Download Artificial Intelligence Winter 2004

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Artificial intelligence in video games wikipedia , lookup

AI winter wikipedia , lookup

Technological singularity wikipedia , lookup

Embodied cognitive science wikipedia , lookup

Knowledge representation and reasoning wikipedia , lookup

Philosophy of artificial intelligence wikipedia , lookup

Intelligence explosion wikipedia , lookup

History of artificial intelligence wikipedia , lookup

Ethics of artificial intelligence wikipedia , lookup

Existential risk from artificial general intelligence wikipedia , lookup

Transcript
CPSC 533 - Artificial Intelligence
Winter 2004
Michael M. Richter
ICT 550
[email protected]
Office hours: M 14-17
CPSC 533 - Artificial Intelligence 2004
Main Goals of the Course
Introduction into artificial intelligence as a discipline of
computer science:
„ Theoretical foundations, logic and mathematics
„ Methods, languages and algorithms
„ Major application areas: Where can AI be used
„ Very important:
z Modeling the Reality
z Abstract thinking
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Prerequisites
„ It is also very important to mention that mathematical
knowledge is needed, in particular interest in
methods of formal logic, some calculus is also good.
„ This is not a math course but you should be afraid to
mathematical techniques.
„ In addition, abstract thinking is important in order to
deal with general practical problems from an abstract
point of view: Any computer program can only grasp
a small part of the real world.
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Grading
„ Exam component:
a:
Midterm
20%
b:
Final
30%
„ Assignment component
c:
Project presentation
d:
Implemented System/Demo
„ Both components must be passed
„
CPSC 533 - Artificial Intelligence 2004
individual
individual
20%
30%
Michael M. Richter
team
team
Other organizational items
„ Office hours: Monday 14:00 - 17:00
or by appointment
(make use of the "by appointment"!)
„ Project Presentation: February 10
„ Midterm:
TBA
„ Labs: Start next week.
Participation in first lab required!
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Intelligent Behavior An Example
„ Suppose someone wants to get a contract for a cell
phone.
„ There are many possibilities for such a contract that may
be more or less useful.
„Hence a discussion between the customer and the sales
person starts in order to find out what may be most useful.
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Starting the Discussion
Customer: I would like to get a contract for a cell phone
Sales Person: Which of the following images best describes
how you will be using your phone ?
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Following Questions
Will you use the phone mainly during business times?
„ Y
N
„ How many minutes do you expect to spend on calls per day?
„ 120
„ What is your monthly cell-phone allowance (in Canadian $)?
„ 500
„ Do you need to control the cost of your usage?
Y
N
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
The Intelligent Action: A
Recommendation
„ You are a business user with a cell-phone
allowance of $500 per month.
„ We are pleased to recommend Talk 100+
which we feel will meet your needs.
„ To see more click here.
(It follows e.g. the offer of special numbers with
low costs).
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Example What is involved (I)
“I would like to get a contract for a cell phone”
Tasks:
„ Hear the sentence and transfer it in a written version
„ Parse and “understand” the written version
(i.e. transfer it into an internal representation)
) Natural language understanding
„ Seeing the image:
„ Identify the icons and understanding the meaning
) Vision and image understanding
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Example What is involved (II)
„ Knowledge involved:
z About different types of customers (represented by the icons)
z About the many possibilities for contracts (constraints)
z About the aspects of usefulness
z About the consequences of partial decisions
„ Planning how the sales person should proceed:
z Selecting the next question
z Drawing consequences from the answer
z Reducing the possible answers
z Making the final (optimal) recommendation
) Deduction, Reasoning
) Planning
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
What is Artificial Intelligence (1) ?
„ We do not intend to define what intelligence is, we take this
term in a naïve sense.
„ AI deals with the development of systems either displaying a
behavior humans associate with intelligence or solving a
problem humans think only an intelligent being can solve.
„ Intelligent solutions are good for performing better actions: The
better the solutions, the better the actions.
„ Intelligent actions need two things:
z Knowledge
z Reasoning
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
So, what is Artificial Intelligence (2)
„ Hence AI systems need
z A way to represent knowledge (e.g. certain need
certain languages)
z Methods for reasoning (knowledge processing) :
Here logic plays á central role
„ Therefore the term Knowledge-based Systems is
often used. There are, however, other views in which
computation is dominant.
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Intelligence Revisited
„ Combinatorial intellegence: This person is
intelligent because s/he is so good on combinatorial
puzzles!
„ Evaluation: Objective tests
„ Emotional intelligence: This person is intelligent
because s/he really understands what I mean!
„ Evaluation: My subjective opinion
„ Social intelligence: This person is intelligent
because s/he fits so good into our group!
„ Evaluation: The common opinion of our group
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Comparing AI Systems and
Procedural Programming
„ In a procedural program all steps are sequentially
organized: To run the program one only has to follow
carefully the instructions (this is the only intelligence
needed).
„ In an AI system certain knowledge is represented and
reasoning methods are provided. The programmer
does not prescribe the order in which the steps are
performed. This leads to the problem of searching for
useful steps what looks intelligent.
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Fully Automatic versus Assistant
Systems
In the beginning of AI: The GPS program (General
Problem Solver) “describe problem in logic and it will
solve it using logical inferences”
Result: Complete failure and enormous negative
reaction in media
Specialization: Fully automatic systems for special
tasks: Was partially successful if the problem could
be formulated in a logical way.
Current trend: Assistant systems that employ humans
as well as machines using the strengths of both.
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Major classical areas of AI
Mathematical reasoning
Natural language understanding and dictating
Vision
Robotics
Expert systems (For configuration, diagnosis and
planning)
„ Machine learning
„ Neural networks
„ General Knowledge Based Systems: Many aspects
of this course are in this area !
„
„
„
„
„
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Diagnosis of Robots
Diagnosis in Industry (1)
Diagnosis of Telecom
Switches at Ericsson
Diagnosis in Industry (2)
Diagnosis in Industry (3)
Troubleshooting CFM56-3 Engines
for the Boeing 737
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Developing an Application
Identify
Data sources
Meta Model
Description
CPSC 533 - Artificial Intelligence 2004
Knowledge
Input
Generate
Interface
Michael M. Richter
Problems with Common Sense
„ Common sense is something you learn implicitly as a
child, not in the school or the university.
„ Common sense is underlying many decisions
Problems:
„ Common sense is not recorded explicitly and people
cannot easily telling when they applied which principle.
„ Therefore it is difficult to represent common sense in a
computer system
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Manipulation on Informal Objects:
A Major Challenge
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
AI The Present (I)
New developments and Applications:
„ Multi-Agent Systems:
z Cooperation concepts to bring together the many specialized
systems to tackle harder tasks
z User has many special (and possibly rather dumb)
assistants
„ The Internet
z Requires agents for “intelligent” routine tasks that are
specialized and limited
z “Real world” is inside the computer
) environment is AI friendly
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
AI The Present (II)
„ The Semantic Web:
z Enhance web sites with computer processable
annotations
z Decentralization of knowledge representation and
processing
„ Relating logical and approximative reasoning
z Some problems do not have exact solution or they
may be too hard to find
„ Improvement of systems by learning and adaptation
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
The Landscape of the Semantic Web (1)
A rough division, no sharp boundaries:
Both parts influence
Law
each other in many Economic
respects
Models
Marketing
Etc.
CPSC 533 - Artificial Intelligence 2004
Computer
Technology
Michael M. Richter
The Landscape of the Semantic Web (2)
Computers and the Web :
Again:
Both sides
influence
each other
Intelligent
Software
Content-dependent
CPSC 533 - Artificial Intelligence 2004
Basic technology:
- hardware
- networks
- system
programming
- XML
etc.
Content-independent
Michael M. Richter
The Landscape of the Semantic Web (3)
Logic:
Data bases
Formal semantics
Deduction
etc
Logic
oriented
Approximation:
Inexactness
Subjectivity
Informal notions
Approximation
oriented
Two aspects of intelligent behavior
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Socio-Technical Processes (1)
„ These are processes were the participating actors („agents“) can be
humans as well as machines.
„ This requires a careful organization of the division of labor:
z What do humans?
z What do machines?
„ A particular problem is the communication between humans and
machines:
z Humans have difficulties to understand the results of machines
z Machines have difficulties to understand the results of humans
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Socio-Technical Processes (2)
„ Conflicting demands:
z Machines need precise instructions
z Humans want to use creativity.
„ Plans and Executions:
z They alternate, before all requirements are present and
before planning is finished execution of some actions start.
„ We will consider some examples that deal with different
aspects:
z Software development processes
z Dialogs and sales processes in electronic commerce
z Search in the internet.
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
AI The Future ? (I)
„ Computer science of today is what computers can do, computer
science of tomorrow is what users can do.
„ The old evalution of computers uses megahertz and gigabytes.
The new evaluation is about „How many bids can I make in
auction?“ or „How long does it take to find something in the
internet?“
(Ben Shneiderman: Leonardo‘s Laptop, chapter 1)
One way to look at the course is to make these statements
precise: How humans be supoprted in an intelligent way?
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
AI The Future ? (II)
„ Specialized agents (avatars) helping and
representing users (in the new digital world)
„ “Command-and-Control” systems for management
and CEOs in large companies
) Management by exception
„ Cooperating intelligent robots in disaster areas and
hostile environments represent humans acting out of
save environments
„ Automatization of trade and commerce
) Management by Policies
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
AI The Future ? (III)
In general:
Use of AI enhances what a human being can do
intellectually
Still the hard problems:
„ Achieving necessary flexibility to interact with human
beings in real world
„ Find concepts to combine the methods of the
different fields
„ Common sense reasoning
CPSC 533 - Artificial Intelligence 2004
Michael M. Richter
Content
„ 1. Introduction
„ 2. General Aspects of
Knowledge Processing
„ 3. Search
„ 4. Logic Oriented Knowledge
Representation
„ 5. Rules
„ 6.Impreciseness
„ 7. Constraints
„ 8. Diagnosis
„ 9.Planning
CPSC 533 - Artificial Intelligence 2004
„ 10. Learning
„ 11. Neural Networks
„ 12. Agents
„ Remark: Some of the
material can be regarded as
add-ons. This will be
explained in the lecture
Michael M. Richter