EVOLUTION OF ARTIFICIAL INTELLIGENCE

10.3 EVOLUTION OF ARTIFICIAL INTELLIGENCE

The development of artificial intelligence includes four major stages. Figure 10.1 shows the evolution from 1960 to now. First, in 1956, a group of computer scientists gathered at Dartmouth College to discuss the great potential of computer applications. They were confident that computers, given enormous computing power, would be able to solve many complex problems and outperform human beings in many areas. At that time, scientists had little understanding of the complexity of human intelligence and were overly optimistic about what the computer could achieve. Many solutions created

at that time were primitive, and hence the stage is called the naive solution stage.

A f t e r several years of trial and error, scientists started focusing on devel opi ng more effective problem-solving methods, such as knowledge-representation schemes, reasoning strategies, and effective search heuristics. Since the feature of this stage is the

development of general purpose methods, it is named the general method stage.

After building enough general purpose methods, people started applying them to real-world applications. The application at this stage is different from the first one in that we already knew that solving c o m m o n sense problems is difficult. Therefore, most applications were targeted at a narrowly defined domain with specialized knowledge. Systems of this kind are called expert systems (ES). The feature is that acquisition of

expert knowledge plays a key role in development such systems. We call it the domain

C H A P T E R 1 0 ARTIFICIAL INTELLIGENCE A N D EXPERT SYSTEMS: K N O W L E D G E - B A S E D SYSTEMS 5 4 3

Domain Knowledge

General Methods

Na'fve Solutions

F I G U R E 1 0 . 1 FOUR STAGES O F A ! EVOLUTION

A I S I N FOCUS 10.2 ARTIFICIAL INTELLIGENCE VERSUS NATURAL INTELLIGENCE

Hie potential value of artificial intelligence can be bet- are erratic; they do not always perform consis- ter understood by contrasting it with natural, or human,

tently.

intelligence. Al has several important advantages: • Al can be documented. Decisions made by a com- puter can be easily documented by tracing the

• Al is more permanent. Natural intelligence is per- activities of the system. Natural intelligence is diffi- ishable from a commercial standpoint in that work-

cult to document. For example, a person may reach ers can change their place of employment or forget

a conclusion but at some later date may be unable information. However, Al is permanent as long as

to re-create the reasoning process that led to it, or the computer systems and programs remain

to even recall the assumptions that were part of the unchanged.

decision.

• Al offers ease of duplication and dissemination. • Al can execute certain tasks much faster than a

Transferring a body of knowledge from one person

human.

to another usually requires a lengthy process of apprenticeship; even so, expertise can seldom be

• Al can perform certain tasks better than many or

duplicated completely. However, knowledge

even most people.

embodied in a computer system can be easily trans- Natural intelligence does have several advantages ferred from that computer to any computer on the

over Al, such as:

Internet or on an intranet. • Natural intelligence is creative, whereas Al is

• Al can be less expensive than natural intelligence.

rather uninspired. The ability to acquire knowledge There are many circumstances in which buying

is inherent in human beings, but with A l , tailored computer services costs less than having corre-

knowledge must be built into a carefully con- sponding human power carry out the same tasks.

structed system.

This is especially true when knowledge is dissemi- • Natural intelligence enables people to benefit from nated over the Web.

and use sensory experience directly, whereas most

• Al, as a computer technology, is consistent and thor-

Al systems must work with symbolic input and rep- ough. Natural intelligence is erratic because people

resentations.

P A R T I V INTELLIGENT DECISION SUPPORT SYSTEMS

Since 1990, more advanced problem-solving methods have been developed. There is a strong n e e d to integrate multiple techniques and solve problems in multiple domains. Hybrid systems such as integrating rule-based and case-based systems, or integrating artificial neural networks and genetic algorithms, b e c o m e necessary. We

call it the integration stage.

The use of artificial intelligence in decision support systems has advantages and limitations. See A I S in Focus 10.2 for a comparison of artificial and natural intelligence.