Introduction to Artificial Intelligence

Lecture 0: Artificial Intelligence

Prof. Gilles Louppe
g.louppe@uliege.be

Today

Course outline
Introduction to Artificial Intelligence
Intelligent agents

Artificial intelligence

"With artificial intelligence we are summoning the demon" -- Elon Musk

???

Triggers a rich imagination fueled by science-fiction.

"We're really closer to a smart washing machine than Terminator" -- Fei-Fei Li, Director of Stanford AI Lab.

???

The reality is quite different...

.width-25.circle[] .width-25.circle[]

]

Edsger Dijkstra: .italic[What do you work on?]
Leslie Valiant (very proudly): .italic[Artificial Intelligence.]
Edsger Dijkstra: .italic[Why don't you work first on the .bold["Intelligence"] part?]

???

-> No more "artificial"

.width-30.circle.center[]

.italic[".bold[What is intelligence, anyway?] It is only a word that people use to name those unknown processes with which our brains solve problems we call hard. But whenever you learn a skill yourself, you're less impressed or mystified when other people do the same.

This is why .bold[the meaning of "intelligence" seems so elusive]: it describes not some definite thing but only the momentary horizon of .bold[our ignorance about how minds might work]. It is hard for scientists who try to understand intelligence to explain precisely what they do, since our working definitions change from year to year. But it is not at all unusual for sciences to aim at moving targets." -- Marvin Minsky]

???

-> No more "intelligence"

A definition?

Artificial intelligence is the science of making machines or programs that: .center.grid[ .kol-1-4[] .kol-1-4[ .caption[Think like people] .width-100[] ] .kol-1-4[ .caption[Think rationally] .width-100[]] ] .grid[ .kol-1-4[] .kol-1-4[ .caption[Act like people] .width-100[] ] .kol-1-4[ .caption[Act rationally] .width-100[] ] ]

???

Textbook definition(s) from AIMA.

Acting humanly

The Turing test

A computer passes the Turing test (aka the Imitation Game) if a human operator, after posing some written questions, cannot tell whether the written responses come from a person or from a computer.

.grid[ .kol-2-3[ .width-80.center[
] ] .kol-1-3.center[ .width-100.circle[] .caption[Can machines think?
(Alan Turing, 1950)] ] ]

???

The Turing test is an operational definition of intelligence.

An agent would not pass the Turing test without the following requirements:

natural language processing
knowledge representation
automated reasoning
machine learning
computer vision (total Turing test)
robotics (total Turing test)

Despite being proposed almost 70 years ago, the Turing test is still relevant today.

Limitations of the Turing test

The Turing test tends to focus on human-like errors, linguistic tricks, etc.

However, it seems more important to study the principles underlying intelligence than to replicate an exemplar.

.center.width-80[]

Aeronautics is not defined as the field of making machines
that fly so exactly like pigeons that they can fool even other pigeons.

Thinking humanly

Cognitive science

Study of the human mind and its processes.

The goal of cognitive science is to form a theory about the structure of the mind, summarized as a comprehensive computer model.
It includes language, problem-solving, decision-making and perception.
A cognitive architecture usually follows human-like reasoning and can be used to produce testable predictions (time of delays during problem solving, kinds of mistakes, learning rates, etc).

???

Grew out of psychology.

The modern SOAR cognitive architecture.

???

The SOAR architecture is both:

a theory of cognition
a computational implementation of that theory

Neurobiology and neuroscience

Study of the anatomy and physiology of neural tissue.

Neurobiology is concerned with the the anatomy and physiology of the brain, from major structures down to neurons and molecules.
Neuroscience adds to that the study of how the brain works, mechanistically, functionally, and systematically to produce observable behavior.

.center.width-50[] .caption[Can we build a computer model of the brain?]

???

Grew out of clinical neurology and neurobiology.

Limitations of cognition and neuroscience for AI

In linguistics, the argument of poverty of the stimulus states that children do not receive sufficient input to generalize grammatical rules through linguistic input alone.
- A baby hears too few sentences to deduce the grammar of English before he speaks correctly.
(Controversial) Therefore, humans must be biologically pre-wired with innate knowledge for representing language. ] .kol-1-3.center[ .circle.width-100[] .caption[How do we know what we know? (Noam Chomsky, 1980)] ] ]

Therefore, it may not be possible to implement a fully functioning computer model of the human mind without background knowledge of some sort.

???

This is a huge technical obstacle, as accessing this knowledge would require reverse-engineering the brain.

Thinking rationally

The logical approach

The rational thinking approach is concerned with the study of irrefutable reasoning processes. It ensures that all actions performed by a computer are formally provable from inputs and prior knowledge.
The "laws of thought" were supposed to govern the operation of the mind. Their study initiated the field of logic and the logicist tradition of AI (1960-1990).

The Zebra puzzle

There are five houses.
The English man lives in the red house.
The Swede has a dog.
The Dane drinks tea.
The green house is immediately to the left of the white house.
They drink coffee in the green house.
The man who smokes Pall Mall has birds.
In the yellow house they smoke Dunhill.
In the middle house they drink milk. ] .kol-1-2[
The Norwegian lives in the first house.
The man who smokes Blend lives in the house next to the house with cats.
In a house next to the house where they have a horse, they smoke Dunhill.
The man who smokes Blue Master drinks beer.
The German smokes Prince.
The Norwegian lives next to the blue house.
They drink water in a house next to the house where they smoke Blend. ] ]

Who owns the zebra?

select([A|As],S):- select(A,S,S1),select(As,S1).
select([],_).

next_to(A,B,C):- left_of(A,B,C) ; left_of(B,A,C).
left_of(A,B,C):- append(_,[A,B|_],C).

zebra(Owns, HS):-  % color,nation,pet,drink,smokes
      HS =    [ h(_,norwegian,_,_,_), _,  h(_,_,_,milk,_), _, _],
      select( [ h(red,englishman,_,_,_),  h(_,swede,dog,_,_),
                h(_,dane,_,tea,_),        h(_,german,_,_,prince) ], HS),
      select( [ h(_,_,birds,_,pallmall),  h(yellow,_,_,_,dunhill),
                h(_,_,_,beer,bluemaster) ],                         HS),
      left_of(  h(green,_,_,coffee,_),    h(white,_,_,_,_),         HS),
      next_to(  h(_,_,_,_,dunhill),       h(_,_,horse,_,_),         HS),
      next_to(  h(_,_,_,_,blend),         h(_,_,cats, _,_),         HS),
      next_to(  h(_,_,_,_,blend),         h(_,_,_,water,_),         HS),
      next_to(  h(_,norwegian,_,_,_),     h(blue,_,_,_,_),          HS),
      member(   h(_,Owns,zebra,_,_), HS).

:- ?- time(( zebra(Who, HS), maplist(writeln,HS), nl, write(Who), nl, nl, fail
             ; write('No more solutions.') )).

Output =

h(yellow,norwegian, cats,  water, dunhill)
h(blue,  dane,      horse, tea,   blend)
h(red,   englishman,birds, milk,  pallmall)
h(green, german,    zebra, coffee,prince)
h(white, swede,     dog,   beer,  bluemaster)

german

No more solutions.
% 5,959 inferences, 0.000 CPU in 0.060 seconds (0% CPU, Infinite Lips)

Limitations of logical inference

Representation of informal knowledge is difficult.
Hard to define provable plausible reasoning.
Combinatorial explosion (in time and space).
Logical inference is only a part of intelligence. It does not cover everything:
- e.g., might be no provably correct thing to do, but still something must be done;
- e.g., reflex actions can be more successful than slower carefully deliberated ones.

Acting rationally

A rational agent acts so as to achieve the best expected outcome.

Correct logical inference is just one of several possible mechanisms for achieving this goal.
Perfect rationality cannot be achieved due to computational limitations!
- The amount of reasoning is adjusted according to available resources and importance of the result.
The brain is good at making rational decisions but not perfect either.

Rationality only concerns what decisions are made (not the thought process behind them, human-like or not).

Goals are expressed in terms of the performance or utility of outcomes. Being rational means maximizing its expected performance. The standard of rationality is general and mathematically well defined.

In this course, Artificial intelligence = Maximizing expected performance ]

AI prehistory

Philosophy: logic, methods of reasoning, mind as physical system, foundations of learning, language, rationality.
Mathematics: formal representation and proof, algorithms, computation, (un)decidability, (in)tractability, probability.
Psychology: adaptation, phenomena of perception and motor control, psychophysics.
Economics: formal theory of rational decisions.
Linguistics: knowledge representation, grammar.
Neuroscience: plastic physical substrate for mental activity.
Control theory: homeostatic systems, stability, simple optimal agent designs.

A short history of AI

1940-1950: Early days

1943: McCulloch and Pitts: Boolean circuit model of the brain.
1950: Turing's "Computing machinery and intelligence".

1950-1970: Excitement and expectations

1950s: Early AI programs, including Samuel's checkers program, Newell and Simon's Logic Theorist and Gelernter's Geometry Engine.
1956: Dartmouth meeting: "Aritificial Intelligence" adopted.
1958: Rosenblatt invents the perceptron.
1965: Robinson's complete algorithm for logical reasoning.
1966-1974: AI discovers computational complexity.

.width-60.center[]

The Darthmouth workshop (1956)

.italic[The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.]

1970-1990: Knowledge-based approaches

1969: Neural network research almost disappears after Minsky and Papert's book.
1969-1979: Early development of knowledge-based systems.
1980-1988: Expert systems industrial boom.
1988-1993: Expert systems industry busts (AI winter).

1990-Present: Statistical approaches

1985-1995: The return of neural networks.
1988-: Resurgence of probability, focus on uncertainty, general increase in technical depth.
1995-2010: New fade of neural networks.
1995-: Complete intelligent agents and learning systems.
2000-: Availability of very large datasets.
2010-: Availability of fast commodity hardware (GPUs).
2012-: Resurgence of neural networks with deep learning approaches.

What can AI do at present?

Translate spoken Chinese to spoken English, live?
Answer multi choice questions, as good as an 8th grader?
Converse with a person for an hour?
Play decently at Chess? Go? Poker? Soccer?
Buy groceries on the web? in a supermarket?
Prove mathematical theorems?
Drive a car safely on a parking lot? in New York?
Perform a surgery?
Identify skin cancer better than a dermatologist?
Write a funny story?
Paint like Vangogh? Compose music?
Show common sense?

Speech translation and synthesis (2012)

Speech synthesis and question answering (Google, 2018)

Playing Atari games

Beat the best human Go players (2016)

Beat teams of human players at real-time strategy games (Dota 2) (2018)

Playing soccer (2018)

Learning to walk (2017)

???

Single algorithm for learning! Nothing is hardcoded.

Similar to a baby learning to walk.

Driving a car (NVIDIA, 2016)

... and preventing accidents.

Semantic segmentation (2017)

Generating image descriptions (2015)

Detecting skin cancer (2017)

Folding laundry (2010)

Compose music (NVIDIA, 2017)

Learning to sort waste, before training
(Norman Marlier, ULiège, 2018)

Learning to sort waste, after training
(Norman Marlier, ULiège, 2018)

What is missing?

Intelligence is not just about pattern recognition, which is something most of these works are based on.

It is about modeling the world:

explaining and understanding what we see;
imagining things we could see but haven't yet;
problem solving and planning actions to make these things real;
building new models as we learn more about the world.

The end.

References

Turing, Alan M. "Computing machinery and intelligence." Mind 59.236 (1950): 433-460.
Newell, Allen, and Herbert Simon. "The logic theory machine--A complex information processing system." IRE Transactions on information theory 2.3 (1956): 61-79.
Chomsky, Noam. "Rules and representations." Behavioral and brain sciences 3.1 (1980): 1-15.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

lecture0.md

lecture0.md

Introduction to Artificial Intelligence

Today

Artificial intelligence

A definition?

Acting humanly

The Turing test

Limitations of the Turing test

Thinking humanly

Cognitive science

Neurobiology and neuroscience

Limitations of cognition and neuroscience for AI

Thinking rationally

The logical approach

The Zebra puzzle

Limitations of logical inference

Acting rationally

AI prehistory

A short history of AI

1940-1950: Early days

1950-1970: Excitement and expectations

The Darthmouth workshop (1956)

1970-1990: Knowledge-based approaches

1990-Present: Statistical approaches

What can AI do at present?

What is missing?

References

Files

lecture0.md

Latest commit

History

lecture0.md

File metadata and controls

Introduction to Artificial Intelligence

Today

Artificial intelligence

A definition?

Acting humanly

The Turing test

Limitations of the Turing test

Thinking humanly

Cognitive science

Neurobiology and neuroscience

Limitations of cognition and neuroscience for AI

Thinking rationally

The logical approach

The Zebra puzzle

Limitations of logical inference

Acting rationally

AI prehistory

A short history of AI

1940-1950: Early days

1950-1970: Excitement and expectations

The Darthmouth workshop (1956)

1970-1990: Knowledge-based approaches

1990-Present: Statistical approaches

What can AI do at present?

What is missing?

References