318 lecture 1

AI definitions

Moving definition:

Whenever AI techniques solve a problem, Very soon thereafter, the technique is no longer considered an AI technique, but becomes more mainstream CS.

The two definitions try to avoid the whole issue of defining Artificial or defining Intelligence and will do for now.

At least they define the boundaries of the problem.

Early AI (symbolic):

One possible Classification:

Mundane problems, Formal problems, Expert Problems.

MUNDANE TASKS

Perception

Natural Language

Common-sense Reasoning

Simple reasoning and logical symbol manipulation

Robot Control

FORMAL TASKS

Games

To solve these problems we must explore a large number of solutions quickly and choose the Best One.

Mathematics

Proving Properties of Programs

Manipulate Symbols and reduce problem (usually recursively), until the answer is obvious. That is, it can be looked up in a table.

EXPERT TASKS

Engineering

Scientific Analysis

Medical Diagnosis

Financial Analysis

Rule based systems

AI's Underlying assumption:

The PHYSICAL SYMBOL SYSTEM HYPOTHESIS

A physical symbol system (such as a computer) has the necessary and sufficient means for general intelligent action

In other words:

Computers (Turing Machines) have the power for general intelligent action.

AI's Goals:

Winston:

ENGINEERING goals:

SCIENTIFIC goals:

Goals for this COURSE:

Learn a set of tools for REPRESENTING and MANIPULATING knowledge.

Project:

Learn basic approaches that deal with LEARNING or ADAPTATION

After this course:

You can confidently solve any problem that you could be confronted with. Or at least say that it is too hard with current technology. Point out areas of improvement (AI related) in any software you develop.You will be in a better position to handle hard problems and will be able to advise on whether AI will be useful for a particular project.

Software developers with an AI background are some of the best developers. They not only have to deal with hard problems but also limited resources. This is one area where computing power is never enough and you need to be creative in implementation.

LANGUAGES:

LISP is often called the language of AI but this is not a programming course and so no LISP...

Use any language you like for the assignment.

AI and other disciplines:

SCIENCE and ENGINEERING:

Knowledge based systems: Areas of study for which you require expertise.

PSYCHOLOGY:

AI and psychology are related because psychology is the study of human behavior (intelligent behavior), so can we learn by copying? AI want to make COMPUTATIONAL equivalents of anything learned from study of human behavior. Unfortunately copying does not always work consider the aeroplane.

LINGUISTICS...similar to the above

AI and BIOLOGY are related for similar reasons. Neural Networks (NNs) came out of this kind of study of Brains (human and animal).

For example, studying the human visual system gives a good model for computational vision.

EVOLUTIONARY THEORY

Genetic Algorithms (GAs) are algorithms that mimic evolution to learn, design, and adapt over time.

Characteristics of Intelligent systems:

One basic theme in AI

REPRESENTATION:

A representation is a set of conventions about how to describe a class of entities. A good representation often makes it very easy to solve a problem.

Problem :

farmer, hungry fox, fat goose, delicious grain.

River

Constraints:

How to cross the river?

English language representation is hard.

Try visual/graphical representation:

F=Fox

G=Goose

M=Farmer

X=Grain

~=River

F G M X ~

F X ~ M G

M F X ~ G

F ~ M G X      or     X ~ M F G

M F G ~ X             M G X ~ F

G ~ M F X

M G ~ F X

~ M G F X

The representation is more visual

This is an example of a SEMANTIC NET

The REPRESENTATION PRINCIPLE

A Good Representation:

links define the possible paths from a given node;

A transition is possible if there is a link else NOT

Some Terminology

A Representation has four PARTS

SEMANTIC NETS

A SEMANTIC NET is a representation

In which

Constructors construct nodes

Readers