INTRODUCTION TO

ARTIFICIAL
INTELLIGENCE

Nguyễn Ngọc Thảo

[email protected]
Outline
• What is Artificial Intelligence (AI)?
• The foundations of AI
• A brief history of AI
• AI applications in various fields
• What are we going to learn?

2
What is AI?
AI: A dream for everyone

4
AI Innovations: Atlas Robot

Source: YouTube video

5
AI Innovations: DALL·E 2

Source: YouTube video 6

Tesla’s Autopilot: Automated driver

Source: Autopilot | Tesla 7

Amazon Go: A store of the future

Source: YouTube video 8

AI Innovations: Deep Blue – AlphaGo

AlphaGo vs. Lee Sedol

(03/2016)

Deep Blue vs. Kasparov

(02/1996 and 05/1997)

9
The complexity of Chess and GO

Source: YouTube video 10

AI Innovations: OpenAI Five

Source: OpenAI Five

11
Intelligence vs. Artificial Intelligence
Intelligence includes the capacity for logic, understanding,
learning, reasoning, creativity, and problem solving, etc.

Artificial intelligence (AI) attempts not just to understand

but also to build intelligent entities.
12
The field of Artificial Intelligence
• AI is one of the newest fields in science and engineering.
• Work started in earnest soon after World War II
• The name was coined at a conference at Dartmouth College in
1956.

John McCarthy Marvin Minsky Allen Newell Arthur Samuel Herbert Simon
(1927 – 2011) (1927 – 2016) (1927 – 1992) (1901 – 1990) (1916 – 2001)

13
The field of Artificial Intelligence
• AI research builds intelligent entities that simulate humans
in different aspects.

✓ Thinking: learning, planning, and

refining knowledge

✓ Perception: see, hear, feel, etc.

✓ Communication in natural languages

✓ Manipulation and moving objects

14
What is Artificial Intelligence?

15
What is Artificial Intelligence?
Thought processes and reasoning

Systems that Systems that

Rationality
think think
Humans

like humans rationally

Systems that Systems that

act act
like humans rationally

Behavior 16
Systems that act like humans
• The Turing Test approach (Alan Turing, 1950)

A computer passes the test if a human interrogator, after posing several written
questions, cannot tell whether the written responses come from a person or from a
computer.

17
Systems that act like humans

• Problems with the Turing Test

ntelligent
nintelligent
Turing ehavior
human
test humans
ehavior
don t do

• Variations
• Reverse Turing Test: CAPTCHA
• Total Turing Test: additionally examine the perceptual (computer
vision) and the objects manipulation (robotics) abilities of the subject.
18
Turing Test: Variations
• Reverse Turing Test: CAPTCHA

19
Turing Test: Variations
• Total Turing Test: additionally examine the perceptual
(computer vision) and the objects manipulation (robotics)
abilities of the subject.

Sheep dog
or mop?

20
A better Turing Test?
• AI researchers have devoted little effort to pass the test.
• It is more important to study the underlying principles of
intelligence than to duplicate an exemplar.

Image credit

21
Systems that think like humans
• General Problem Solver – GPS (Newell and Simon, 1961)
• Not merely solve problems correctly
• Compare the trace of its reasoning steps to traces of human subjects
while solving the same problems

• Cognitive Science
• Computer models from AI precise and testable
theories of
• Experimental techniques from psychology the human mind

• These approaches are now distinct from AI

• Share the available theories but do not explain anything resembling
human intelligence
• All share a principal direction

22
Systems that think rationally
• The laws of thought approach
• “Right thinking” = irrefutable reasoning processes
• E.g., Aristotle’s syllogisms provided patterns for argument structures
that always yielded correct conclusions when given correct
premises.

All men are mortal. x.man(x) mortal(x)

Socrates is a man. man(Socrates)
Therefore,
Socrates is mortal. mortal(Socrates)
Aristotle
(381BC – 322BC)
23
Systems that think rationally
• Problems with the logicist approach
• Not all intelligence is mediated by logic behavior
• Solving a problem “in principle” is different from doing in practice
• Both obstacles apply to any attempt to build computational
reasoning systems

24
Systems that act rationally
• The rational agent approach
• Rational behavior = “doing the right thing”,
• “Right thing”: what is expected to maximize goal achievement given
the available information
• An agent is just something that perceives and then acts
𝒇: 𝓟 → 𝓐
• A rational agent acts to achieve the best outcome or, when
there is uncertainty, the best expected outcome.
• Include thinking, inference as a part of being rational agent
• Include more: action without thinking, e.g., reflexes

25
Systems that act rationally
• A behavior is either a reflex action or an intelligent one.
• A reflex action can be rational or not, while an intelligent
action is usually rational.
• An intelligence behavior is usually obtained via a learning process.

A man withdraws his fingers from a hot stove. Two people cross the street
at the crosswalk.
26
Systems that act rationally
• More general than the “laws of thought” approach
• Correct inference is not all of rationality.
• In some situations, there is no provably correct thing to do, but
something must still be done.
• Amenable to scientific development than those based on
human behavior or human thought

27
Major roles and Goals of AI
AI studies the intelligent part concerned with human and
represents those actions using computers.

Goals of AI

28
Foundations
of AI
Research fields related to AI

Control theory Mathematics

and
cybernetics Philosophy

Linguistics Neuroscience

Economics Computer Psychology

Engineering
30
Research fields related to AI
Field Description
Logic, methods of reasoning, mind as physical
Philosophy system, foundations of learning, language,
rationality.
Formal representation and proof, algorithms,
Mathematics computation, (un)decidability, (in)tractability,
probability.

Economics Utility, decision theory, rational economic agents

Neuroscience Neurons as information processing units.

Psychology/ How do people behave, perceive, process
Cognitive Science information, represent knowledge.
Computer Engineering Building fast computers
Design systems that maximize an objective
Control Theory
function over time
Linguistic Knowledge representation, grammar 31
Research fields related to AI

32
Pros and Cons of AI
More powerful and more useful computers
New and improved interfaces
Solve new problems
Better handling of information
Relieve information overload
Conversion of information into knowledge

Increased costs
Difficulty with software development - slow and expensive
Few experienced programmers
33
A brief
history of AI
A brief history of AI
• 1940-1950: Early days
• 1943: McCulloch & Pitts: Boolean circuit model of brain
• 1950: Turing's “Computing Machinery and ntelligence”
• 1950—70: Excitement: Look, Ma, no hands!
• 1950s: Early AI programs, including Samuel's checkers program, Newell & Simon's
Logic Theorist, Gelernter's Geometry Engine
• 1956: Dartmouth meeting: “Artificial ntelligence” adopted
• 1965: Robinson's complete algorithm for logical reasoning
• 1970—90: Knowledge-based approaches
• 1969—79: Early development of knowledge-based systems
• 1980—88: Expert systems industry booms
• 1988—93: Expert systems industry busts: “A Winter”

35
A brief history of AI
• 1990—: Statistical approaches
• Resurgence of probability, focus on uncertainty
• General increase in technical depth
• Agents and learning systems… “A Spring”?
• 2000—: Where are we now?

Source: Stack Exchange

36
AI Applications
Autonomous Planning and Scheduling

Autonomous rovers
Autonomous rovers

Telescope scheduling

Analysis of data 38
Medicine

Classification on
medical images

Have you obtained positive cultures?

Yes.
What type of infection is it?
Primary bacteremia.
When did the symptoms first appear?
May 5
Diagnosis system
I recommend gentamycin using a doze of ...
(e.g., MYCIN)
39
Games and Entertainment

40
What are we
going to learn?
 Back-propagation algorithm
• One of the key supervised learning algorithm used for
training artificial neural networks.
• Goal: Minimize the error between the predicted output and
the actual target output by adjusting the model's weights.

42
Image credit: MLK
Reinforcement learning (RL)
• An agent learns to interact with an environment based on
feedback signals it receives from the environment.

43
RL as a trial-and-error process
• The learner is not told which actions to take, instead he
discovers which actions yield the most reward by trial.

Learning to ride a bike requires trial and

error, much like reinforcement learning.
(Video courtesy of Mark Harris, who says he
is “learning reinforcement” as a parent.)

Image credit: Commoncog

44
Initial performance After 15 mins of training After 30 mins of training

45
Federated learning (FL)
• Privacy-preserving models can be trained in heterogeneous
and distributed networks.

46
Meta-learning (ML)
• Meta-learning focuses on designing models or algorithms
that can learn from and adapt to new tasks more efficiently.
• The model aims to learn a set of skills that can be applied
across a range of tasks, rather than being specialized for a
specific task.

47
Image credit: Cloudera Fast Forward
Few-shot learning (FSL)
• It is about classifying new samples when you have only a
few training samples with labels.
• FSL refers to 𝑁-way-𝐾-shot-classification.
• 𝑁 stands for the number of classes, and 𝐾 for the number of
samples from each class to train on.

𝐾 shots

𝑁 ways 49
One-shot learning (OSL)
• It assesses the similarity and difference between two images,
when only a single sample of each class is available.

50
Zero-shot learning (ZSL)
• This aims to categorize objects from previously unseen
classes without any training examples.

A child would have no problem recognizing a zebra if

• He has seen a horse before, and
• He has read somewhere that a zebra looks like a
horse but has black-and-white stripes.

51