Transcript Beautiful Thinking - Bradley University

Introduction to Artificial Intelligence
for Bradley University – CS 521
Anthony (Tony) J. Grichnik
Visiting Scientist to Bradley University
Caterpillar Inc.
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
6
9
1
8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
7
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
1
4
3
Copyright 2006, Tony Grichnik ~ All Rights Reserved
8
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
7
3
5
4
1
2
6
Copyright 2006, Tony Grichnik ~ All Rights Reserved
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
6
9
7
2
Copyright 2006, Tony Grichnik ~ All Rights Reserved
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
3
8
4
5
9
7
6
1
2
Copyright 2006, Tony Grichnik ~ All Rights Reserved
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
8
9
1
6
Copyright 2006, Tony Grichnik ~ All Rights Reserved
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
4
5
9
7
1
Copyright 2006, Tony Grichnik ~ All Rights Reserved
2
6
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
6
1
9
5
Copyright 2006, Tony Grichnik ~ All Rights Reserved
8
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
5
1
8
2
Copyright 2006, Tony Grichnik ~ All Rights Reserved
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
1
5
8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
9
4
1 3 5 7 9
2 4 6 8
Copyright 2006, Tony Grichnik ~ All Rights Reserved
3
7
2
6
Copyright 2006, Tony Grichnik ~ All Rights Reserved
Are you ready?
Quiz

Who goes first?

How many digits are there?

What is the object of the game?

How did you figure this out?
Copyright 2006, Tony Grichnik ~ All Rights Reserved
What do you think of when you
think of Artificial Intelligence
(AI)?
Copyright 2006, Tony Grichnik ~ All Rights Reserved
To understand AI is to move beyond
programming.
The key to AI is the “I” – Intelligence!
- for computers to understand, adapt
and improve.
AI moves us toward understanding
ourselves as much as it does solving
problems.
Copyright 2006, Tony Grichnik ~ All Rights Reserved
Goals for This Class

Gain a basic overview of the three major fields in Artificial
Intelligence (AI).

Complete one project in each field to build up your academic
resume.

Become sufficiently exposed to AI to see if you’d like to pursue a
career in the field or go deeper into AI academia.


You will not be an expert at the end of the class…or at the end of your
life…there should always be something left to learn.
You will be “dangerously competent” if you complete all the projects
successfully.
Copyright 2006, Tony Grichnik ~ All Rights Reserved
Expectations of Students

Learn! (Well duh….)

If you’re given something to read, read it and read all of it.


If you don’t know, ask.


We can’t help much if you don’t!
Don’t show up with things that don’t compile or execute.


If you don’t “get it,” speak up! (It won’t get better if you keep plodding
along and are lost from the beginning.)
They may run wrong or poorly, but this is a Master’s level class. Basic
programming should be old hat by now.
Complete the assignments to the best of your ability.
Copyright 2006, Tony Grichnik ~ All Rights Reserved
Introduction to Artificial Intelligence
for Bradley University – CS 521
Anthony (Tony) J. Grichnik
Visiting Scientist to Bradley University
Caterpillar Inc.
Outline

Introduction:
The Clans of Artificial Intelligence (AI)

Logical – Be the Expert

Statistical – Would you like to play a game?

Biological – Solutions…naturally

The Future – Hybrids and more
Copyright 2006, Tony Grichnik ~ All Rights Reserved
Introduction
(Welcome to the neighborhood…)

Contrary to popular belief, AI is not a new field.




In general there are three major fields of AI.



In fact it’s founded on principles that are hundreds – sometimes thousands – of years old.
Modern computers simply implement the techniques faster, more effectively and on a
broader range of problems.
They key is that computers make AI a practical part of daily life.
Until recently (the last 25 years or so) they absolutely couldn’t stand each other – and in
some ways still can’t stand each other.
Think of them like warring clans bent on each other’s destruction, because for generations
they have competed for resources as such.
As a result, if you find a “deep expert” in one one the three fields, the following things
will likely be true.



They will be able to tell you what is really great about their clan’s approach.
Their techniques will probably be able to solve your problem, although how efficient and
effective the solution will be may not be optimal. (Generally the tools from any clan can be
bent into doing whatever you need done.)
They will be able to tell you why the other clans are worthless pigs and should be ignored.
Conversely they rarely know what the other clans do well.
Copyright 2006, Tony Grichnik ~ All Rights Reserved
The three major clans of AI
Bayesian Inference
Surfaces & Splines
Support Vector Machines
Clustering
Regression Models
Fuzzy Logic
Coevolution
Ants & Swarms
*
Expert Systems
Inference Engines
Neural Networks
Genetic Algorithms
“BSB” models
Heuristic systems
Copyright 2006, Tony Grichnik ~ All Rights Reserved
*heretics
The Statistical Clan
Foundation
•
Based on mathematical processes dating back hundreds of years
Common Methods
•
•
•
Regression Methods – Data is fit to a equation of known form, then metrics (R, R2, covariance, etc.) are used to measure the quality of the
fit. (See http://www.statsoft.com/textbook/stgrm.html )
Bayesian Inference – Probabilities are calculated to confirm or reject future conditions based on previous (posterior) data. (See
http://yudkowsky.net/bayes/bayes.html )
Clustering Models – Common groupings of vectors are identified via search. (See http://www.statsoft.com/textbook/stcluan.html )
Key Strength
•
Statistical methods work well for large data sets, allowing abstraction to a few key parameters like mean, standard deviation, moment and
kurtosis.
Key Weakness
•
Statistical methods must make distribution assumptions to be effective. When these are not met statistical methods often break down.
Common Uses
•
•
•
Political and economic analysis (like trending, consumer metrics)
Transactional processing (like banking – especially fraud detection)
Medicine (drug testing, etc.)
Copyright 2006, Tony Grichnik ~ All Rights Reserved
The Logical Clan
Foundation
•
Based on principles established in philosophy and rhetoric (logical thinking), in some cases dating back to ancient Greece
Common Methods
•
•
Expert Systems – A framework of logic statements capture the experience of a person or people knowledgeable in a particular domain.
See http://www.amazon.com/gp/reader/0824799275/ref=sib_dp_pop_ex/102-7050084-5599306?ie=UTF8&p=S00P#reader-link and read
the entire “Excerpt.”
Rule Induction / Rule Abduction – Similar to expert systems, Rule Induction works from historical data to extract a framework of rules that
describe the underlying knowledge structures present in the data (e.g. “Result R is a consequence of Rule A operating on facts F1..Fn”).
Rule Abduction is similar but focuses on explanation of rules that exist (e.g. “Rule A is explained or supported by evidence E1..En”). See
http://en.wikipedia.org/wiki/Logical_reasoning and also http://en.wikipedia.org/wiki/Abductive_reasoning
Key Strength
•
Logical methods are highly explainable to the general public. It is easy to see “why” a logical AI system behaves as it does.
Key Weakness
•
Logical systems often fail as the number of facts represented increases. A bivalent or Boolean system expands a 2 n, where n = the
number of conditions evaluated in the system.
Common Uses
•
•
Simple control systems (like your microwave or simple industrial automation)
Product selection guides (like those found on many websites or at the auto parts store)
Copyright 2006, Tony Grichnik ~ All Rights Reserved
The Biological Clan
Foundation
•
Biology, naturally! More specifically, computing principles derived from biology. These techniques are relatively young, with the oldest
dating back 50 – 100 years.
Common Methods
•
•
Neural networks – Computational models of brain functions. Generally fall into synthesizing (interpolation) and classifying (categorization)
models with hundreds of variations. See http://en.wikipedia.org/wiki/Neural_network as a good starting point.
Genetic algorithms – Maps complex searches into a “survival of the fittest” approach. See
http://www.rennard.org/alife/english/gavintrgb.html and especially the Java applet http://www.rennard.org/alife/english/gavgb.html )
Key Strengths
•
•
Biological methods are often faster than equivalent methods from other clans, often by orders of magnitude.
Biological methods are often more powerful solutions to highly complex problems. Biologically-inspired computing techniques are
sometimes the only means to solve complex problems in reasonable amounts of time.
Key Weakness
•
Biological AI often lacks explainability provided by the other two clans. The solutions can often be verified but the process of obtaining the
solution can be difficult to follow.
Common Uses
•
•
•
Pattern recognition (like vision systems, voice print analysis)
Optimization processes (like for industrial processes)
Complex control systems (like some advanced aircraft and engines)
Copyright 2006, Tony Grichnik ~ All Rights Reserved
The Heretics…Hybrid Systems
Foundation
•
In the last 15-20 years a significant expansion has occurred in hybrid systems containing elements of each major clan.
Common Methods
•
•
•
Fuzzy logic – Combines logical AI with statistical concepts. See http://www-bisc.cs.berkeley.edu/ and view the PowerPoint presentation
http://www-bisc.cs.berkeley.edu/BISCProgram/PBIDS.ppt
Heuristic systems – Logical AI combined with biological concepts. See http://www.mpibberlin.mpg.de/en/forschung/abc/forschungsziele.htm
Swarm computing – Uses independent biological agents (“ants”) to statistically explore potential solutions. See
http://en.wikipedia.org/wiki/Swarm_intelligence and linked sites as needed.
Key Strengths
•
Combines the best elements of each field to make even more powerful AI systems.
Key Weakness
•
•
“A camel is a horse designed by committee.”” – Anonymous
If poorly constructed, a hybrid will bring together the worst elements of each clan rather than the best ones.
Common Uses
•
Few so far…see http://proceeed.statsoft.com for one commercial hybrid.
Copyright 2006, Tony Grichnik ~ All Rights Reserved
If you learn nothing else from this class,
learn this:
USE THE RIGHT TOOL FOR THE
JOB!!!!
Memorize the strengths and
weaknesses of each clan and use that
to your advantage!
Copyright 2006, Tony Grichnik ~ All Rights Reserved
Introduction to Artificial Intelligence
for Bradley University – CS 521
Anthony (Tony) J. Grichnik
Visiting Scientist to Bradley University
Caterpillar Inc.