Reflex actions can be rational

Florin Leon - Inteligenta artificiala 26


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Acting rationally
 Acting rationally is emphasized by most AI today
 Rational agents
 In solving actual problems, the result is what
really matters
 Rather than imitating humans trying to solve
hard problems, just try to solve hard problems

Florin Leon - Inteligenta artificiala 27


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Difference in approaches
 Thought vs. behavior
 Human vs. rational

 Behavior is observable and more scientifically


testable than thought
 Rationality is clearly defined: optimal actions
 In some complex environments, it may not be
possible to always find the optimal actions

Florin Leon - Inteligenta artificiala 28


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Natural and artificial intelligence
 Electronic computers have a different architecture
than biological systems
 They have each different types of performance
 Language recognition vs. complex computations
 Common sense is an immense, hidden quantity of
implicit knowledge accumulated from experience
 The ultimate challenge for building a human-like machine
 This knowledge cannot be programmed, it must be learned

Florin Leon - Inteligenta artificiala 29


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
General artificial intelligence
 Potential directions that should be merged
into a single mechanism:
 Connectionist methods (deep networks)
 Can learn from data, possibly noisy
 Typical application: image recognition
 Symbolic methods (logic methods, ontologies)
 Can combine knowledge, can reason
 Typical application: automated theorem proving, in the
sense of applying general knowledge to particular cases

Florin Leon - Inteligenta artificiala 30


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
General artificial intelligence
 Potential directions that should be merged
into a single mechanism:
 Connectionist methods (deep networks)
 Symbolic methods (logic methods, ontologies)
 Probabilistic methods (Bayesian networks)
 Evolutionary methods (genetic/evolutionary
algorithms)
 Analogy methods (transfer learning)

Florin Leon - Inteligenta artificiala 31


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Introduction to Artificial
Intelligence
1. Overview
2. Intelligence and artificial intelligence
3. Definitions of artificial intelligence
4. History of artificial intelligence
5. Subfields and applications
6. Conclusions

Florin Leon - Inteligenta artificiala 32


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
The beginnings
 1943: first model of an artificial neuron
(McCulloch & Pitts)
 1949: neural learning (Hebb)
 1950, 1951: studies about chess games
(Shannon, Turing)
 1951: SNARC, the first neural computer
(Minsky & Edmonds)

Florin Leon - Inteligenta artificiala 33


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
The birth of AI
 1956: “Artificial Intelligence” term: McCarthy, Dartmouth College
workshop
 Logical Theorist (Newell & Simon)
 Early enthusiasm
 1957: General Problem Solver
 1958: Lisp, time sharing, theoretical Advice Taker (McCarthy)
 1959: Geometry Theorem Prover
 1965: resolution method
 Checkers program that learns from playing
 Calculus program
 “Microworlds”
 Predictions that AI would eventually do almost anything

Florin Leon - Inteligenta artificiala 34


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Perceptrons
 Rosenblatt (1957)
 Linear classifier
 Perceptron convergence theorem (1962)
 Training algorithm

Only linearly separable problems


Florin Leon - Inteligenta artificiala 35
https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
A dose of reality
 The programs that worked well for the microworlds failed when
applied to real-life problems
 Solving in principle is not the same as solving in practice
 Combinatorial explosion
 Automatic translation from Russian to English
 The spirit is willing but the flesh is weak
 The vodka is good but the meat is rotten
 Minsky and Papert’s Perceptrons (1969)
 Non-linearly separable problems (e.g. XOR) cannot be modelled by
a perceptron
 The Lighthill Report (1973) cancelled AI funding in UK except
for three universities
Florin Leon - Inteligenta artificiala 36
https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Knowledge-based systems
 Dendral, the first expert system (1965)
 Used for spectrometry

 Later versions included the idea of Advice Taker:


separating knowledge (rules) from the program
that performs reasoning
Florin Leon - Inteligenta artificiala 37
https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Eliza
 The computer therapist (Weizenbaum & Colby, 1966)

Florin Leon - Inteligenta artificiala 38


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Other achievements
 1965: Fuzzy logic (Zadeh)
 1968: Knowledge representation with semantic networks
(Quillian)
 1970: Prolog (Colmerauer)
 1970: MYCIN expert system (Shortliffe et al.)
 Written in Lisp, 450 rules
 Blood infections, explanations (“why?”), confidence factors
 EMYCIN (1979), expert system shell
 1973: Genetic algorithms (Holland)
 1975: Knowledge representation with frames (Minsky)
 1980: Prospector, detection of mineral deposits
Florin Leon - Inteligenta artificiala 39
https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
AI becomes an industry
 1978: XCON, commercial expert system for the
configuration of VAX computers
 $40 million savings a year in 1986
 750 rules in 1979, 5500 rules in 1995

Florin Leon - Inteligenta artificiala 40


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Success stories: the 80s
 1984: The Cyc project (Lenat), collection of common sense
knowledge. In 2006: 47,000 concepts, 306,000 facts
 1985: Bayesian networks (Pearl), probabilistic reasoning
 1986: The back-propagation algorithm for training multilayer
perceptron neural networks (Rumelhart, Hinton & Williams),
actually discovered in 1969 (Bryson & Ho)
 1987: SOAR (Newell, Laird & Rosenbloom)
 1988: The Society of Mind (Minsky)
 1988: HiTech (CMU) defeats chess grandmaster Denker

Florin Leon - Inteligenta artificiala 41


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
NetTalk
 Neural network that learns to read texts in English
(Sejnowski & Rosenberg, 1986)
 309 neurons, 18,629 connections

Florin Leon - Inteligenta artificiala 42


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Success stories: the 90s
 1991: During the Gulf war, the DART system is used to
optimize and schedule the transportation of supplies and
personnel involving over 50,000 vehicles and people
 1995: TD-Gammon (Tesauro), backgammon
 1995: ALVINN, autonomous vehicle (4585 km across USA)
 1995: A.L.I.C.E. chatterbot
 1996: Robbins conjecture from Boolean logic, unsolved for
decades, is proven with an automated theorem prover

Florin Leon - Inteligenta artificiala 43


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Robot soccer

Florin Leon - Inteligenta artificiala 44


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Chess
 1997: Deep Blue, developed by IBM, defeats world
champion Garry Kasparov

Florin Leon - Inteligenta artificiala 45


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Expressive robots
 Cog Project  Kismet
(Brooks, MIT) (Breazeal, MIT)

Florin Leon - Inteligenta artificiala 46


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Explorers
 2000: Nomad, meteorite search in Antarctica

Florin Leon - Inteligenta artificiala 47


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
“Massive” algorithm (2000)
 Visual effects with a very large number of agents

The Lord of the Rings, 2001-2003

Florin Leon - Inteligenta artificiala 48


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Computer art
 Aaron (Cohen, Kurzweil CyberArt Technologies, 2001)

Florin Leon - Inteligenta artificiala 49


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
And more...
 1999: Real-time expert system for handling routine self-repair
spacecraft tasks (Remote Agent for Deep Space 1)
 2002: Kramchik and Deep Fritz make a draw (latest version,
Fritz 16 was released in 2017)
 2006: Google Translate, now over 100 languages
 2007: Urban Challenge, driverless cars, with a prize of
2 million $ (96 km urban area course)
 2009: Google introduces automatic subtitles in YouTube videos
 2014: “Eugene Goostman”, a program claiming to be a
13 year old Ukrainian boy passes the Turing test, deceiving 33%
of the judges

Florin Leon - Inteligenta artificiala 50


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Answers in natural language
 IBM Watson
 Jeopardy! (2011)

Florin Leon - Inteligenta artificiala 51


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Military applications
 X-47B drone, USA (2011)
 First autonomous aircraft entirely controlled by an AI
 Flight autonomy: 6 hours, 3900 km without refuelling
 Cost: 813 million $

Florin Leon - Inteligenta artificiala 52


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Google car
 Google self-driving car on city streets (2014)

Florin Leon - Inteligenta artificiala 53


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Deep Reinforcement Learning
 Google DeepMind (2015)
 Learned to play 49 Atari 2600 games by only
watching the display and the score

Florin Leon - Inteligenta artificiala 54


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
AI “dreams”
 Google, deep learning (2015)

Florin Leon - Inteligenta artificiala 55


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Image identification
 Wolfram (2015)

Florin Leon - Inteligenta artificiala 56


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Artificial paintings
 Neural artistic style (2015)

Subject (photo) + Style (van Gogh) → Generated painting

Florin Leon - Inteligenta artificiala 57


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
New Rembrandt painting
 “The Next Rembrandt”
project
 Computer analysis of
168,263 painting
fragments from 346
original paintings
 April 2016: new painting
revealed, 3D printed

Florin Leon - Inteligenta artificiala 58


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
AlphaGo
 Google DeepMind
 March 2016: won 4-1 against Lee Sedol, professional go player
of 9-dan rank, prize: 1 million $
 May 2017: won against Ke Jie, the world’s best go player

Florin Leon - Inteligenta artificiala 59


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
AlphaZero
 AlphaGo Zero (2017) learned go by self-play,
without learning from expert human plays
 AlphaGo Zero defeated AlphaGo (100-0)
 AlphaZero (Dec. 2017) learned chess in 4 hours and
defeated Stockfish (28 wins, 72 draws, 0 losses)
 AI has clearly achieved super-human abilities for
virtually all games

Florin Leon - Inteligenta artificiala 60


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Poker
 Libratus, developed at Carnegie Mellon university
 January 2017: defeated four of the best players in the world
 120,000 games/hands in 20 days
 Won the equivalent of 1,766,250 $
 Pluribus (Facebook & Carnegie Mellon University)
 July 2019: defeated top human playes in a multi-player
format (5 people + 1 AI)
 Won the equivalent of over 1 million $

Florin Leon - Inteligenta artificiala 61


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Google Duplex
 Conversations in natural language (2018)

Florin Leon - Inteligenta artificiala 62


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
The Atlas robot
 Boston Dynamics (2018)

Florin Leon - Inteligenta artificiala 63


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Text generation
 GPT-2 (OpenAI, 2019)
 https://talktotransformer.com/

Florin Leon - Inteligenta artificiala 64


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
In România
 Antivirus software
 BitDefender, Kaspersky, RAV (now in Windows Defender)
 Classification and clustering methods
 GECAD ePayment
 Detection online payment fraud
 Procter & Gamble
 Data mining for sale optimization
 Ropardo
 Production optimization solutions, business intelligence
 Research institutes
 Automatic translation for Romanian
 Traffic flow optimization

Florin Leon - Inteligenta artificiala 65


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Introduction to Artificial
Intelligence
1. Overview
2. Intelligence and artificial intelligence
3. Definitions of artificial intelligence
4. History of artificial intelligence
5. Subfields and applications
6. Conclusions

Florin Leon - Inteligenta artificiala 66


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
AI principles
 Bounded rationality implies oportunistic search
 A physical symbol system has the necessary and
sufficient means for general intelligent action
 The magical number 70.000 ± 20.000
 Search compensates for the lack of knowledge
 Knowledge compensates for the lack of search

Florin Leon - Inteligenta artificiala 67


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
AI subfields Computer
vision

Speech
recognition

Natural
language
processing

Robotics
Florin Leon - Inteligenta artificiala 68
https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
AI subfields Fuzzy logic Neural
Reasoning networks

Game
playing

Genetic
Search
algorithms
heuristics

Expert
systems
Artificial
life (?) Machine
learning Intelligent
agents

Florin Leon - Inteligenta artificiala 69


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm
Conclusions
 Intelligence is a measure of the capability to achieve goals (to
solve problems) in a complex, dynamic environment
 Intelligence is a collection of strategies used to optimize the
interaction with the environment
 There are AI approaches based on thought or behaviour,
modelling the human mind or rationality
 AI has made substantial progress in:
 Pattern recognition and learning
 Planning and reasoning problems
 More recently, image and language recognition
 Many problems are still unsolved
 AI is an interesting research field!

Florin Leon - Inteligenta artificiala 70


https://sites.google.com/view/iafii/home - http://florinleon.byethost24.com/curs_ia_info.htm

You might also like