Download Systems that act like humans

Computer Science CPSC 422
Intelligent Systems
Cristina Conati
Super Brief Intro
• Advanced AI course
• Builds upon 322 (and 312)
• 322 gave a broad, high level overview of main research areas in AI (logic,
search, planning, reasoning under uncertainty, decision making)
• We will go into more depth on some of the topics
• Look at “Learning”
• Study some applications, in the field of Intelligent User Interfaces
Overview
• Administrivia
• Let’s connect back to 322: what is AI?
• Refresher
• Examples
People
Instructor
• Cristina Conati ( [email protected]; office CICSR 125)
Teaching Assistant
• Hajir Roozbehani ([email protected])
Course Pages
• Course website:
http://www.cs.ubc.ca/~conati/422/422-2010World/422-2010.html
• This site also includes a calendar with a tentative scheduling of
topics.
http://www.cs.ubc.ca/~conati/422/422-2010World/schedule-422-2010.html
CHECK IT OFTEN!
 Lecture slides
 Assignments/Solutions
 Other material
Course Material
• Main Textbook
• Artificial Intelligence: A Modern Approach (AIMA). 3rd edition, Russell
and Norvig
• Can buy cheaper e-version (http://www.coursesmart.com/9780136067337)
• Additional textbook: Artificial Intelligence: Foundations of
Computational Agents. by Poole and Mackworth. (P&M)
• I’ll post the relevant chapters in WebCT as needed
• This textbook it is not a substitute for the AIMA textbook
Course Material
• Lecture Slides
• I'll post a version of each lecture's slides just after class.
• But I won't post material that I write on the slides or on the board during
class. You'll have to come to class to get that .
• You are responsible for all the material in the readings for each
class, regardless of whether it has been explicitly covered in
class.
• You will also need to know all the material covered in class,
whether or not it is included in the readings or available online.
Readings
• It is strongly recommended that you read the assigned readings
before each class. It will help you understand the material better
when I lecture
• However, there will be some classes that are centered around the
discussion of one or more research papers.
• You MUST read the papers before coming to class, because
• you will have to come up with questions on them and participate to class
discussion (more on this later)
How to Get Help?
• Use the WebCT Discussion Board for questions on course
material (so check it frequently)
• That way others can learn from your questions and comments
• Use email for personal questions (e.g., grade inquiries or health
problems).
• Go to office hours (Discussion Board is NOT a good substitute
for this) – times below are still tentative, will be finalized next week
• Cristina: likely Tu: 3:30-4:30
• Hajir:
TBA
• Can schedule by appointment if you have a class conflict with the
official office hours
Evaluation
• Final exam (45%)
• midterm exams (30%)
• Assignments (15 %)
• Class participation and questions for classes based on paper
discussion (10%)
But, if your final grade is 20% higher than your midterm grade:
• Midterm: 15%
• Final: 60 %
To pass: at least 50% in both your overall grade and your final
exam grade
Coursework: Assignments
• Submit via Handin by the appointed deadline.
• See syllabus for late assignment policy
• You will have one Late Assignment Bonus, i.e. you will be
allowed to submit one of the assignments up to 2 days late with no
penalty (see details in syllabus).
Coursework: assignments
Is collaboration on assignments allowed? Yes, you may work with
one (and only one) other person.
•That person must also be a student in CPSC 422 this term.
•You will have to officially declare that you have collaborated with this
person when submitting your assignment.
•There will be more details about collaboration policy included in the
assignment itself.
Talking about the assignments with anybody other than an official
teammate, looking at existing solutions, submitting solutions not
worked out by the team members constitute plagiarism
See UBC official regulations on what constitutes plagiarism
(pointer in syllabus)
Coursework: discussion-based classes
• Discussion-based classes
• There will be a few classes during the course that will be
centered on reading and discussing one or more research
papers
• You will have to
 come up with critical questions (discussion points) on each of the
assigned readings (I will give you the exact number for each set of
readings)
 Hand in your questions (I’ll give you details on when and how to do
this as we go)
 Be prepared to present and discuss your questions in class
Coursework: discussion-based classes
• First discussion-based class: Tuesday, Jan. 19
• Paper (available on-line from class schedule):
• Conati C., Gertner A., VanLehn K., 2002. Using Bayesian Networks to
Manage Uncertainty in Student Modeling. User Modeling and UserAdapted Interaction. 12(4) p. 371-417.
• Make sure to have at least two questions on this reading
to discuss in class.
• Send you questions to *both* [email protected] and
[email protected] by 9am on Tuesday, Jan. 19
• Hand in a written copy of the questions at the end of class
Questions on papers
• Clarification questions are welcome, but there should
be at least two that can be used as “discussion points”,
i.e. that
• Question elements of the presented research (i.e. point out
weaknesses)
• make connections with the relevant techniques presented in
class (Bnets in case of the first discussion paper)
• Make connections/comparisons with other papers (once we
have covered enough papers to do this)
• Examples of questions of different quality are available
from the syllabus
Missing Assignments or Exams
• If serious circumstances (like an illness or other personal
matters) cause you to be late for an assignment or to miss an
exam:
• If you miss an assignment, your score will be reweighed to exclude that
assignment
• If you miss the midterm, its weight will be shifted to the final
 thus, your total grade will be 75 % final, 25% coursework
• If you miss the final, you'll have to write a make-up final as soon as
possible
• If you have flu-like symptoms, you do not need to bring a
doctor’s note, but you do need to report your illness at
http://www.students.ubc.ca/health/flu.cfm
• For all other circumstamces, you do need official certification
(e.g. doctor’s note)
To Summarize
• All the course logistics are described in the course
syllabus
• http://www.cs.ubc.ca/~conati/422/422-2010World/422-2010.html
• Make sure to read it and that you agree with the course
rules before deciding to take the course
Department of Computer Science
Undergraduate Events This Week
How to Prepare for the Tech Career
Fair
Date:
Wed. Jan 6
Time:
5 – 6:30 pm
Location: DMP 110
Resume Writing Workshop (for
non-coop students)
Date:
Thurs. Jan 7
Time:
12:30 – 2 pm
Location: DMP 201
CSSS Movie Night
Date:
Thurs. Jan 7
Time:
6 – 10 pm
Location: DMP 310
Movies:
“Up” & “The Hangover”
(Free Popcorn & Pop)
Drop-In Resume Edition Session
Date:
Mon. Jan 11
Time:
11 am – 2 pm
Location: Rm 255, ICICS/CS Bldg
Industry Panel
Speakers: Managers from Google, IBM,
Microsoft, TELUS, etc.
Date:
Tues. Jan 12
Time:
Panel: 5:15 – 6:15 pm;
Networking: 6:15 – 7:15 pm
Location: Panel: DMP 110;
Networking: X-wing Undergrad
Lounge
Tech Career Fair
Date:
Wed. Jan 13
Time:
10 am – 4 pm
Location: SUB Ballroom
Overview
• Administrivia
• Let’s connect back to 322: what is AI?
• Refresher
• Examples
What is Artificial Intelligence?
From: Russell S. and Norvig, P.``Artificial Intelligence: A Modern Approach.''
Systems that act like humans
“The study of how to make
computers do things at which, at
the moment, people are
better”(Rich and Knight, 1991)
Systems that think rationally
“The study of mental faculties
through the use of computational
models” (Charniack and
McDermott, 1985).
Systems that think like humans Systems that act rationally
“The automation of activities that “AI..is concerned with intelligent
we associate with human
behavior in artifacts (Nilsson,
thinking, such as decision
1998)
making, problem solving,
learning”(Bellman, 1978)
Systems that act like humans
• Turing test (1950): Can a human interrogator tell
whether (written) responses to her (written) questions
come from a human or a machine?
•
•
•
•
Natural Language Processing
Knowledge Representation
Automated Reasoning
Machine Learning
• Total Turing Test (extended to include physical aspects of human
behavior)
• Computer Vision
• Robotic
Has any AI System Passed the Tutoring Test?
• Not the full blown one (see http://www.loebner.net/Prizef/loebnerprize.html)
• Loebner Prize initiative has a 100,000 and a Gold Medal for the first
computer whose responses were indistinguishable from a human's. No one has
won this yet
• Each year a monetary prize and a bronze medal are awarded to the most
human-like computer.
• The winner is the best entry relative to other entries that year, irrespective of
how good it is in an absolute sense
• Variations restricted to specific tasks requiring some form of
intelligence
• Check Winner 2009 (http://www.worldsbestchatbot.com/Do_Much_More)
• And you can play with winner from 2008 (Elbot )
But why do we want an intelligent system
to act like a human?
• Because for many tasks, humans are still the Gold Standard
But why do we want an intelligent system
to act like a human?
Generating Multimedia Presentations
Zhou, Wen, and Aggarwal. A Graph-Matching Approach to Dynamic Media Allocation
in Intelligent Multimedia Interfaces. Best Paper Award at Intelligent User Interfaces 2005.
• Algorithm to effectively allocate text and graphics in
multimedia presentations
• Empirical Validation
• System (RIA) output on 50 user queries (real estate and tourist guide
application)
• Media allocation on same queries by two multimedia UI designers
• Third expert “blindly” ranked all responses
• Results
• RIA best/co-best in 17 cases
• Minor differences in 28 of the remaining 33 cases
Why Replicate Human Behavior, Including
its “Limitations”?
• AI and Entertainment
• E.g. Façade, a one-act interactive drama
http://www.quvu.net/interactivestory.net/#publications
• Sometime these limitations can be useful, e.g.
• Supporting Human Learning via Peer interaction
(Goodman, B., Soller, A., Linton, F., and Gaimari, R. (1997) Encouraging Student Reflection
and Articulation using a Learning Companion. Proceedings of the AI-ED 97 World Conference
on Artificial Intelligence in Education, Kobe, Japan, 151-158.)
• Supporting Human Learning via teachable agents
(Leelawong, K., & Biswas, G. Designing Learning by Teaching Agents: The Betty's Brain
System, International Journal of Artificial Intelligence in Education, vol. 18, no. 3, pp. 181208, 2008
What is Artificial Intelligence?
Systems that act like humans
“The study of how to make
computers do things at which, at
the moment, people are
better”(Rich and Knight, 1991)
Systems that think rationally
“The study of mental faculties
through the use of computational
models” (Charniack and
McDermott, 1985).
Systems that think like humans Systems that act rationally
“The automation of activities that
we associate with human
thinking, such as decision
making, problem solving,
learning”(Bellman, 1978)
“The branch of computer science
that is concerned with the
automation of intelligent
behavior (Luger and Stubblefield
1993)
Systems That Think Like Humans
• Use Computational Models to Understand the Actual
Workings of Human Mind
• Devise/Choose a sufficiently precise theory of the mind
• Express it as a computer program
• Check match between program and human behavior (actions
and timing) on similar tasks
• Tight connections with Cognitive Science
• Also known as descriptive approaches to AI
Some Examples
• Newell and Simon’s GPS (General Problem Solver, 1961) to test
means-end approach as general problem solving strategy
• John Anderson’s ACT-R cognitive architecture (http://act-r.psy.cmu.edu/)
•
Anderson, J. R. & Lebiere, C. (1998). The atomic components of thought. Erlbaum;
•
Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C., & Qin, Y . (2004). An
integrated theory of the mind. Psychological Review 111, (4). 1036-1060.
• SOAR cognitive architecture (http://sitemaker.umich.edu/soar)
•
Newell, A. 1990. Unified Theories of Cognition. Cambridge, Massachusetts: Harvard
University Press.
ACT-R Models for Intelligent Tutoring Systems
ACT-R Models for Intelligent Tutoring Systems
Intelligent Tutoring Systems (ITS)
Cognitive Science
Computer Science
(AI, HCI)
ILE
Education


Intelligent agents that support human learning and
training
By autonomously and intelligently adapting to
learners’ specific needs, like good teachers do
ACT-R Models for Intelligent Tutoring Systems
• One of ACT-R main assumptions: Cognitive skills
(procedural knowledge) are represented as production rules:
IF this situation is TRUE, THEN do X
• An ACT-R model representing expertise in a given
domain requires writing a set of production rules
mimicking how a human would reason to perform tasks
in that domain
• An ACT-R model for an ITS encodes all the reasoning
steps necessary to solve problems in the target domain
• Example: rules describing how to solve
5x+3=30
ACT-R Models for Intelligent Tutoring Systems
Eq: 5x+3=30 ; Goals: [Solve for x]
• Rule: To solve for x when there is only one occurrence, unwrap (isolate) x.
Eq:5x+3=30 ; Goals: [Unwrap x]
• Rule: To unwrap ?V, find the outermost wrapper ?W of ?V and remove ?W
Eq: 5x+3=30; Goals: [Find wrapper ?W of x; Remove ?W]
• Rule: To find wrapper ?W of ?V, find the top level expression ?E on side of
equation containing ?V, and set ?W to part of ?E that does not contain ?V
Eq: 5x+3=30; Goals: [Remove “+3”]
• Rule: To remove “+?E”, subtract “+?E” from both sides
Eq: 5x+3=30; Goals: [Subtract “+3” from both sides]
• Rule: To subtract “+?E” from both sides ….
Eq: 5x+3-3=30-3
Model Tracing
• Given a rule-based representation of a target domain (e.g.
algebra),
• an “expert model” can trace student performance by firing rules
and do a stepwise comparison of rule outcome with student
action
• Mismatches signal incorrect student knowledge that requires
tutoring
• Knowledge tracing extends model tracing to assess probability
that a student knows domain rules given observed actions
• These models showed good fit with student performance,
indicating value of the ACT-R theory
• Also, the Cognitive Tutors based on this model are great
examples of AI success – used in thousands of high schools in
the USA (http://www.carnegielearning.com/success.cfm)
What is Artificial Intelligence?
Systems that act like humans
“The study of how to make
computers do things at which, at
the moment, people are
better”(Rich and Knight, 1991)
Systems that think rationally
“The study of mental faculties
through the use of computational
models” (Charniack and
McDermott, 1985).
Systems that think like humans Systems that act rationally
“The automation of activities that
we associate with human
thinking, such as decision
making, problem solving,
learning”(Bellman, 1978)
“The branch of computer science
that is concerned with the
automation of intelligent
behavior (Luger and Stubblefield
1993)
Systems that Think Rationally
• Logic: formalize right thinking, i.e. irrefutable reasoning
processes.
• Logistic tradition in AI aims to build computational frameworks based on
logic.
• Then use these frameworks to build intelligent systems
• You have seen some examples in 322 (Propositional Logic) and
312 (Logic Programming)
• We will look at more advanced logic-based representations
• Semantic Networks
• Ontologies
Systems that Think Rationally
• Main Research Problems/Challenges
• Proving Soundness and Completeness of various formalisms
• How to represent often informal and uncertain domain
knowledge and formalize it in logic notation
• Computational Complexity
• Tradeoff between expressiveness and tractability in logicbased systems H. J. Levesque and R. J. Brachman. Expressiveness and tractability in
knowledge representation and reasoning. Computational Intelligence, 3(2):78--93, 1987.
What is Artificial Intelligence?
Systems that act like humans
“The study of how to make
computers do things at which, at
the moment, people are
better”(Rich and Knight, 1991)
Systems that think rationally
“The study of mental faculties
through the use of computational
models” (Charniack and
McDermott, 1985).
Systems that think like humans Systems that act rationally
“The automation of activities that
we associate with human
thinking, such as decision
making, problem solving,
learning”(Bellman, 1978)
“The branch of computer science
that is concerned with the
automation of intelligent
behavior (Luger and Stubblefield
1993)
Systems that Act Rationally
• The “think rationally” approach focuses on correct
inference
• But more is needed for rational behavior, e.g.
• How to behave when there is no provably correct thing to do
(i.e. reasoning under uncertainty)
• Fully reactive behavior (instinct vs. reason)
AI as Study and Design of Intelligent
Agents (Poole and Mackworth, 1999)
• An intelligent agent is such that
• Its actions are appropriate for its goals and circumstances
• It is flexible to changing environments and goals
• It learns from experience
• It makes appropriate choices given perceptual limitations and limited
resources
• This definition drops the constraint of cognitive plausibility
• Same as building flying machines by understanding general principles of
flying (aerodynamic) vs. by reproducing how birds fly
• Normative vs. Descriptive theories of Intelligent Behavior
Intelligent Agents
• In AI, artificial agents that have a physical presence in the world
are usually known as Robots
• Robotics is the field primarily concerned with the implementation
of the physical aspects of a robot (i.e. perception of the physical
environment, actions on the environment)
• Another class of artificial agents include interface agents, for either
stand alone or Web-based applications (e.g. intelligent desktop
assistants, recommender systems, intelligent tutoring systems)
• Interface agents don’t have to worry about interaction with the
physical environment, but share all other fundamental
components of intelligent behavior with robots
• We will focus on these agents in this course
Intelligent Agents in the World
Knowledge Representation
Machine Learning
Reasoning +
Decision Theory
Natural Language
Generation
Natural Language
Understanding
+
Computer Vision
Speech Recognition
+
Physiological Sensing
Mining of Interaction Logs
+
Robotics
+
Human Computer
/Robot
Interaction
The “Act Rationally” view
This is the view that was adopted in cpsc322, and that we will
continue to explore in the first part of the course
• Reasoning under uncertainty: Bayesian networks and Hidden
Markov Models
• Brief review, some applications, approximate inference
• Decision Making: planning under uncertainty
• Markov Decision Processes, review
• Partially Observable Markov Decision Processes (POMDP)
• Learning
• Decision Trees, Neural Networks, Learning Bayesian Networks,
Reinforcement Learning
Next Class
Review of Bayesian networks, and representational issues
IMPORTANT: You must be familiar with the basic
concepts of probability theory. I won’t go over them.
For are refresher: Ch. 13 in textbook and review slides
posted in the schedule page (Th. slot)