CSC384: Intro to Artificial Intelligence
Instructor: Fahiem Bacchus
– Office D.L. Pratt, Room 398B
– Office Hours: Monday 3:00pm to 4:00pm and Friday 10:00
am to 11:00 am.
Intro to Artificial Intelligence – MWF 1:00 pm – 2:00 pm. GB 119
– Note Fridays will be partly a tutorial, but some lectures will
be given that day.
Fahiem Bacchus, University of Toronto, 2
CSC384: Reference Materials CSC384: Reference Materials
Recommended Textbook: Alternate Book:
Artificial Intelligence: A Modern Approach
Stuart Russell and Peter Norvig Computational Intelligence: A Logical Approach by David
Poole and Alan Mackworth.
- Complete book is available on line!
– Older editions are also useable---but you will http://artint.info/
have to search the text for the relevant
sections (they have renumbered the sections).
– 2 copies on 24hr reserve in the Engineering and
Computer Science Library Online Course:
– Sections most related to the lecture material - Lectures are on line via you-tube (linked from the
will be indicated in the slides course website).
Fahiem Bacchus, University of Toronto, 3 Fahiem Bacchus, University of Toronto, 4
CSC384: Prerequisites CSC384: Website
• Some probability (STA247H/STA255H/STA257H). Some •The course web site:
knowledge of functional programming and logic
programming is useful (CSC324H). http://www.cs.toronto.edu/~fbacchus/csc384Winter/
• This year the course will use Python in the assignments. information,
– Primary source of more detailed information
• You will be responsible for any background material – Check the web site often.
that you don’t know, you will have to learn on your own. – Updates about assignments, clarifications etc. will be
posted only on the web site.
•The course bulletin board:
Will not be monitored.
Fahiem Bacchus, University of Toronto, 5 Fahiem Bacchus, University of Toronto, 6
CSC384: How You Will be Graded Plagiarism
• Course work:
– 3 Assignments (programming, theory, short answers): 10% each • See
– Two Term tests (ap o . 1 hour each): 15% each
o e es s (aprox. ou eac ): 5% eac http://www.cs.toronto.edu/~fpitt/documents/plagiarism.html
– A Final Exam (3hrs): 40% for the meaning of plagiarism, how to avoid it, and the U of T
policies about it.
– You need a minimum of 40% on the Final to pass the course
• Late policy for Assignments: • All assignments are to be done individually!
– Start Early, late assignments will not be accepted.
• You can discuss the assignments with other students, but you
should not give your code (or parts of your code) to other
students. You should not look at another student’s code until
after you have handed in your assignment (and the due date
Fahiem Bacchus, University of Toronto, 7 Fahiem Bacchus, University of Toronto, 8
CSC384: Email Policy CSC384: Important Dates
• I don’t answer questions about course content or the • See the Course Information Sheet.
assignments by email. • The dates for the tests and assignments might in unusual
• I will read short and to the point email. circumstances have to be slightly changed (but hopefully not).
• Come to my office hours, talk to me before or after class
• If you have an unavoidable scheduling conflict we can
arrange a mutually acceptable alternative meeting time.
Fahiem Bacchus, University of Toronto, 9 Fahiem Bacchus, University of Toronto, 10
Are these Intelligent?
Wh t is Artificial Intelligence?
What i A tifi i l I t lli ?
What is Intelligence?
Fahiem Bacchus, University of Toronto, 12
What is Intelligence? Artificial Intelligence
• Webster says:
– The capacity to acquire and apply knowledge.
– The faculty of thought and reason
How to achieve intelligent behavior
• What features/abilities do humans (animals? animate through computational means
objects?) have that you think are indicative or
characteristic of intelligence?
• Abstract concepts, mathematics, language, problem
solving, memory, logical reasoning, planning ahead,
emotions, morality, ability to learn/adapt, etc…
Fahiem Bacchus, University of Toronto, 13 Fahiem Bacchus, University of Toronto, 14
Classical Test of (Human) Intelligence Human Intelligence
• The Turing Test: •Turing provided some very persuasive
– A human interrogator. Communicates with a hidden
subject that is either a computer system or a human.
j p y arguments that a system passing the Turing test
is i t lli t
If the human interrogator cannot reliably decide
whether or not the subject is a computer, the – We can only really say it behaves like a human
computer is said to have passed the Turing test. – Nothing guarantees that it thinks like a human
• Weak Turing type tests:
•The Turing test does not provide much traction
on the question of how to actually build an
See Luis von Ahn, Manuel Blum, Nicholas Hopper, and John Langford.
CAPTCHA: Using Hard AI Problems for Security. In Eurocrypt.
Fahiem Bacchus, University of Toronto, 15 Fahiem Bacchus, University of Toronto, 16
Human Intelligence Human Intelligence
•Is imitating humans the goal? • In general there are various reasons why trying to mimic
humans might not be the best approach to AI:
– Computers and Humans have a very different architecture with
quite different abilities
– Numerical computations
– Visual and sensory processing
– Massively and slow parallel vs. fast serial
Computer Human Brain
•Cons? 1011 neurons
Computational Units 4 CPUs, 109 gates
Storage Units 1010 bits RAM 1011 neurons
1013 bits disk 1014 synapses
Cycle time 10-9 sec 10-3 sec
Bandwidth 1010 bits/sec 1014 bits/sec
Memory updates/sec 1010 1014
Fahiem Bacchus, University of Toronto, 17 Fahiem Bacchus, University of Toronto, 18
Human Intelligence Rationality
•But more importantly, we know very little about •The alternative approach relies on the notion of
how the human brain performs its higher level rationality.
processes Hence, this point of view provides
very little information from which a scientific
understanding of these processes can be built. •Typically this is a precise mathematical notion
of what it means to do the right thing in any
•Nevertheless, Neuroscience has been very particular circumstance. Provides
influential in some areas of AI. For example, in – A precise mechanism for analyzing and
sensing processing etc.
robotic sensing, vision processing, etc understanding the properties of this ideal behavior
we are trying to achieve.
•Humans might not be best comparison?
– A precise benchmark against which we can measure
– Don’t always make the best decisions
the behavior the systems we build.
– Computer intelligence can aid in our decision
Fahiem Bacchus, University of Toronto, 19 Fahiem Bacchus, University of Toronto, 20
Rationality Computational Intelligence
• Mathematical characterizations of rationality have •AI tries to understand and model intelligence as
come from diverse areas like logic (laws of thought) a computational process.
and economics (utility theory how best to act
under uncertainty, game theory how self-interested
agents interact). •Thus we try to construct systems whose
computation achieves or approximates the
• There is no universal agreement about which desired notion of rationality.
notion of rationality is best, but since these notions
are precise we can study them and give exact Science.
•Hence AI is part of Computer Science
characterizations of their properties, good and bad. – Other areas interested in the study of intelligence lie in other
areas or study, e.g., cognitive science which focuses on human
intelligence. Such areas are very related, but their central focus
• We’ll focus on acting rationally tends to be different.
– this has implications for thinking/reasoning
Fahiem Bacchus, University of Toronto, 21 Fahiem Bacchus, University of Toronto, 22
Four AI Definitions by Russell + Norvig Subareas of AI
•Perception: vision, speech understanding, etc.
Like humans Not necessarily like humans
•Machine Learning, Neural networks
Systems that think like
Systems that think rationally
•Natural language processing
•Reasoning and decision making OUR FOCUS
– Knowledge representation
Systems that act like Systems that act rationally
g( g ,p )
– Reasoning (logical, probabilistic)
humans Our focus – Decision making (search, planning, decision theory)
Fahiem Bacchus, University of Toronto, 23 Fahiem Bacchus, University of Toronto, 24
Further Courses in AI What We Cover in CSC384
• Perception: vision, speech understanding, etc.
– CSC487H1 “Computational Vision”
•Search (Chapter 3, 5, 6)
– CSC420H1 “Introduction to Image Understanding” – Uninformed Search (3.4)
• Machine Learning, Neural networks – Heuristic Search (3.5, 3.6)
– CSC321H “Introduction to Neural Networks and Machine Learning”
– Game Tree Search (5)
– CSC411H “Machine Learning and Data Mining”
– CSC412H1 “Uncertainty and Learning in Artificial Intelligence” – Constraint Satisfaction Problems, Backtracking Search (6)
– Engineering courses •Knowledge Representation (Chapter 8, 9)
• Natural language processing
– First order logic for more general knowledge (8)
– CSC401H1 “Natural Language Computing”
– CSC485H1 “Computational Linguistics” – Inference in First-Order Logic (9)
• Reasoning and decision making
– CSC486H1 “Knowledge Representation and Reasoning”
• Builds on this course
Fahiem Bacchus, University of Toronto, 25 Fahiem Bacchus, University of Toronto, 26
What We Cover in CSC384 AI Successes
•Classical Planning (Chapter 10) • Games: chess, checkers, poker, bridge, backgammon…
– Predicate representation of states
• Physical skills: driving a car, flying a plane or helicopter,
y g y g p p
– Planning Algorithms vacuuming...
• Language: machine translation, speech recognition, character
•Quantifying Uncertainty and Probabilistic recognition, …
Reasoning (Chapter 13, 14, 16)
– Uncertainties, Probabilities • Vision: face recognition, face detection, digital photographic
processing, motion tracking, ...
– Probabilistic Reasoning, Bayesian Networks
– Decision Making under Uncertainties, Utilities and Influence • Commerce and industry: page rank for searching, fraud detection,
diagrams trading on financial markets…
Fahiem Bacchus, University of Toronto, 27 Fahiem Bacchus, University of Toronto, 28
Recent AI Successes Degrees of Intelligence
• Building an intelligent system as capable as humans
• Re-integrating the diverse subfields of AI remains an elusive goal.
• Darpa Grand Challenges However,
• However systems have been built which exhibit various
– Goal: build a fully autonomous car that can drive a 240 km specialized degrees of intelligence.
course in the Mojave desert
– 2004: none went further than 12 km
• Formalisms and algorithmic ideas have been identified
– 2005: 5 finished
as being useful in the construction of these “intelligent”
– 2007: Urban Challenge: 96 km urban course (former air force
base) with obstacles, moving traffic, and traffic regulations: 6 • Together these formalisms and algorithms form the
finishers foundation o ou a e p to u de s a d intelligence as
ou da o of our attempt o understand e ge ce
– 2011: Google testing its autonomous car for over 150,000 km on a computational process.
• 2011: IBM Watson competing successfully against two • In this course we will study some of these formalisms and
Jeopardy grand-champions see how they can be used to achieve various degrees
Fahiem Bacchus, University of Toronto, 29 Fahiem Bacchus, University of Toronto, 30
– 1.1: What is AI?
– 2: Intelligent Agents
•Other interesting readings:
– 1.2: Foundations
– 1.3: History
Fahiem Bacchus, University of Toronto, 31