Transcript inteligentni sistemi in agenti - Department of Intelligent Systems

INTELLIGENT SYSTEMS
3.
M. Gams
Intelligent systems

IN. SOCIETY
ENGINEERING, TECHNOLOGY
ARTIFICIAL INTELLIGENCE
Definition (scientific):
Intelligent ststem is a system that learns during its existence.
It senses its environment and learns, for each situation,
which action permits it to reach its objectives.
It continually acts, mentally and externally, and by acting
reaches its objectives.
To reach its objective it has to select its response. A simple
way to select a response is to select one that was favorable in
a similar previous situation.
Engineering definition: more advanced than current ones.
Properties / Learning
Learning
 Flexibility
 Adaptation
 Explanation
 Discovery


(engineering – mouse?)
Intelligent systems
Engineering, invisible intelligence
Practical directions, real-life problems
Verified AI methods: rule-based systems, trees,
expert systems, fuzzy systems, neural networks,
genetic algorithms, hybrid systems
Intelligent systems often simulate human
bureaucrats, expert systems simulate experts
Motivation
People are expensive (to buy or maintain),
computers cheap: computers work 24 hours a
day, no vacations, network accessibility is
worldwide, only 3% microprocessors in
computers, an average car 16
microprocessors, exponential trend (faster,
cheaper, more applications)
Intelligent systems are more friendly, more
flexible than classical systems (not truly
intelligent, just a bit more than classical)
S. Goonatilake, P. Treleaven:
I. S. for Finance and Business
•20 years ago substantial increase in IS
Killer applications - breakthrough
•Visa, 6 G trans. ann., 550G$, security;
American Express, 15$ > 1.4$
•typical: lots of data, new AI and HW cap.
•quality improvement, lower costs,
Killer application
American Express, Visa
Authorizer’s Assistant - an expert system
before: simple rigid rules, majority left to
human supervisors, many people with
different performance
Then new: an expert / intelligent system
with many rules, copies expert supervisors,
faster, cheaper, more equilibrated
15$ > 1.4$ per one transaction
(Visa - an neural network – DM and ML
prevail)
Benefits
The key question – trust – can IS be
trusted - obviously good enough (actually
as good as average humans)
Intelligent systems enabled
organizational changes in terms of HW,
SW and humans
Work done better and faster, more
profits, cheaper transactions
Less employed, more work done by
computers
Problem - unemployment
Conclusion
Intelligent systems apply AI methods and
introduce intelligent services
IS combine advantages of computer systems
(cost, availability) with some human properties
(simple engineering intelligence – learning,
adapting, reasoning), and achieve better
cost/benefit for several tasks
Especially appropriate for mundane
bureaucratic tasks in information society