presentation source

Download Report

Transcript presentation source

Artificial Intelligence:
Research and Collaborative
Possibilities
a presentation by:
Dr. Ernest L. McDuffie, Assistant Professor
Department of Computer Science, Florida State University
at the:
First Annual Africa-America Cooperative Workshop
in Computational Science & Engineering
University of the Western Cape
Cape Town, South Africa
21 July 2000
What is Intelligence?







A single faculty or just a collection of distinct and unrelated
abilities?
Exactly what happens when learning occurs?
What is intuition?
What is self-awareness?
Can intelligence be inferred from observable behavior, or does it
require evidence of a particular internal mechanism?
How is knowledge represented in nerve tissue?
Is it even possible to achieve intelligence on a computer, or
does an intelligent entity require the richness of sensation and
experience that might be found only in a biological existence?
What is AI?
The study of intelligent behavior.
 The goal being a theory of intelligence
that can account for the behavior of all
naturally occurring intelligent entities.
 Then use the theory to guide the
creation of artificial entities capable of
intelligent behavior.

AI Fields
Natural Language Processing
 Game Playing
 Automatic Theorem Proving
 Pattern Recognition & Computer Vision
 Expert Systems
 Modeling Forms of Reasoning
 Automatic Learning
 Robotics

Knowledge
We assume that intelligent entities have
knowledge about their environment.
 What can we say about such
knowledge? What forms can it take?
What are its limits? How is it used? How
is it acquired?
 We are beginning to understand how
neurons process simple signals, but
how the brain processes and represents
knowledge is still not well understood.

How Computers Represent Knowledge
There are two major ways we think of
machines having knowledge of the
world.
 Clarification about the distinction
between the two is on going.
 They are implicit or procedural
knowledge and explicit or declarative
knowledge.

Implicit Knowledge
In a computer this type of knowledge
takes the form of stored procedures.
 The knowledge would manifest itself
when the procedure is run.
 In humans it is often called tacit
knowledge and can be difficult or
impossible to describe.
 It is difficult to easily modify this type of
knowledge in a computer.

Explicit Knowledge
Complex tasks that we usually think of
as requiring intelligence tend to use
explicit knowledge representations.
 A tabular database of salary data would
be one example of explicit knowledge.
 Particularly useful are explicit
representations that can be interpreted
as making declarative statements.

Efficiency vs. Flexibility
Using declarative knowledge usually is
more costly and slower than is directly
applying procedural knowledge.
 Declarative knowledge can also be
accessed by introspective programs so
a machine can then answer questions
about what it knows.
 Generally, we give up efficiency to gain
flexibility and vice versa.

AI Needs Both
Procedural and Declarative types of
knowledge.
 Most flexible kinds of intelligence seem
to depend strongly on declarative
knowledge.
 AI has concerned itself more and more
with this type of knowledge.
 Procedural knowledge still has a role to
play.

Computer Learning
To assimilate new information or
procedures without a programmer
writing a new program.
 This is different from discovery
programs like those designed to
formulate new mathematical theorems.
 A range of different techniques are used
in computer learning programs.

Some Techniques Are:




Induction - learning by generalization from specific
examples.
Candidate Elimination - a specific method of
induction; testing rules and a method for generating
new one.
Genetic Algorithms - finding better and better
versions of rules/programs/strings by using random
repeated mutations and selection.
Neural Net - a method of training to modify the
connections between neurons; back propagation.
Progress Has Been Slow
Learning from experience is difficult in
any domain that is not very restricted or
has formal contexts.
 It seems that even simple animals like
flies or slugs have better learning ability.
 Studies of these types of animals have
been used as background for some
neural net approaches.

A New Direction - MIT
Alternative Essences of Intelligence
 An attempt at building complex
machines with human like capabilities.
 Four essences - development, social
interaction, physical coupling to the
environment, and integration.
 Dr. Rodney Brooks, Director AI Lab,
MIT.

My Research

Temporal Reasoning
• Allen Relationships

Automatic Scheduling
• Lots of manufacturing applications

Second Generation Hybrid Expert
Systems
• Combining learning and decision making

Applied AI to real world problems
• Network security, intrusions detection
Any Questions?
Any Comments?
[email protected]
Work Phone: 850.644.3861
Fax: 850.644.0058
Thank you for your attention!