Upload
others
View
36
Download
2
Embed Size (px)
Citation preview
Introduction to Artificial Intelligence
Course Overview and Introduction
CE417: Introduction to Ar tificial Intelligence
Sharif University of Technology
Spring 2012
Soleymani
In which we try to explain why we consider artificial intelligence to be a subject most worthy of study, and in which we try to decide what exactly it is, this being a good thing to decide
before embarking.
Instructor: Mahdeih Soleymani
Email: [email protected]
Lectures: Sun-Tue (15-16:30), Room 201
Website: http://ce.sharif.edu/cources/90-91/2/ce417-1
Course Info
Artificial Intelligence: A Modern
Approach
by Stuar t Russell and Peter Norvig
3 rd Edition, 2009
Text Book
http://aima.cs.berkeley.edu/
Mid Term Exam: 30%
Final Exam: 40%
Homeworks (written & programming) : 15% (+ 5% Optional)
Two quizzes: 15%
Class part icipation & work, project : +X (<= 1 .5 point) Optional
Rules:
If the exams sum < 50%, it would be multiplied in other items.
No late homework is accepted (no exceptions )
Homeworks must be completed individually (although encouraging you to
help each other to learn, assignments must be your own work)
Marking Scheme
Getting a feeling of Ar tificial Intelligence (AI), i ts aims, fields,
abil it ies, open problems
Learning fundamentals of AI
Learning some basic tools for AI and a l ittle experience with
AI
Class Target
Introduction and Agents (chapters 1-2)
Search & game paying (chapters 3-6)
Logical systems & reasoning (chapters 7-9)
Planning (chapter 10)
Uncertainty & probabilistic reasoning (chapter 13-14)
+ Some additional par ts (if possible)
Main Topics of Course
Search
Heuristic Search (Chapter 3,4) Search spaces & heuristic guidance
Game tree search (Chapter 5) Working against an opponent
Backtracking Search (Chapter 6) Constraint Satisfaction Problems
Reasoning and knowledge Representation (Chapter 7-9)
Logical agents and First Order Logic for more general knowledge
Planning (Chapter 10)
Predicate representation of states, planning graphs, reachability heuristics
Uncertainty (Chapter 13-14) Probabilistic reasoning, Bayes networks
Course Overview
AI is one of the newest science (AI phrase was coined in 1956)
However, the quest for AI begins with dreams (thousands of
years ago)
One of the most preferred science (along with molecular
biology)
Stil l has openings for several full t ime Einsteins
Huge variety of subfields
Ranging from general purpose areas such as learning and perception
to such specific tasks as playing chess, proving theorems, writing
poetry, and diagnosing diseases
Relevant to any intellectual task (universal f ield)
AI is exciting
Learning
Reasoning (focus of our course)
Reasoning (logical, probabilistic)
Knowledge representation
Decision making or Problem solving (search, planning, decision theory)
Perception (Vision, Speech Recognition, …)
Robotics (abil ity to move and manipulate objects)
Natural Language Processing
Subareas of AI
What is AI?
What is i ntelligence?
What are features that make humans (animals, or animate
objects) intell igent?
What is Artificial Intelligence?
The abil ity to carr y out abstract thinking (Terman, 1921)
The capacity for knowledge, and knowledge possessed (Henmon,
1921)
The capacity to learn or to prof it by experience (Dearborn, 1921)
Intell igence is what is measured by intell igence tests (Boring, 1923)
A global concept that involves an individual's abil ity to act
purposefully, think rationally, and deal effectively with the
environment (Wechsler, 1958)
Intelligence Definitions
http://internal.psychology.illinois.edu/~lyubansk/IQdef.htm
A general factor that runs through all types of per formance (Jensen)
Intell igent activ ity consists of grasping the essentials in a given situation and responding appropriately to them (Heim 1970)
A person possesses intell igence insofar as he had learned, or can learn, to adjust himself to his environment (Colvin 1982)
Intell igence is adaptation to the environment (unknown)
Intell igence is that faculty of mind by which order is perceived in a situation previously considered disordered (R.W. Young, 1999)
Intell igence is the ability to use optimally l imited resources - including t ime - to achieve goals. (Kurzweil , 1999)
Intelligence Definitions
http://internal.psychology.illinois.edu/~lyubansk/IQdef.htm
You beat somebody at Chess (or Checkers or Backgammon or Go or Othel lo or Poker ) .
You prove a mathematical theorem using a set of known axioms.
Asked to prove a geometr y theorem, you draw a diagram and use i t to help you.
You are tr ying to come up with an algor i thm to solve a problem using other algor i thms you are famil iar with. You have meta-knowledge about how one should think about algor i thms, and you use this knowledge to cr i tique your algor i thm as you create i t.
A stranger passing you on the street notices your watch and asks, "Can you tel l me the time?" You say, " It's 3:00," and not simply "Yes."
On a particular day, you need to buy a bunch of things, meet three di fferent people, return some books to the l ibrar y, and get a certain amount of exercise. You plan the day in such a way that ever ything is achieved in an effic ient manner.
Samples of Human Level Intelligence
www.cs.indiana.edu/classes/b551/Notes/tasks.html
You are a lawyer who is asked to defend someone. You recall three similar cases in which the defendant was found gui l ty, and you turn down the potential c l ient.
You are told to find a large Phil lips screwdriver in a cluttered workroom. You enter the room (you've never been there before) , look around, make your way around without fal l ing over things, and eventual ly find the screwdriver.
You are shown five letters in an unfamil iar font. Later you recognize this font w hen you see a di fferent character from i t, and you are also able to make good guesses as to what the other characters in the font would look l ike.
You're a 6-month-old infant. You can produce sounds with your vocal organs, and you can hear speech sounds around you, but you don't know how to make the sounds you're hearing. In the next year you figure out what the sounds of your parents' language are and how to make them.
Someone taps out a rhythm, and you are able to beat along with i t and to continue to even after i t stops.
Samples of Human Level Intelligence
www.cs.indiana.edu/classes/b551/Notes/tasks.html
Formal Definitions of Artificial Intelligence
Human intelligence Rational
Thinking Thinking humanly Thinking rationally
Behavior Acting humanly Acting rationally
Rationality: doing the right thing
Mathematical characterizations of rationality have come from
diverse areas l ike:
Logic (laws of thought)
Economics (utility theory: how best to act under uncertainty, game
theory: how self-interested agents interact)
Rationality
Turing Test (Turing, 1950) - Operational test for intell igent
behavior:
A human interrogator communicates (through a teletype) with a
hidden subject that is either a computer system or a human. If the
human interrogator cannot reliably decide whether on not the subject
is a computer, the computer is said to have passed the Turing test.
5 minutes test, it passes by fooling the interrogator 30% of time
Turing predicted that by 2000 a computer could pass the test.
He was wrong.
Acting Humanly
To pass the basic Turing test:
Natural Language Processing (communication)
Knowledge Representation (storing what it knows or hears)
Automated Reasoning (using the stored info to draw new conclusions or answer questions)
Learning (adapting to new circumstances and recognition)
To pass the total Turing test (in addition to above):
Vision
Speech recognition and synthetize
Robotics
Emotions, Characteristics
Anticipated most of AI major fields (60 years ago)
Problem: Tur ing test i s not reproducible , con structive, or amenable to mathematical analysis
Acting Humanly (Cont.)
Needs some way of determining how humans thinks Brain imaging (observing brain in action)
Introspection (catching our thoughts as they go)
Psychological experiments (observing a person in action)
Scientific theories of internal activities of the brain
1) Experimental investigation of actual human or animal behavior (top-
down) -- Cognitive Science
2) Direct identification from neurological data (bottom-up) -- Cognitive
Neuroscience
Cognitive Science and AI are now distinct sciences (while continuing to fer tilize each other)
Precise theory of mind is not available and seems mysterious.
Thinking humanly: cognitive modeling
Aristotle codified the right thinking and correct
arguments/inference processes
“Socrates is a man, all men are mortal, therefore Socrates is mortal”
Direct l ine through mathematics and philosophy to modern AI
However, intelligent behaviors do not necessarily result from logical
deliberation
Main obstacles:
Not easy to convert informal knowledge to formal ones
Reasoning usually needs high computational resource
Thinking rationally: "laws of thought"
Rational agent does the right thing achieving the best
outcome or expected outcome (given what it knows)
Thinking rationality is sometimes par t of being a rational
agent (it is not al l of rationality)
Rational behavior doesn't necessarily involve thinking (e.g., blinking
reflex)
There may be no provable correct thing to do but something must be
done
Compared to approaches based on human (behavior or
thinking), it can be more scientif ic
Well-defined mathematically and completely general
Compared to “laws of thought”, i t is more general
Acting rationally: rational agent
Bounded rationality – design best agent for given resources
when not enough time available to do all computations
Perfect rationality as a good starting point
We’ll focus on acting rationally in this course.
Acting rationally: rational agent (cont.)
Rational agents
An agent is an entity that perceives and acts
Abstractly, an agent is a function from percept histories to actions:
[f: P* A]
For any given class of environments and tasks, we seek the agent (or agents) with the best performance
Agents acting rationally have been gradually more popular
than systems based on human intelligence (thinking or acting
humanly)
Definition of AI has also been changed during the time.
Despite successes, founders of AI including McCarthy &
Minsky have expressed discontent with the progress of AI
AI should put less emphasis on creating ever- improved version of
applications that are good at a specific task
AI should return to its roots “machines that think, that learn, and that
create” (Human-level AI)
AI evolution
Phi losophy Logic, mind as physical system, inductive learning
Mathematics Formal logic, algori thms, computation (computabili ty, decidabili ty, and
tractability), probability
Economics “preferred outcome” or utility, decision theory, game theory, delayed payoff
Neuroscience Physical substrate for mental activity
Psychology Experimental techniques, brain as an information processing device
Computer engineer ing Building efficient computers (making AI applications possible)
Control theor y & cybernetics Self-organizing machines, homeostatic systems and stabi li ty, optimal system
designs
Linguistics Relation of language and thought (knowledge representation)
Foundations of AI (Related Fields)