UCCD2063
Artificial Intelligence Techniques
Unit 1:
Introduction to Artificial Intelligence (AI)
Outline
• Introduction
• What is AI
• Applications of AI
• Limitations of AI
What is AI?
3
What is AI?
▪ Sci-Fi AI :
[Link]
▪ Where are we now?
• Object detection
[Link]
• Human pose and face detection
[Link]
[Link]
• Semantic Segmentation
[Link]
• Robotics
[Link]
4
What is AI?
“ Artificial intelligence (AI) is the science and
engineering of making intelligent machines,
especially intelligent computer programs.”
[John McCarthy, father of AI]
“ AI is a computer program/machine that can solve
problem like a human.”
5
What is Human Intelligence?
Problem Solving Learning Reasoning
Perception (Vision) Language Understanding
6
History of AI
▪ 1950—70: Early AI, great ▪ 1980s-late—90s: First AI Winter
expectations
• 1950: Alan Turing developed the Turing ▪ 1990-2010: Machine Learning
test to test machine intelligence • Many learning models: neural
• 1950s: Early AI programs, including network, SVM, decision tree, knn,...
Samuel's checkers program, Gelernter's • Use of probability to model
Geometry Engine uncertainty
• 1956: Dartmouth meeting: “Artificial • AI Spring!
Intelligence” adopted
• 1957: Frank Rosenblatt invented the
perceptron algorithm ▪ 2012-present: Rise of Deep Learning
• 2012: AlexNet beats previous
• benchmark on the ImageNet
▪ 1960s-late—70s: First AI Winter competition
• Neural networks gets deeper
▪ 1970—90: Knowledge-based AI • Availability of very large datasets and
• 1969—79: Early development of fast GPU processor
knowledge-based systems • Data will drive future discoveries and
• 1980—88: Expert systems industry alleviate the complexity in AI
booms, AI becomes an industry
• 1986: “Backpropogation” algorithm is
able to train multi-layer perceptrons
7
History of AI
8
How Does AI work?
▪ Example: Detect credit card fraud
• Collect past transaction data (normal + fraudulent)
• Define/learn dominant features:
− Transaction time
− Transaction amount
− Transaction location
− Type of purchase
− ...
• The AI algorithm learns what feature combinations suggest
unusual activity from the data
• Use the model to predict whether a new transaction is
normal or fraudulent.
9
Evolution of AI Technique
Rule-based
Expert System
Manual
Classic Machine
Learning
Learn from features
Deep
Learning
Learn directly from data
10
Machine Learning vs Deep Learning
Feature
Vector
Note: This course covers traditional machine learning and AI
techniques such as search and statistical inference. Deep
learning is covered in UCCD3074.
11
The AI Technology Landscape
12
13
Applications of AI – Computer Vision
Face Detection & Recognition
▪ Input data: images
▪ Output: location of all faces in
the image
Visual Object Detection
▪ Input data: Images
▪ Output: presence/absence of
particular objects in the
image
Video Surveillance
▪ Input data: video
▪ Output:
• Human tracking
• Human activity detection
• Abnormal event detection 14
Applications of AI – Image Processing
Medical Imaging
▪ Input data: images
▪ Output: presence/absence of
Tumor in the image
Image Processing
▪ Input data: Blur/noisy image
▪ Output: clear image
Dehazing Impainting 15
Application of AI – Voice Recognition
Speech Technologies (e.g., Siri, Alexa)
▪ Input data: Audio waveform
▪ Output:
• Text characters
• Automatic speech recognition (ASR)
16
Application of AI – Natural Language Processing
Language Processing Technologies
▪ Input data: text
▪ Output:
▪ Machine Translation
▪ Question answering
▪ Text classification, spam filtering, etc.
17
Application of AI – Autonomous Cars
▪ Autonomous Cars
• Input data: sensor data, video camera, LIDAR system
• Output: Driving Control (steering rotation, brake level,
accelerator level)
18
Application of AI – Games
▪ Games!
Chess game (1997): Kasparov
vs. Deep Blue
Jeopardy! (2011): Humans vs
IBM Watson
Go (2016): Lee Sedol vs.
Google Alpha Go
19
AI Industry
20
Limitations of AI
"We're very far from building truly intelligent
machines.
All you’re seeing now — all these feats of AI like self-
driving cars, interpreting medical images, beating the
world champion at Go and so on — these are very
narrow intelligences,
and they’re really trained for a particular purpose.
Yann LeCunn, head of AI @
Facebook [link] They’re situations where we can collect a lot of
data."
We’re very far from having machines that can learn the most basic things about the world
in the way humans and animals can do. Like, yes, in particular areas machines have
superhuman performance, but in terms of general intelligence, current AI is not even
close to a rat.
21
Limitations of AI
What Artificial Intelligence Can and
Can’t Do Right Now [link]
▪ Despite AI’s breadth of impact, the
types of it being deployed are still
extremely limited.
Andrew Ng, Adjunct Professor ▪ Types of AI’s being deployed is still
@ Stanford University
limited to A→B
22
What AI can and cannot do
A (image) → B (car)
A (image) → B (??)
23
Three Types of AI
Machine Machine Machine
Learning Intelligence Consciousness
24
Where human beat machines
▪ Creative endeavors:
• Entrepreneurship, scientific discovery, creative writing, etc.
▪ Social interactions:
• AI lack of emotions, not sensitive to the need of others
• Not good for any jobs requiring emotions: Teachers, nurses,
caretakers, negotiator, managers, etc.
▪ Physical finesse and mobility:
• Picking up a pen, dancing, swimming, hiking mountains,
gardening, housekeeping
▪ Problems that is hard to describe:
• The true challenge for AI are problems that are easy for people
to perform but hard for people to describe – problems that we
solve intuitively, that feel automatic
25
Next:
The Fundamentals of Machine Learning
