Download - INTELLIGENT SYSTEMS IN BUSINESS - Wiley: · PDF fileINTELLIGENT SYSTEMS IN BUSINESS 12 For generations people have attempted to make smart machines to perform tasks that require

INTELLIGENT SYSTEMS IN BUSINESS

12

For generations people have attempted to make smart machines to perform tasks thatrequire intelligence. In this chapter we present the achievement of this goal as it is re-lated to the world of business. Available intelligent systems range from expert systemsto industrial robots. These systems can be used by themselves or in conjunction withother systems, to increase productivity, quality, and customer service, and to reducecycle time. They can also be integrated among themselves or embeded in other infor-mation, electrical, or mechanical systems to improve the functionality of those sys-tems. Intelligent systems can be also used to facilitate communication andcollaboration among people within and between organizations, expanding the capa-bilities of the latter. Intelligent systems help us to communicate better with peoplewho speak other languages as well to communicate with computers. In addition, intel-ligent systems help us to overcome the information overload, enabling us to quicklyfind, compare, and analyze data and to better conduct electronic commerce and cus-tomer service. Finally, such systems can act as advisers and tutors to people.

CHAPTER PREVIEW

12.1 Artificial Intelligence and Intelligent Systems

12.2 Expert Systems

12.3 Other Intelligent Systems

12.4 Intelligent Agents

12.5 Virtual Reality: An Emerging Technology

12.6 Ethical and Global Issues of Intelligent Systems

CHAPTER OUTLINE

1. Describe artificial intelligence and compare it to

conventional computing.

2. Identify the characteristics, structure, benefits, and

limitations of expert systems.

3. Describe the major characteristics of natural language

processing and voice technologies.

4. Describe neural computing and its capabilities.

5. Define intelligent agents and their role in IT.

6. Describe virtual reality.

LEARNING OBJECTIVES

ANALOG DEVICES USES INTELLIGENT SYSTEMS TO SUPPORTCOLLABORATION WITH CUSTOMERS

Analog Devices Inc. designs, manufactures, and markets a broad line of precision in-tegrated circuits (ICs) used in analog and digital signal-processing applications. Theproducts are sold to various manufacturers that use ICs in their own products. Themarket for such ICs is very competitive.

With thousands of products, Analog Devices hadbeen printing new catalogs and data sheets each year,some of which measured up to two feet thick. The cost ofprinting and shipping catalogs to 50,000 customers world-wide reached about $3 million each year.

Analog’s sales engineers used to take customer re-quirements by phone and try to manually find a match inthe company’s product range. This complex process in-volves considering dozens of constraints while interactingwith customers (usually design engineers) to ascertaintheir needs and priorities. Only well-trained engineersare successful in this lengthy process. Since it is very diffi-cult to find and retain such engineers, there were prob-lems with matching products to customers.

Analog Devices, like its competitors, tried to find a better way to service its customers. Initially, projects were developed using an algorithm-based search engine combined with a standard database management system.This approach failed because this type of search returns a value only when all conditionsare exactly met. When customers provided an incomplete set of specifications, the answer was “no match”; when the customers relaxed some specifications, hundreds ofmatches were found. Thus, the process was lengthy, expensive, and error-prone.

Analog Devices decided to change the process by improving the customer–companycommunication. Using an IT-based intelligent system called case-based reasoning,which derives conclusions from historical cases, and combining it with a decision-support optimization model, the company allows customers to specify product requirements interactively, online, in order to find the right product, or the one closestto their needs, by themselves.

With this system the customer enters specifications directly on the screen. Valuescan be numbers or information such as “the best,” “sort of,” or “less than.” The sys-tem always provides an answer, a list of the top-ten Analog Devices products thatmost closely meet the specified requirements. The customer then examines the prod-ucts online. If the customer is not satisfied, another search begins with new specifica-tions and priorities. This process can continue until the right product is identified.

The system was initiately delivered to customers with the entire catalog on a CD-ROM. Today, the system is accessible to customers through an extranet, so they canuse the inexpensive Internet to save money on communication and collaboration. Thecompany’s engineers can also use the system on the company’s intranet.

Analog Devices saved initially about $2 million a year, because the cost to produce andship 120,000 CD-ROMs was far less expensive than the paper version of the catalog.

www.analog.com

Web site of semi-conductor manu-facturer AnalogDevices.

T h e B u s i n e s s P r o b l e m

T h e I T S o l u t i o n

T h e R e s u l t s

The extranet savings are much larger since there is no need to make and ship the CD-ROMs whenever significant changes were made. In addition, the quality of the serviceprovided by Web-based intelligent systems makes a difference in this competitive mar-ket. Now, when customers call sales support they usually know exactly what they want,and what they order is exactly what they need. The company’s support engineers nowhave time to accommodate more complex customer requirements.

As the system tracks customers’ requests, a future extension will use data collectedto analyze this information and input them to an improved design of new products.Sources: Compiled from D. Kress, “AI at Work,” PC AI (March/April 1998), and from analogdevices.com(2002).

The opening case demonstrates how an intelligent system solved a difficult businessproblem by improving and expediting communication and collaboration betweenthe company and its customers. The system facilitated the work of the sales engi-neers. Previously, only well-trained engineers were successful, but it took them aconsiderable amount of time. Now much of the search is done by the customersthemselves, freeing the sales engineers to do other tasks. The opening case alsodemonstrates that the intelligent system solution was integrated with other informa-tion technologies (CD-ROM, Internet, extranet, search engine) as well as with aDSS. Case-based reasoning is one of several intelligent systems that businesses canuse to improve their operations. This case is also an example of collaborative commerce (Chapter 9).

The fundamentals of the major intelligent systems and the support they providefor problem solving and seizing business opportunities are the subjects of this chapter.We will also discuss a related application, virtual reality.

As the opening case illustrated, the introduction of an intelligent system enabled non-experts to perform a task previously done by experts. This is only one benefit of intel-ligent systems, which are the commerical applications of artificial intelligence (AI).

A r t i f i c i a l I n t e l l i g e n c e a n d I n t e l l i g e n t B e h a v i o rArtificial intelligence (AI) is a term that encompasses many definitions. Most expertsagree that AI is concerned with two basic ideas. First, it involves studying the thoughtprocesses of humans; second, it deals with representing those processes via machines(computers, robots, and so on). One well-publicized definition of AI is “behavior by amachine that, if performed by a human being, would be called intelligent.” The threeobjectives of artificial intelligence are (1) to make machines smarter, (2) to under-stand what intelligence is, and (3) to make machines more useful.

What is the meaning of the term intelligent behavior? Several capabilities are con-sidered to be signs of intelligence:

• Learning or understanding from experience

• Making sense of ambiguous or contradictory messages

• Responding quickly and successfully to a new situation

• Using reasoning to solve problems and direct actions effectively

• Dealing with complex situations

W h a t W e L e a r n e d f r o m T h i s C a s e

12.1 ARTIFICIAL INTELLIGENCE AND INTELLIGENT SYSTEMS

• Understanding and inferring in ordinary, rational ways

• Applying knowledge to manipulate the environment

• Recognizing the relative importance of different elements in a situation

Although AI’s ultimate goal is to build machines that will mimic human intelligence,current intelligent systems found in commercial AI products are far from exhibitingany significant intelligence. Nevertheless, intelligent systems are currently conductingmany tasks that require some human intelligence, for a significant improvement ofproductivity, quality, and cycle time.

An interesting test to determine whether a computer exhibits intelligent behaviorwas designed by Alan Turing, a British AI pioneer. According to the Turing test, a com-puter could be considered “smart” only when a human interviewer, conversing with bothan unseen human being and an unseen computer, cannot determine which is which.

So far we have concentrated on the notion of intelligence. According to anotherdefinition, artificial intelligence is the branch of computer science that deals with waysof representing knowledge using symbols in addition to numbers and using heuristics(rules of thumb) rather than just algorithms for processing information.

Knowledge and AI. AI is frequently associated with the concept of knowledge. Thecomputer cannot have experiences or study and learn as the human mind can, but itcan use knowledge given to it by human experts. Such knowledge consists of facts,concepts, theories, heuristic methods, procedures, and relationships. As defined inChapters 1 and 11, knowledge is information organized and analyzed to make it un-derstandable and applicable to problem solving or decision making, and to incorpo-rate procedures, ideas, and human experience. The collection of knowledge related toa problem (or an opportunity) to be used in an intelligent system is organized andstored in what we call a knowledge base, and it is specific to a problem.

C o m p a r i n g A r t i f i c i a l a n d N a t u r a l I n t e l l i g e n c eThe potential value of AI can be better understood by contrasting it with natural(human) intelligence. AI has several commercial advantages over natural intelligence:

• AI is more permanent. Natural intelligence is perishable from a commercial stand-point, because workers may take knowledge with them when they leave their placeof employment, or they may forget their knowledge. AI, however, is permanent aslong as the computer systems and programs remain unchanged.

• AI can be less expensive than natural intelligence. There are many circumstances inwhich developing or buying an intelligent system costs less than having human be-ings carry out the same tasks, as was shown in the opening case.

• AI is consistent and thorough. Natural intelligence is erratic because people are er-ratic; they may not perform consistently.

• AI can be documented. Decisions made by a computer can be easily documented bytracing the activities of the system. Natural intelligence is difficult to document.

• AI offers ease of duplication and dissemination. Transferring a body of knowledgefrom one person to another usually requires a lengthy process of apprenticeship;some expertise can never be duplicated completely. Knowledge embodied in a com-puter system can be copied and easily moved to another computer, anywhere andany time. For example, the knowledge needed to match customers’ needs with Ana-log Devices’ products is now available on the extranet for the company’s customersto use when needed; knowledge distribution can be made rapidly and inexpensivelyto thousands of customers worldwide.

392 Chapter 12 Intelligent Systems in Business

On the other hand, natural intelligence has severaladvantages over AI:

• Natural intelligence is creative, whereas AI is ratheruninspired. The ability to acquire knowledge is in-herent in human beings. But with AI, tailored knowl-edge must be built into a carefully constructedsystem.

• Natural intelligence enables people to benefit fromand directly use sensory experiences. Many AI sys-tems must first interpret information collected bysensors, thus providing users with indirect sensoryexperiences.

• Natural intelligence enables people to recognize relationships between things, to sense qualities, and to spot patterns that explain how various items interrelate.

• Perhaps most important, human reasoning is alwaysable to make use of a wide context of experiences andbring that to bear on individual problems. In contrast, AI systems typically gaintheir power by having a very narrow focus.

Despite their limitations, AI methods can be extremely valuable. They can makecomputers easier to use and make knowledge more widely available. Furthermore,with the passage of time, the magnitude of these limitations is decreasing. The majorpotential benefits of AI are shown in Manager’s Checklist 12.1.

C o n v e n t i o n a l v e r s u s A I C o m p u t i n gConventional computer programs are based on algorithms, mathematical formulas orsequential procedures that lead to a solution. An algorithm is converted into a com-puter program that tells the computer exactly what operations to carry out in order tosolve problems. Conventional computing is therefore done by numerical processing.AI programs go beyond conventional computing by including heuristics, or rules ofthumb that express knowledge.

AI software also uses symbolic processing of knowledge. In AI, a symbol can be aletter, word, or number that represents objects, processes, and their relationships. Ob-jects can be people, things, ideas, concepts, events, or statements of facts. Using sym-bols, it is possible to create a knowledge base of facts and concepts, and therelationships that exist among them.

Section 12.1 Artificial Intelligence and Intelligent Systems 393

Manager’s Checklist 12.1

The Potential Benefits of Artificial Intelligence

• Makes the use of some computer applications very friendly.• Significantly increases the speed and consistency of problem solving.• Helps solve problems that cannot be solved by conventional computing.• Helps solve problems that have incomplete or unclear data.• Helps in handling the information overload (by summarizing or interpreting

information).• Significantly increases the productivity of performing many tasks.• Helps in searching and finding relationships among large amounts of data.• Facilitates the delivery of fast customer service at a low cost.

The major differences between AI computing and conventional computing areshown in Table 12.1.

Does a computer really think? Knowledge bases and search techniques certainlymake computers more useful, but can they really make computers more intelligent?The fact that most AI programs are implemented by search and pattern-matchingtechniques leads to the conclusion that computers are not really intelligent. You givethe computer a lot of information and some guidelines about how to use this informa-tion. The computer can then come up with a solution. But all it does is test the variousalternatives and attempt to find some combination that meets the designated criteria.The computer appears to be “thinking”’ and often gives a satisfactory solution. How-ever, the human mind is just too complex to duplicate. Computers certainly cannotthink in the same way humans do, but they can be very useful for increasing our pro-ductivity. This is done by several commercial AI technologies.

C o m m e r c i a l A r t i f i c i a l I n t e l l i g e n c e S y s t e m sThe development of machines that exhibit intelligent characteristics draws on severalsciences and technologies, ranging from linguistics to mathematics. The major intelli-gent systems are: expert systems, natural language processing, speech understanding,robotics and sensory systems, computer vision and scene recognition, intelligent com-puter-aided instruction, fuzzy logic, neural computing, and case-based reasoning. Acombination of two or more of the above is considered a hybrid intelligent system. Anoverview of some of these intelligent systems follows.

Expert systems. Expert systems (ESs) are computerized advisory programs that at-tempt to imitate the reasoning processes of experts in solving difficult problems. Ex-pert systems are of great interest to organizations because they can increaseproductivity and augment work forces in specialty areas where it is becoming increas-ingly difficult to find and retain human experts. Expert systems are further discussedin Section 12.2.

Natural language technology. Natural language processing (NLP) gives computerusers the ability to communicate with computers in human languages. The field of nat-ural language processing is discussed in detail in Section 12.3.


Tab le 12.1 Convent iona l versus AI Comput ing

D imens i on Conven t i ona l A I

Processing Primarily algorithmic Includes symbolic conceptualization

Nature of input Must be complete Can be incomplete

Search approach Frequently based on Frequently uses rules and heuristicsalgorithms (rules of thumb)

Explanation Usually not provided Provided

Focus Data, information Knowledge

Maintenance and Usually difficult Relatively easy changes can be madeupdate in self-contained modules

Reasoning capability No Yes

Speech (voice) understanding. Speech understanding is the recognition and under-standing by a computer of a spoken language. Details are given in Section 12.3.

Robotics and sensory systems. Sensory systems such as vision recognition combinedwith AI define a broad category of systems generally referred to as robotics. A robotis an electromechanical device that can be programmed and repro-grammed to automate manual tasks.

Robots combine sensory systems with intelligent systems and me-chanical motions to produce machines of widely varying abilities. Ro-botics is used mainly in welding, painting, cleaning, and simple materialhandling. Assembly-line operations, particularly those that are highlyrepetitive or hazardous, are also beginning to be performed by robots.Robots are used for finding, moving, and packing items in automated e-commerce warehouses, for example. Robots are getting more andmore capable. They are being put to use in ways that ease life by per-forming tasks in hazardous environments and by performing household tasks rangingfrom cutting the grass in your backyard, to cooking, to cleaning the floor.

Computer vision and scene recognition. Visual recognition has been defined as theaddition of some form of computer intelligence and decision making to digitized vi-sual information received from a machine sensor. The resultant information is thenused to perform or control such operations as robotics movement, conveyor speeds,and production-line quality control. The basic objective of computer vision is to inter-pret scenarios. Computer vision is used extensively in performing industrial-qualitycontrol tasks (such as inspection of products). Would you believe that every Tylenolor other brand-name pill is checked by computer vision for defects? Defective pillsare removed.

Intelligent computer-assisted instruction. Computer-assisted instruction (CAI),which has been in use for several decades and now is the base of e-learning, brings thepower of the computer to the educational process. Now CAI methods are being ap-plied to the development of intelligent computer-assisted instruction (ICAI) systemsthat can tutor humans by shaping their teaching techniques to fit the learning patternsof individual students. To a certain extent, such a machine can be viewed as an expertsystem. However, the major objective of an expert system is to render advice, whereasthe purpose of ICAI is to teach.

ICAI applications are not limited to schools. As a matter of fact, they have founda sizable niche in the military and corporate sectors. ICAI systems are being usedtoday for such various tasks as problem solving, simulation, discovery, learning, drilland practice, games, and testing. Such systems are also used to support people withphysical or learning impairments. An increasing number of ICAI programs are nowoffered on the Internet and intranet, supporting virtual schools and universities. ICAIcan be combined with distance learning, learning situations in which teachers and stu-dents are in different locations. Another application of ICAI is interpretive testing.Using this approach, GMAT (the MBA admission test) and other infamously longtests have shortened their length of testing time. By being able to better interpret theanswers, the test can more accurately pinpoint the strengths and weaknesses of thetest takers by asking fewer but more relevant questions.

Machine learning. Conventional computerized problem-solving techniques cannotsolve complex problems where specialized knowledge is needed. Such knowledge canbe provided, in some cases, by an expert system (ES). However, the use of an ES islimited by factors such as its rule structure, difficulties in knowledge acquisition, and

Section 12.1 Artificial Intelligence and Intelligent Systems 395

Some nonbusiness uses forrobots: (A) In August 1997,the first World Cup RobotSoccer Competition was con-ducted in Nagoya, Japan.Competing robots includedseveral AI technologies. (Seerobocup.org.)

the inability of the ES to learn from experiences. For situations where an ES is inap-propriate, machine learning can be used. Machine learning refers to a set of methodsthat attempt to teach computers to solve problems or to support problem solving byanalyzing (learning from) historical and current cases. This task, however, is not sim-ple. One problem is that there are many models of learning, and sometimes it is diffi-cult to match a learning model with the type of problem that needs to be solved.Three methods of machine learning—neural computing, case-based reasoning, andfuzzy logic—are described in Section 12.3.

Handwriting recognizers. The dream of every post office in the world is to be able toautomate the reading of all handwritten address characters, regardless of their shape.Today’s scanners are good at “reading” typed or printed material, but they are notvery good at handwriting recognition. Handwriting recognition is supported by tech-nologies such as expert systems and neural computing and is available in some pen-based computers. Handwriting interfaces are especially popular with nontypistsbecause they can convert handwritten text into typed digital text. Of special interestare products such as Pen Computing (see Chapter 3 and Palm.com).

Intelligent agents. One of the most interesting applications of intelligent systems istheir inclusion in intelligent agents, which perform a variety of tasks for their mastersmuch like a human agent does. Intelligent agents, as we will see in Section 12.4 and asdemonstrated in Chapter 9, are extremely important for e-commerce and other Web-based applications.

Other applications. Artificial intelligence can be applied to several other tasks suchas automatic programming, which automatically writes computer programs in re-sponse to and in accordance with the specifications of a program developer. Recently,


A b o u t B u s i n e s sA b o u t B u s i n e s sBox 12.1: General Electric’s SCISOR analyzes financial news

General Electric’s Research and Development Centerhas developed a natural language system called SCISOR(System for Conceptual Information Summarization,Organization, and Retrieval) that performs text analysisand question-answering in a limited, predefined subjectarea (called a constrained domain). One application ofthis system deals with analyzing financial news. For ex-ample, SCISOR automatically selects and analyzes sto-ries about corporate mergers and acquisitions from theonline financial service of Dow Jones. It is able toprocess news in less than 10 seconds per story. First, itdetermines whether the story is about a corporatemerger or acquisition. Then, it selects information suchas the target, suitor, and price per share. The system al-lows the user to browse and ask questions such as,“What price was offered for Polaroid?” or “How muchwas Bruck Plastics sold for?”

The system’s effectiveness was demonstrated in test-ing, when it proved to be 100 percent accurate in identi-

fying all 31 mergers and acquisitions stories that were in-cluded in a universe of 731 financial news releases fromthe newswire service.

A similar application is a Web-based personalizednews system that was developed in Singapore to trackbusiness news available in English, Chinese, and Malay,summarize it, and extract desired personalized news inany of these languages.

Questions

1. What are the benefits of analyzing financial newsvia a machine?

2. What other applications might be developed withthis type of system?

3. How could such a system be combined with an In-ternet news dissemination portal such asmoney.cnn.com?

‘s‘s ge.com FIN POM

an automatic translation of Web pages to other languages has become very popular.Other applications are preparation of news summaries and translation from onehuman or computer language to another. Some computer programs, for example,“read” stories in newspapers or other documents, including those available on the In-ternet, and make summaries in English or other languages. This capability helps han-dle the problem of information overload, as described in IT’s About Business 12.1.

Of all the intelligent systems, the one with the most business applications is the expertsystem. When an organization has a complex decision to make or a problem to solve,it often turns to experts for advice. These experts have specific knowledge and experi-ence in the problem area. They are aware of alternative solutions, chances of success,and costs that the organization may incur if the problem is not solved. Companies en-gage experts for advice on many matters, ranging from mergers and acquisitions toadvertising strategy. Experts can diagnose problems correctly and solve them satisfac-torily within a reasonable time frame. The more unstructured the situation, the morespecialized is the advice. However, human experts are expensive, and they may not bereadily available. Expert systems (ES) are an attempt to mimic human experts.

C o n c e p t s o f E x p e r t S y s t e m sIn order to explore the concepts involved in ES, read IT’s About Business 12.2, whichdescribes a well-known application case at General Electric. The case of GE’s DavidSmith demonstrates that the basic idea behind an ES is simple: Expertise is trans-ferred from an expert, or other sources of expertise, to a computer and is stored there.Expertise is the extensive, task-specific knowledge acquired from training, learning,and experience. It enables experts to make better and faster decisions than non-experts in solving complex problems. Expertise takes a long time (usually years) to ac-quire. Users can call on the computer’s stored expertise for specific advice as needed.The computer can make inferences and arrive at a conclusion. Then, like a human ex-pert, the computer program advises the nonexperts and explains, if necessary, thelogic behind the advice. When they contain the wisdom of several experts, expert sys-tems can sometimes perform better than any single expert can.

The goal of an expert system is to transfer expertise from an expert and docu-mented sources to a computer and then to the user. This process involves four activi-ties: knowledge acquisition (from experts or other sources), knowledge representation(in the computer), knowledge inferencing, and knowledge transfer and use to solve aproblem. Through the activity of knowledge representation, acquired knowledge isorganized as rules or objects, and is stored electronically in a knowledge base.

B e f o r e y o u g o o n . . .

1. List the major advantages that artificial intelligence has over natural intelligence.

2. List the major disadvantages of artificial intelligence compared with natural intelligence.

3. List and briefly define the commercial AI application tools.

Section 12.2 Expert Systems 397

12.2 EXPERT SYSTEMS

A unique feature of an expert system is its ability to “reason.” Given the neces-sary expertise stored in the knowledge base and accessibility to databases, the com-puter is programmed so that it can make inferences, based on a search-and-matchprocess. Knowledge inferencing is performed in a component called the inference en-gine and results in advice or a recommendation for novices. ES can explain its recom-mendation through a subsystem called the justifier or the explanation facility.

During the past few years, the technology of expert systems has been successfullyapplied in thousands of organizations worldwide to problems ranging from identifyingcredit card fraud to medical diagnosis to the analysis of dust in mines. Two illustrativeapplications are listed next.

Helping the Navajo Nation. The states of Arizona, New Mexico, and Utah aretransferring management of the welfare program to the Navajo Nation, which nowself-adminsters the program for its own people. The program provides financial andhuman services to approximately 28,000 Navajo clients. An expert system facilitatesself-management of the welfare program. The interactive solution Case Worker Advi-sor (from exsys.com) integrates the tribe’s unique cultural heritage while followingcomplex tribal, federal, and state guidelines.

Carrier configures equipment orders. Carrier Corporation, a major air conditioningmanufacturer, introduced expert systems into its operations. For each customized

EXAMPLES


A b o u t B u s i n e s sA b o u t B u s i n e s sBox 12.2: GE’s expert system models human troubleshooters

General Electric’s top locomotive field service engineer,David I. Smith, had been with the company for morethan 40 years and was expert at troubleshooting dieselelectric locomotive engines. Smith traveled throughoutthe country to places where locomotives were in need ofrepair to determine what was wrong and to advise youngengineers. The company was dependent on Smith. Therewas just one problem: Smith was nearing retirement.

GE’s traditional approach to such a situation was toset up apprenticeship teams that paired senior and juniorengineers for several months or even years. By the timethe older engineers retired, the younger engineers hadabsorbed enough of their expertise to carry on. It was agood short-term solution, but GE still wanted a more ef-fective and dependable way of disseminating expertiseamong its engineers and preventing valuable knowledgefrom retiring with people like David Smith.

GE decided to build an expert system to solve theproblem by modeling the way a human troubleshooterworks. The system builders spent several months inter-viewing Smith and transferring his knowledge to a com-puter. The computer program was developed over athree-year period, slowly increasing the knowledge andnumber of decision rules stored in the computer. The re-sulting diagnostic technology enables a novice engineer

or even a technician to uncover a fault by spending onlya few minutes at the computer terminal. The system canalso explain to the user the logic of its advice, serving asa teacher. Furthermore, the system can lead usersthrough the required repair procedures, presenting de-tailed, computer-aided drawings of parts and subsystemsand providing specific how-to instruction demonstra-tions. It is based on a flexible, humanlike thoughtprocess, rather than rigid procedures expressed in flow-charts or decision trees.

The system is currently installed at every railroad re-pair shop served by GE, thus eliminating delays, preserv-ing Smith’s expertise, and boosting maintenanceproductivity. After his retirement, Smith was hired as aconsultant to help in updating and maintaining the sys-tem’s knowledge.

Questions

1. Why was this application ideal for an expert sys-tem approach?

2. What are the major advantages of the computer-ized ES?

3. How can GE keep the knowledge in the system upto date after Smith stops his consultancy?

‘s‘s ge.com POM HRM

order, Carrier must procure all the parts and subsystems so that orders can be filled ontime. The ES configures a set of part numbers for each customer’s equipment order.Using the ES, Carrier was able to minimize both pricing and configuration errors, reduce cycle time, and increase customer satisfaction as well as profitability (car-rier.com). In summer 2001, Carrier teamed up with IBM to create a Web-enabled airconditioner that allows customers to turn on their home air conditioners before theyreach home. Commands are given from cellphones or PCs, through myappliance.com.The system also can monitor performance for early detection of problems. ●



The Benefits and Limitations of Expert Systems

• Increased output and productivity:Many tasks (e.g., diagnosis) can beperformed much faster.

• Increased quality and reliability:Consistent output, lower error rate,important data not overlooked.

• Capture of scarce expertise: Topexperts’ knowledge is captured fordissemination to many.

• Ability to operate in hazardousenvironment: ES can be installed on robots that operate in toxic andother dangerous environments.

• Improved customer service: Fastaccess to information and to FAQsfacilitates customer service by help-desk employees.

• Humanlike intelligence: Can makeother computer systems smarter andmore powerful when embedded inthem.

• Facilitated fault tolerance: Can workwith incomplete input informationand generate good conclusions evenif some input data are missing.

• Complex problem solving anddecision making: Can integratemultiple opinions, quickly analyzedata, and suggest solutions.

• Training capabilities: Can simulatedecisions and explain reasoningbehind the decisions.

• Reduction of cycle time anddowntime: Advice is availableanywhere, at any time, and analysiscan be performed quickly.

• Embedded systems: ESs are easilyembedded in thousands ofelectrical, mechanical, and ITsystems.

• Limited expertise: Knowledge to becaptured is not always readilyavailable; expertise can be hard toextract from humans. In someinstances, experts may refuse tocontribute their knowledge, or maycontribute incorrect or incompleteknowledge.

• No single correct solution: Theapproach of each expert to a givensituation may be different, yet correct.

• Natural cognitive limits: Users ofexpert systems may not use thebenefits of the system to the fullestextent because of limitedunderstanding.

• Narrowly defined subject areas: ESworks well only for certain taskssuch as diagnosing a malfunction ina machine.

• Occasional incorrect recommen-dations: Many expert systems haveno independent means of checkingwhether their conclusions arereasonable or correct.

• Limited vocabulary, or jargon: Theterminology that experts use forexpressing facts and relations isfrequently not understood by others.

• Cost: Help in building ESs isfrequently required from knowledgeengineers who are rare andexpensive.

• Lack of trust by end users: Usersmay not trust a machine.

• Bias: Knowledge transfer is subjectto perceptual and judgmental biases.

• Liability issues: Liability for badadvice provided by an ES is difficultto assess.

Bene f i t s L im i t a t i on s

B e n e f i t s a n d L i m i t a t i o n s o f E x p e r t S y s t e m sWhy have ESs become popular? Because of the large number of capabilities and ben-efits they provide at a reasonable cost. However, available ES methodologies are notalways straightforward and effective, and some problems have slowed the commercialspread of ES. Despite some limitations, though, the use of ES is growing rapidly. Themajor benefits and limitations are compiled in Manager’s Checklist 12.2.

T h e P r o c e s s a n d C o m p o n e n t s o f E x p e r t S y s t e m sThe process in which expert systems are constructed and used is indicated in Figure12.1 and described here. The major component parts of expert systems shown in thefigure are also discussed.

The process of ES. The process of ES can be divided into two parts: the system de-velopment environment and the consultation environment. The development environ-ment (shown on the left side of Figure 12.1) is the part in which the ES is constructed.The consultation environment (shown on the right side of Figure 12.1) describes howadvice is rendered to the users.

The development process starts with a knowledge engineer, who can also be thesystem builder, acquiring knowledge from experts and/or documented sources. Thisknowledge is programmed in the system’s knowledge base together with facts aboutthe subject area (domain), usually in terms of “if–then” rules.

In the consultation environment, the user contacts the system via the user inter-face to ask for advice. The ES collects information from the user, usually by askingquestions about symptoms and conditions, and then activates the inference engine,which searches the knowledge base for recommended actions. An example of a con-sultation is provided in IT’s About Business 12.3.


Figure 12.1 Structureand process of expert systems.

Recommendedaction

Knowledgerefinement

Explanationfacility

Blackboard (workplace)

User interface

Inference enginedraws conclusions

User

Knowledgeengineer

Expert anddocumentedknowledge

Facts about thespecific incident

Knowledgeacquisition

Consultation EnvironmentDevelopment System

Knowledge baseFacts: What is known about the

domain areaRules: Logical reference (e.g.,

between symptoms andcauses)

The components of ES. The following major components exist in an expert system:a knowledge base, a memory area called the blackboard, an inference engine, a userinterface, and an explanation facility. The functions of these components are brieflydescribed here.

• The knowledge base contains knowledge necessary for understanding, formulating,and solving a specific class of problems. It includes two basic elements: (1) facts,such as the problem and its various states, and (2) rules, that direct the use ofknowledge to solve the specific class of problems.

• The blackboard is an area of working memory set aside for the description of a cur-rent problem, as specified by the input data; it is also used for storing intermediateresults. It is a temporary database used by the inference engine to execute its tasks.


A b o u t B u s i n e s sA b o u t B u s i n e s sBox 12.3: How to select an advertising medium—a sample of an ES consultation

This prototype system will attempt to provide recom-mendation(s) on the advertising mix so as to maximizethe client’s product exposure in the market. Currently,the system makes recommendations on only two types ofadvertising media: television and newspaper.

Sample Printout of Consultation: The system willask the user several questions, such as the one shownbelow, to find the requirements and/or symptoms or theproblem.

The user asks the computer “Why do you need thisinformation?” The computer answers by displaying thepertinent rule (Rule #1).

The computer continues with questions such as:

After the user has answered all questions, the ESdisplays the recommendations:

Source: Printouts were generated with software from Exsys Inc.(exsys.com).

Questions

1. How can this system be adapted for electroniccommerce?

2. Would you trust the recommendations of this sys-tem more than those offered by human experts?Why or why not?

‘s‘s exsys.com MKT

Why_

80000

• The “brain” of the ES is the inference engine, which is essentially a computer pro-gram that provides a methodology for reasoning and formulating conclusions.

• The user interface in ES allows for user–computer dialogue, which can be best car-ried out in a natural language, usually presented as questions and answers, andsometimes supplemented by graphics. The question-and-answer dialogue triggersthe inference engine to match the problem symptoms with the knowledge in theknowledge base and to generate advice.

• The explanation subsystem can trace responsibility for conclusions to their source,which is crucial both in the transfer of expertise and in problem solving. This sub-system explains the ES’s behavior by interactively answering questions such as thefollowing: Why was a certain question asked by the expert system? How was a cer-tain conclusion reached? Why was a certain alternative rejected? What is the planto reach the solution?

Human experts have a kind of knowledge-refining system with which they analyzetheir own performance, learn from it, and improve it for future consultations. Simi-larly, such evaluation is necessary in computerized learning so that the program willbe able to improve by analyzing the reasons for its success or failure. Such a compo-nent is not currently available in commercial expert systems, but it is being developedin experimental expert systems.

I l l u s t r a t i v e A p p l i c a t i o n sExpert systems are in use today in all types of organizations. They are especially use-ful in certain generic categories, displayed in Table 12.2. The following examples illus-trate the diversity and nature of ES applications.


Tab le 12. 2 Gener i c Categor ies of Expert Systems

Ca t ego ry Ta sk o r P rob l em Add re s s ed

1. Interpretation Inferring situation descriptions from observations

2. Prediction Inferring likely consequences of given situations

3. Diagnosis Inferring system malfunctions from observations

4. Design/configuration Configuring objects under constraints

5. Planning Developing plans to achieve goal(s)

6. Monitoring Comparing observations to plans and flagging exceptions

7. Debugging Detecting problems and inconsistencies and prescribingremedies for malfunctions

8. Repair Executing a plan to administer a prescribed remedy

9. Instruction Monitoring performance, diagnosing, debugging, andcorrecting learning

10. Control Monitoring system behavior; interpreting, predicting,repairing, and sometimes alerting operators

11. Data analysis Quantitiative and qualitative analysis of complex data

12. Customer/product support Solving help-desk problems

13. Decision support Interactive advisory systems

The U.S. Treasury fights criminals with an expert system. One of the major tasks ofthe U.S. Financial Crime Enforcement Network (FinCEN), an agency of the U.S. De-partment of the Treasury, is to prevent and detect money laundering. One area of in-vestigation is cash transactions over $10,000, which all banks must report. Theproblem is that there are more than 200,000 such transactions every week (more than12,000,000 per year). FinCEN does not have the budget or the staff to conduct theanalysis necessary to manually examine all of these transactions.

The practical solution is the use of a rule-based expert system that contains theexpertise of FinCEN’s top experts. The expert system is used to automatically detectsuspicious transactions and changes in transaction patterns. Then, these are checkedmanually. Since its inception in 1993, the expert system has helped to uncover cases ofmoney-laundering activities valued at over $250 million annually.

Ticket auditing at Northwest Airlines. When Northwest Airlines (NWA) acquiredRepublic Airlines, its volume of operations increased to 70,000 tickets per day. Thesetickets needed to be audited by comparing a copy of each ticket against fare informa-tion, including travel agent commissions. Manual comparison was slow and expensive.Therefore, only samples of the tickets (about 1%) were audited. The sample indicatedan error rate of about 10 percent (usually a loss to the airline).

NWA’s solution to this problem was to build a ticket-auditing ES that scans alltickets electronically and stores the information in a database. Another databasestores all the fares and commission agreements. Then the expert system goes to work.The ES first determines the correct fare, using only 250 rules. The most favorablecommission to travel agents is determined, and any discrepancy results in a report tothe agent with a debit or credit and an appropriate explanation. The system also pro-vides information for marketing, contract management, planning, and control. The re-duction in agent errors saves NWA about $10 million annually. ●

E x p e r t S y s t e m s a n d t h e I n t e r n e t / I n t r a n e t sThe relationship between expert systems and the Internet and intranets is a two-waystreet. The Net supports ES (and other AI) applications, and expert systems supportthe Net.

One of the justifications for an ES is the potential to provide knowledge and ad-vice to large numbers of users. The widespread use of the Internet and intranets pro-vides the opportunity to disseminate expertise and knowledge to mass audiences. Bydisseminating knowledge to many users, the cost per user becomes small, making anES very attractive. Implementing expert systems (and other intelligent systems) asknowledge servers, it becomes economically feasible and profitable to publish exper-tise on the Net, as the following examples demonstrate.

OSHA uses an expert system. The U.S. Department of Labor Occupational Safetyand Health Administration (OSHA) has an up-and-running Web-based ES. The sys-tem provides guidance to help employers protect workers from the hazards of entryinto permit-required confined spaces. The system helps determine if a space is cov-ered by OSHA’s regulation on such spaces. For example, by going to the site(osha.gov) and using the Web browser interface, you can see if you need a specialgovernment permit to work in your room. (See Problem-Solving Activity 2 at the endof the chapter.)

EXAMPLES

EXAMPLES


Customer service online in Japan. Ebara Manufacturing, a Japanese pump manu-facturer, produces several thousand kinds of pumps for many industries. Tradition-ally, customers would just ask for a pump. Sales personnel had to figure out what kindwas really needed. As products got more numerous and sophisticated, this manualsystem started to break down. Ebara fixed this problem with an online expert system.The system takes customers through a series of questions that connect their needs tospecific products, often in less than a minute. ●

An expert system’s major objective is to provide expert advice. However, expert sys-tems can be used only for the special situations for which they have been developed.Other intelligent systems can broaden the range of applications. Four such technolo-gies are described next.

N a t u r a l L a n g u a g e P r o c e s s i n g a n d V o i c e T e c h n o l o g yTypically, when you want to tell a computer what to do, you type commands on thekeyboard using pre-determined commands, or you click on an icon. In responding toyour commands, the computer outputs message symbols or other short, cryptic notesof information. Many problems would be minimized or even eliminated if we couldcommunicate with the computer in our own languages, rather than the command-language that at this point is the necessary intermediary. We would simply type in di-rections, instructions, or information in our natural language. Better yet, we wouldconverse with the computer using voice. The computer would be smart enough to in-terpret the input, regardless of its format. Natural language processing (NLP) refers tocommunicating with a computer in English or whatever language you speak.

To understand a natural language inquiry, a computer must have the knowledgeto analyze and then interpret the input. This may include linguistic knowledge aboutwords, domain (subject area) knowledge, common-sense knowledge, and even knowl-edge about the users and their goals. Once the computer understands the input, it cantake the desired action. In this section we briefly discuss applications of NLP pro-grams, voice recognition, voice portals, and voice generation.

Applications of natural language processing. Natural language processing programshave been applied in several areas. The most important are human–computer inter-faces (mainly to databases), abstracting and summarizing text, grammar analysis,translation of a natural language to another natural language, translation of a com-puter language to another computer language, speech understanding, and letter com-position. By far the most dominant use of NLP is in interfaces, or “front-ends,” toother software packages, especially databases.

Voice (speech) recognition and understanding. Voice (or speech) recognition is aprocess that allows users to communicate with a computer by speaking to it. The term


1. List and briefly describe the major components of an ES.

2. What are the potential advantages of an ES?

3. List the 13 generic categories of an ES.


12.3 OTHER INTELLIGENT SYSTEMS

speech recognition is sometimes applied only to the firstpart of the process—recognizing words that have been spo-ken without necessarily interpreting their meanings. Theother part of the process, wherein the meaning of speech isascertained, is called speech understanding. It may be possi-ble to understand the meaning of a spoken sentence with-out actually recognizing every word, and vice versa. Whena speech recognition system is combined with a natural lan-guage processing system, the result is an overall system thatnot only recognizes voice input but also understands it.

The ultimate goal of voice recognition is to allow a com-puter to understand the natural speech of any humanspeaker at least as well as a human listener could under-stand it. In addition to the fact that this is the most naturalmethod of communication, voice recognition offers severalother advantages, as shown in Manager’s Checklist 12.3.

There are a few limitations to voice recognition. Themajor limitation is the inability to recognize long sentences, or to recognize speechfast enough. The better the system in this respect, the higher the cost. An additionallimitation of speech recognition systems is that they do not (yet) interface well withicons and windows, so speech may need to be combined with the keyboard operation,which slows down the system.

Voice portals. Use of an ordinary telephone as an Internet appliance is known as avoice portal. Customers dial a toll-free number and use voice to request informationranging from a traffic report to stock prices. That is, the site the customer reaches throughthe phone acts like an Internet portal (like Yahoo! or Lycos). The difference is that infor-mation is accessed by voice rathan than by pointing and clicking a mouse. One of the firstapplications was deployed at autobytel.com. Customers use the portal to obtain informa-tion and transact purchases. The system greets customers (by name if the company knowsyour name) and asks what kind of a car you are interested in buying. The major voiceportal companies in 2001 were Heyanita.com, Tellme.com, and Telsurf.com.

Voice generation (voice synthesis). Natural language generation strives to allow com-puters to produce ordinary English language, on the screen or by voice so that people

Section 12.3 Other Intelligent Systems 405


Benefits of Voice Recognition• Ease of access: Many more people can speak than can type. As long ascommunication with a computer depends on developing typing skills, manypeople may not be able to use computers effectively.

• Speed: Even the most competent typists speak more quickly than they type.It is estimated that the average person can speak twice as quickly as aproficient typist can type.

• Manual freedom: There are many situations in which computers might beuseful to people whose hands are otherwise occupied, such as productassemblers, pilots of military aircraft, and busy executives.

• Remote access: Many computers are set up to be accessed remotely bytelephone. If a remote database includes speech recognition capabilities,you could retrieve information by issuing oral commands into a telephone.

• Accuracy: In typing information, people are prone to make mistakes,especially in spelling. These are minimized with voice input.

can understand computers more easily. The technology by which computers speak isknown as voice synthesis. The synthesis of voice by computer differs from the simpleplayback of a prerecorded voice by either analog or digital means. As the term synthesisimplies, sounds that make up words and phrases are constructed electronically frombasic sound components and can be made to form any desired voice pattern.

In July 2001, AT&T Labs started selling speech software called Natural Voicesthat is so good at reproducing the sound, inflections, and intonations of a human voicethat it can recreate voices—even the voices of long-dead celebrities or any one forwhich a sample voice is available. Priced at thousands of dollars, the software is sellingto producers of animated movies, publishers of video games, car manufacturers thatprovide voice driving directions, and books-on-tape publishers.

Applications of voice technology. As voice recognition and synthesis technology hasimproved in recent years, many applications have been developed for commercial use.Table 12.3 lists a sampling of such applications via input or output devices; two moreare presented in more detail below.


Tab le 12.3 Sample of Vo i ce Techno logy App l i cat ions

Company App l i c a t i on

Scandinavian Airlines, other airlines Answering inquiries about reservations, schedules, lost baggage, etc.a

Citibank, many other banks Informing credit card holders about balances and credits, providing bankaccount balances and other information to customersa

Delta Dental Plan (CA) Verifying coverage informationa

Federal Express Requesting pickups, responding to inquiries about delivery of specificshipmentsa,b

Illinois Bell, other telephone companies Giving information about services, receiving ordersa,b

Domino’s Pizza Enabling stores to order supplies, providing price informationa,b

General Electric, Rockwell Allowing inspectors to report results of quality assurance testsb

International, Austin Rover,Westpoint Pepperell, Eastman Kodak

Cara Donna Provisions Allowing receivers of shipments to report weights and inventory levels ofvarious meats and cheesesb

Weidner Insurance, AT&T Conducting market research and telemarketingb

U.S. Department of Energy, Idaho Notifying people of emergencies detected by sensorsa

National Engineering Laboratory,Honeywell

New Jersey Department of Education Notifying parents when students are absent and about cancellation of classesa

Kaiser-Permanente Health Calling patients to remind them of appointments, summarizing and reportingFoundation (HMO) resultsa

Car manufacturers Activating radios, heaters, and so on, by voiceb

Texoma Medical Center Logging in and out by voice to payroll departmentb

St. Elizabeth’s Hospital Prompting doctors in the emergency room to conduct all necessary tests, reporting results by doctorsa,b

Hospital Corporation of America Sending and receiving patient data by voice, searching for doctors, preparingschedules and medical recordsa,b

aVia an output device.bVia an input device.

Minnesota wrestles with tax inquiries. With limited staffing and many Minnesotaresidents without touch-tone dialing service, the Department of Revenue needed tofind an innovative solution to meet its increasing tax inquiry phone load and expandits service. The department put speech recognition technology into place with an in-teractive voice system via the telephone. As a result, it was able to respond immedi-ately to an additional 100,000 taxpayer phone inquiries per year that could not havebeen handled otherwise.

American Express books flights. American Express Travel Related Services(AETRS) uses a voice recognition system that allows its customers to check and bookdomestic flights by talking to a computer over the phone. The system asks customersquestions such as: Where do you want to travel to? From where? When? and so on.The system can recognize 350 city and airport names, and it lets callers use more than10,000 different ways to identify a location. Compared to telephone service by an op-erator, reservation transaction cost is reduced by 50 percent. The average transactiontime is reduced from 7 to 2 minutes. AETRS offers a similar service on the Web. ●

N e u r a l C o m p u t i n gThe tools of AI described so far have been mostly restricted to stored knowledge andlogic. A different approach is intelligent systems that use architecture that mimicscertain processing capabilities of the brain. The results are knowledge representa-tions and processing based on massive parallel processing rather than sequential in-formation processing, fast retrieval of large amounts of information, and the abilityto recognize patterns based on experiences. One technology that attempts to achievesuch results is called neural computing or artificial neural networks (ANNs).

An artificial neural network is a computer model that emulates a biological neuralnetwork. Today’s neural computing uses a very limited set of concepts from biologicalneural systems to implement software simulations of massively parallel processes. An

EXAMPLES


Figure 12.2 Neural Internet-based optical character recognizer.

Featureextractionneurons

Classifierneurons

Recognizedcharacter

Optical Character Recognizer (OCR) Neural Net

0 1 2 3 4 5 6 7 8 9

artificial neuron receives inputs analogous to the electrochemical impulses that bio-logical neurons receive from other neurons. The neurons in an ANN receive informa-tion from other neurons or from external sources, transform the information, and passit on to other neurons or as external outputs.

The value of neural network technology includes its usefulness for pattern recog-nition, learning, and the interpretation of incomplete inputs. Figure 12.2 is a simplifiedexplanation of how a neural network can recognize characters. Sets of neurons extractfeatures from the input images. (Here, neurons extract the locations of vertical, hori-zontal, and diagonal strokes.) The locations of these features indicate possible choicesof the character class. Most of the evidence shows that 6 is the best choice.

Neural networks have the potential to provide some of the human characteristicsof problem solving that are difficult to simulate using the logical, analytical techniquesof DSSs or even expert systems. For example, neural networks can analyze largequantities of data to discover patterns and characteristics in situations where the logicor rules are not known. An example would be loan applications. By reviewing manyhistorical cases of applicants’ responses to questionnaires and the granting decisions(yes or no), the ANN can create “patterns” or “profiles” of applications that shouldbe approved, or those that should be denied. A new application is matched against thepattern. If it comes close enough, the computer classifies it as a “yes” or “no”; other-wise it goes to a human to decide. Applications can thus be processed more rapidly.The benefits of neural networks are summarized in Manager’s Checklist 12.4.

Specific business areas that are well-suited to the assistance of ANNs include thefollowing:

• Data mining: Finding data in large and complex databases, as explained in Chap-ter 11

• Tax fraud: Identifying, enhancing, and finding irregularities

• Financial services: Identification of patterns in stock market data and assistance inbond trading strategies, mortgage underwriting, and foreign rate exchange forecast

• Loan application evaluation: Judging worthiness of loan applications based onpatterns in previous application information (e.g., customer credit scoring)

• Solvency prediction: Assessing the strengths and weaknesses of corporations andpredicting possible failures

• New product analysis: Sales forecasting and targeted marketing evaluation

• Airline management: Seat demand forecasting and crew scheduling

• Prediction of consumer behavior on the Internet: Predicting consumer behavior inorder to plan e-commerce advertisement



Benefits of Neural Networks • Pattern recognition: Can analyze large quantities of data to establishpatterns and characteristics in situations where the logic or rules are notknown.

• Fault tolerance: Since there are many processing nodes, damage to a fewnodes or links does not bring the system to a halt.

• Generalization: When a neural network is presented with an incomplete orpreviously unseen input, it can generalize to produce a reasonable response.

• Adaptability: The network learns in new environments. New cases are usedimmediately to retrain the program and keep it updated.

• Forecasting capabilities: Similar to statistics, predictions can be made basedon historical data.

• Evaluation of personnel and job candidates: Matching personnel data to job re-quirements and performance criteria

• Resource allocation based on historical, experiential data: Finding allocationsthat will maximize outputs

• Foreign exchange rate evaluation: Evaluating exchange rates of various curren-cies, including country risk rating

• Identifying takeover targets: Predicting which companies are most likely to be ac-quired by other companies

• Stocks, bonds, and commodity selection and trading: Analyzing various invest-ment alternatives, including pricing of initial public offerings

• Signature validation: Matching against previous signatures

• Human resources prediction: Predicting employee performance and behavior andanalyzing personnel requirements

• Credit card fraud detection: Detecting fraud by analyzing purchasing patterns.

For a more specific example of the use of neural network technology in a credit cardfraud detection application, see IT’s About Business 12.4.

C a s e - b a s e d R e a s o n i n gThe idea of case-based reasoning (CBR) is to adapt successful solutions used in thepast in order to solve new problems. Case-based reasoning first finds the solutions thatsolved problems similar to the current problem. It then adapts the previous solution


A b o u t B u s i n e s sA b o u t B u s i n e s sBox 12.4: Visa cracks down on credit card fraud

Only 0.2% of Visa International’s revenues in 1995 werelost to fraud, but when that percentage means a loss of$655 million, it is well worth addressing. In 1996, Visabegan using neural network technology in order to re-verse the number of fraudulent transactions. By 2002 thecompany was using the system globally.

Most people stick to well-established patterns ofcard use and only rarely splurge on expensive nonessen-tials. Neural networks are designed to notice when acard that is usually used to buy gasoline once a week issuddenly used to buy a number of tickets to the latesttheater premiere on Broadway.

Bank of America field-tested Visa’s cardholder riskidentification system (CRIS) and believes that the sys-tem cuts fraudulent card use by up to two-thirds. TheToronto Dominion Bank found that losses were re-duced, and overall customer service improved, with theintroduction of neural computing. Another bankrecorded savings of $5.5 million in six months. Visamember banks cut their losses by more than 16 percentin the first year of the system’s use. With numbers likethat, the $2 million Visa spent to implement CRIS cer-

tainly seems worth the investment. In fact, Visa says,CRIS paid for itself in less than a year.

In 1995, CRIS conducted more than 16 billion trans-actions. In 2001, VisaNet (a data warehouse and e-mailoperation) and CRIS handled more than 200 billiontransactions.

The only downside to CRIS is that occasionally thesystem prompts a call to a cardholder’s spouse when anout-of-the-ordinary item, such as a surprise vacation tripor a diamond ring, is charged. After all, no one wants tospoil surprises for loved ones.

Questions

1. How does the system detect fraud?

2. What is the advantage of CRIS over an automaticcheck against the balance in the account andagainst a set of rules such as “Call a human autho-rizer when the purchase price is more than 200percent of the average previous bill”?

3. Why do you think neural networks were used butnot expert systems?

‘s‘s visa.com ACC FIN

to fit the current problem, taking into account any differences between the currentand previous situations. Finding relevant cases involves several steps: (1) characteriz-ing the input problem by assigning appropriate features to it, (2) retrieving frommemory the cases with those features, and (3) picking the case or cases that bestmatch the input.

Case-based reasoning has proved to be an extremely effective approach in com-plex cases. The basic justification for the use of CBR is that it processes the right in-formation retrieved at the right time. Case-based reasoning can be used by itself orcan be combined with other reasoning paradigms. Target applications include tacti-cal planning, political analysis, situation assessment, legal planning, diagnosis, frauddetection, design/configuration, message classification, and complex searches. Ana-log Devices’ IT system described in the opening case is an example of case-basedreasoning.

F u z z y L o g i cFuzzy logic deals with uncertainties by simulating the process of qualitative humanreasoning, allowing the computer to behave less precisely and logically than do con-ventional computers. The rationale behind this approach is that decision making isnot always a matter of true or false, black and white. It often involves gray areaswhere the terms approximatly, possible and similar are more appropriate.

Take, for example, the variable “height.” Most people would agree that if you areabove 6 feet, you are tall. Similarly, if your height is less than 5 feet, you are short. Butbetween 6 feet and 5.75 feet, there is less probability that you will be considered tall.Similarly, between 5 and 5.25 feet some will consider you short. Notice that in thearea between 5.25 and 5.75 feet you have a chance for being considered either short or tall.

Currently there are only a few examples of fuzzy logic applications in business,but the results are significant improvements in productivity. More often, fuzzy logic isused together with other intelligent systems, as illustrated in the following example.

Developing marketing strategy. An international investment firm used IT systemsto develop marketing strategy. Developing marketing strategy is a complex processperformed sequentially, with contributions from corporate experts. Numerous mar-keting strategy models were developed over the years to support the process. Unfor-tunately, most of the models supported only one goal (e.g., to perform forecasting).However, one firm developed a system that integrates expert systems, fuzzy logic, andANN, shown in Figure 12.3, to solve this problem. The systems components are:

• Neural networks. These are used to predict future market share and growth.

• Expert systems. These provide intelligent advice on developing market strategy toindividuals and to a planning team.

• Fuzzy logic. This helps managers handle uncertainties and fuzzines of data and in-formation.

The integration of the technologies helps in sharing information, coordination, andevaluation. The system is designed to support both individuals and groups. ●

Any of the intelligent systems described in this and the previous sections can alsobe used as a knowledge component of an intelligent agent, the topic we turn to in thenext section.

EXAMPLE


Intelligent agents (IAs) have the potential to become one of the most important toolsof information technology in the twenty-first century. IA can alleviate the most criti-cal limitation of the Internet—information glut or overflow—and can facilitate elec-tronic commerce. Before we look at their capabilities, let’s determine what we meanby IAs.


1. Describe natural language processing and its major benefits.

2. List the major benefits of voice recognition.

3. What are the major benefits of neural computing? What are some popular applications of ANNs?

4. Define case-based reasoning and fuzzy logic.

Section 12.4 Intelligent Agents 411

Figure 12.3 The archi-tecture of hybrid intelligentsystems. [Source: S. Li,“The Development of a Hy-brid Intelligent System forDeveloping Marketing De-cision Strategy,” DecisionSupport Systems, January2000, Fig. 1, p.399.Reprinted with permissionfrom Elsevier Science.]

Market growth history Market share history

Neuralnetworks

model

Individual orgroup

assessmentmodel

Fuzzification component

Fuzzy expert systemGraphical display module

Graphical portrayalof strategic positions

Legends:

: functional module : data file : relationships

Intelligent advice onmarketing strategy

Databasemanagement

system

Future market growth

Market attractiveness information

Market attractiveness factors

Future market share

Business strength information

Mangerial judgement and intuition

Porter's five forcesBusiness strengths factors

Fuzzified business strengthFuzzified market attractiveness

12.4 INTELLIGENT AGENTS

The term agent is derived from the concept of agency, which means employingsomeone to act on your behalf. A computerized agent represents a person and inter-acts with others to accomplish a predefined task. Several names are used to describeintelligent agents: software agents, wizards, knowbots, and softbots. The names reflectthe nature of the agent. A good working definition is this: An intelligent agent is asoftware entity that senses its environment and then carries out some operations onbehalf of a user (or a program), with a certain degree of autonomy, and in so doingemploys knowledge or representation of the user’s goals or desires.

C h a r a c t e r i s t i c s o f I n t e l l i g e n t A g e n t sThere are several traits or abilities that many people think of when they discuss intel-ligent agents: capability to work on their own (autonomy); exhibition of goal-orientedbehavior; mobility (transportable over networks); dedication to a single repetitivetask ability to interact with humans, systems, and other agents; inclusion of a knowl-edge base; and ability to learn. Although not all intelligent agents have all of these ca-pabilities, they are very useful in facilitating tasks such as the following.

Information access and navigation. Information access is today’s major applicationof intelligent agents, and it is done by use of different search engines.

Decision support and empowerment. Knowledge workers need support, especiallyin decision-making. Intelligent agents can facilitate decision making and empoweremployees, as shown in IT’s About Business 12.5.

Repetitive office activities. There is a pressing need to automate repetitive tasksperformed by administrative and clerical personnel in functional areas, such as salesor customer support, in order to reduce labor costs and increase office productivity.


A b o u t B u s i n e s sA b o u t B u s i n e s sBox 12.5: Empowering employees by using intelligent agents

Fringe benefits are frequently likened to a cafeteria—people mix and match what they like within the con-straints of what is available and how much they can use.The management of fringe benefits is a very resource-intensive process, especially when thousands of employ-ees are involved. Nike and Signet Bank both installed intelligent agent software that empowers employees tomanage their own fringe benefits selections online. Em-ployees access the human resources databases and conduct activities such as selecting and changing benefitsor making charitable contributions through payroll deductions.

The software agent that supports these activities iscalled Electronic Workforce (from Edify Corp.). It en-ables employers to delegate to employees some time-consuming and repetitive tasks that were previouslyconducted by human resources (HR) employees. Em-ployees enter and delete data, command the computer toperform certain transactions, and interpret information.

If they make mistakes or request benefits for which theyare not eligible, the agent immediately alerts them to theproblem. Previously, paperwork would have to berouted to an employee for corrections and then back tothe HR department. The use of the agent enables com-panies to increase benefits options and employee satis-faction, with the same or even fewer human resourcesemployees. Some new applications involve speech recog-nition capabilities and enhanced customer e-mail.

Questions

1. Can you imagine what would happen if there wereno agents in this case?

2. How can an agent know that an employee made amistake?

3. Enter edify.com and examine the new capabilitiesof eletronic workforce and other agent-basedproducts.

‘s‘s edify.com

FIN

MKT

POM

HRM

ACC

Today, labor costs are estimated to be as much as 60 percent of the total cost of infor-mation delivery.

Mundane personal activities. In our fast-paced society, time-strapped individualsneed new ways to minimize the time spent on routine personal tasks like booking air-line tickets. One specific form of intelligent agents is the voice-activated interface agentthat reduces the burden on the user of having to explicitly command the computer.

Search and retrieval. It is not economically possible to directly manipulate a data-base system in a business setting that involves millions of data objects. Users have todelegate to agents the tasks of searching and cost comparison. These agents performthe tedious, time-consuming, and repetitive tasks of searching databases, retrievingand filtering information, and delivering results to the user.

Electronic commerce agents. Some of the agents described earlier are used in vari-ous EC activities (see Chapter 9).

Domain experts. It is advisable to capture costly expertise, model it, and make itwidely available. “Expert” software agents can be models of real-world agents, suchas translators, lawyers, diplomats, union negotiators, and even clergy.

Management activities. Intelligent agents can even be used to assist managers inperforming their activities. Some management-oriented tasks that an agent can doare: advise, alert, broadcast, browse, critique, distribute, enlist, empower, explain, fil-ter, guide, identify, match, monitor, navigate, negotiate, organize, present, query, re-port, remind, retrieve, schedule, search, secure, solicit, store, suggest, summarize,teach, translate, and watch.

M o b i l e A g e n t sAgents may be either static, residing on the client machine to manage a user interace, forinstance, or mobile. Mobility is the degree to which the agents travel through networks.Mobile agents can move from one Internet site to another and can send data to and re-trieve data from the user, who can focus on other work in the meantime. This can be veryhelpful to users. For example, if users want to continuously monitor an electronic auctionthat takes a few days, they essentially would have to be online continuously for days.

Section 12.4 Intelligent Agents 413

Tab le 12.4 App l i cat ions of In te l l igent Agents (Non- Internet )

App l i c a t i on De s c r i p t i on

User interface agents Monitor usage and suggest improvement. Example:Microsoft’s wizards.

Operating systems agents Add accounts, do group management, manage access,add/remove programs and devices, monitor licenses.

Spreadsheet agents Offer suggestions for improvements. Can tutor noviceusers. Sometimes called wizards.

Workflow and task Administer workflow management—monitor management agents activities, alert, and remind. Example: Ginkgo from IBM.

(See networking.ibm.com/iag/iaghome, and IBM’sintelligent agent custom services.)

Software development agents Assist in routine activities such as data filtering.

Mobile agents that automatically watch auctions and stocks for you are readily available.Another example of a mobile agent is one that travels from site to site, looking for infor-mation on a certain stock as instructed by the user. When the stock price hits a certainlevel, or if there is news about the stock, the agent alerts the user.

A p p l i c a t i o n s o f I n t e l l i g e n t A g e n t sMost intelligent agent applications can be classified into three categories: (1) Internetagents, (2) electronic commerce agents, and (3) other agents. The first two categoriesare illustrated in Chapters 7 and 9. Examples of “other agents” are shown in Table 12.4.

W h a t I s V i r t u a l R e a l i t y ?There is no universal definition of virtual reality (VR). The most common definitionsimply that virtual reality is interactive, uses computer-generated, three-dimensionalgraphics, and is delivered to the user through a head-mounted display. Defined tech-nically, VR is an “environment and/or technology that provides artificially generatedsensory cues sufficient to engender in the user some willing suspension of disbelief.”The user get the feeling of physically being in an environment by interacting with asimulation of it.

The benefits of virtual reality are obvious: More than one person and even a largegroup can share and interact in the same environment. VR thus can be a powerfulmedium for communication, collaborative entertainment, and learning. The user cangrasp and move virtual objects. In VR a person “believes” that what he or she is doingis real, even though it is artificially created. This capability can be utilized for gaining acompetitive business advantage.

Sophisticated VR systems simulate sight, sound, and touch and combine thesesenses with computer-generated input to users’ eyes, ears, and skin. By using a head-mounted display, gloves, and a bodysuit, or by means of large projected images in sim-ulator cabs, users can “enter” and interact with artificially generated environments.For example, Figure 12.4 shows a skier in the NEC Corporation (Japan) Lab. NECused the laboratory to develop a ski simulator, which is available in amusement cen-ters ands and is also used for training.

B u s i n e s s A p p l i c a t i o n s o f V i r t u a l R e a l i t yExtensive use of virtual reality is expected in marketing. For example, Tower Recordsoffers a virtual music store on the Internet; customers can “meet” each other in frontof the store, go inside, and preview CDs and videos. They select and purchase theirchoices electronically and interactively from a sales associate. Similarly, virtual super-markets could spark interest in home grocery shopping. In the future, shoppers will


1. Define intelligent agents.

2. List the major characteristics of intelligent agents.

3. List some typical applications of intelligent agents.


Figure 12.4 Developingvirtual skiing in Japan.

12.5 VIRTUAL REALITY: AN EMERGING TECHNOLOGY

enter a virtual supermarket, walk through the virtual aisles, select virtual products,and put them in a virtual cart; actual groceries will later be delivered to customers’homes. Several real estate companies use VR to present properties even before theyare built. Applications in these and other areas are shown in Table 12.5.

Virtual reality is just beginning to move into many business applications. Three-dimensional worlds on the Internet should prove popular because they providemetaphors to which everyone can relate.

There are many issues related to the implementation of intelligent systems. We’ll dis-cuss three of these issues here: ethical and societal issues, legal issues, and global issues.

E t h i c a l a n d S o c i e t a l I s s u e sIn general, scientists and managers are concerned with the possibility of power misuseand harm to people from the use of intelligent systems. Professor Rheingold, a virtualreality pioneer, raised the issue of behavior in a world where the distinction betweenthe real and the virtual is unclear. For example, when the line between reality and vir-tual reality is blurred, some people might use real weapons with as little thought forthe actual consequences as they use virtual ones. Several of the ethical issues dis-cussed in Chapter 15 are directly related to intelligent systems. For example, privacy is


1. Define virtual reality.

2. List the advantages of virtual reality.

3. Describe some business applications of virtual reality.

Section 12.6 Ethical and Global Issues of Intelligent Systems 415

Tab le 12.5 Examples of V i r tua l Rea l i ty App l i cat ions

I ndu s t r y App l i c a t i on s

Manufacturing • Worker training• Design, testing, and virtual prototyping of products and processes• Engineering and ergonomic analysis• Simulation of assembly, production, and maintenance

Transportation • Virtual aircraft mockups• New-car design and testing of cars in virtual accidents• Simulation of flying first class in airplanes

Finance • View stock prices and characteristics

Architecture • Display of building and other structures

Military • Training (pilots, astronauts, drivers) and battlefield simulation

Medicine • Training of surgeons (with simulators) and planning surgeries• Planning physical therapy

Marketing • Store and product display• Electronic shopping

12.6 ETHICAL AND GLOBAL ISSUES OF INTELLIGENT SYSTEMS

a major concern in knowledge bases (who will have access to information stored inthe knowledge base?), and experts’ knowledge is related to intellectual property.

If you have watched or read any science fiction, you no doubt have found scenar-ios in which robots turn against humans. Isaac Asimov, a famous science fictionwriter, suggested that the following laws of robotics be followed in developing roboticapplications:

1. A robot may not injure a human being or, through inaction, allow a human to beharmed.

2. A robot must obey orders given by humans except when that conflicts with the firstlaw.

3. A robot must protect its own existence unless that conflicts with the first or secondlaws.

HAL and Dave. If you saw the classic science fiction film 2001: ASpace Odyssey, you might remember the astronaut, Dave, saying to therobot (which was named HAL), “Open the pod bay door, HAL.” HALreplies, “I’m sorry, Dave, I can’t do that.” A similar scenario occurredwith a robot named David in the movie Artificial Intelligence (AI). Asthese scenes make clear, computers with intelligent systems might beable to refuse human orders, and they might hurt people. This examplerelates to the issue of how much decision making power to delegate tocomputers and how to guard against malfunctions. ●

The knowledge embedded in intelligent systems is often acquired from human ex-perts. This knowledge is frequently available in terms of rules, some of which are diffi-cult to explain whereas others have not even been tested. Therefore, an ES based onsuch rules may not be accurate. Furthermore, the use of such rules may damage notonly property, but people as well. Like any other knowledge, computerized rules mustbe tested before they are used and then must be constantly updated and maintained.

L e g a l I s s u e sThe use of intelligent systems raises interesting legal issues such as:

• What happens if a manager or expert enters an incorrect judgment value into an ESand the result is damage or a disaster?

• Who is liable for wrong advice (or information) provided by an ES? For example,what happens if a physician accepts an incorrect diagnosis made by a computer andperforms an act, based on this diagnosis, that results in the death of a patient? Is itthe fault of the physician, the knowledge engineer that solicited the expertise, theknowledge contributor, the software manufacturer, the system builder, or somecombination of them?

• Who owns the knowledge in a knowledge base?

• Who is an expert? What if several experts disagree?

• Can management force experts to contribute their expertise?

• Should royalties be paid to experts who provide the knowledge to ES, and if so inwhat amount?

• What is the value of an expert opinion in court when the expertise is encoded in acomputer?

EXAMPLE


2001: A Space Odyssey’sHAL computer with friendDave.

These legal issues are largely not addressed by current laws. There are many pendinglawsuits related to IT, the Internet, and e-commerce, some of which are directly re-lated to intelligent systems.

G l o b a l A s p e c t s o f I n t e l l i g e n t S y s t e m sIntelligent systems are used in many applications related to global trade. Althoughmost intelligent systems, whether used domestically or globally, are similar to anyother information system, some do have unique global aspects. The following exam-ples demonstrate the variety of global applications.

Foreign trade. An expert system that advises companies on how to exploit opportu-nities related to the NAFTA agreement (which promotes trade among the UnitedStates, Mexico, and Canada) is available online (corporateinformation.com). For ex-ample, the system determines whether a finished product qualifies for preferential tar-iff treatment, and it helps companies set up buying policies.

Foreign exchange transactions. An application called the FS System advises on for-eign exchange trading—trading of the currencies of various nations as their values riseand fall in relation to each other. It contains trading, hedging, and risk-control strate-gies (athenagroup.com). Similarly, TARA is an intelligent assistant for traders makingforeign exchange transactions. Foreign exchange traders need to consider historicaltrends, a country’s risks, economic directions, and more. Manufacturers HanoverTrust is using this system to facilitate its traders’ investment decisions. (For more seeafexco.com, and forex-trc.com.)

Employee training. Many companies train their employees online before they go toa foreign country. The intelligent systems cut the training time by as much as 50 per-cent, and employees can be trained anywhere.

Weather forecasting. Climatic expert systems provide long-range climate forecastsfor the North Pacific, North Atlantic, North America, Europe, and the European Arc-tic seas. Such forecasts are critical for global commodity traders. The service is pro-vided for free on the Internet (onlineweather.com).

Section 12.6 Ethical and Global Issues of Intelligent Systems 417

A specialized expert systemfor weather forecasting.

Automatic language translations. As countries’ borders begin to disappear inglobal trading, language translation is becoming very important. This topic is very im-portant in e-commerce, where appropriate translation of Web pages is a critical suc-cess factor. The use of intelligent systems in language translation has been progressingrapidly since the mid-1990s.

Many other systems and applications are used to facilitate international trade.Several examples were provided earlier (e.g., the use of ES to fight money launderingacross international borders, and the use of a hybrid intelligent system for developingglobal marketing strategy). As international trade is expanding, mainly due to the In-ternet and trading blocks like the European Union and NAFTA, expertise will beneeded in many areas, ranging from legal issues to export and import licenses. Suchexpertise can be provided to a global audience online. Also, expert systems can pro-vide to users in developing countries the advice of top experts in the fields of medi-cine, safety, agriculture, and crime fighting.


1. List the laws of robotics.

2. Describe some major legal issues in intelligent systems.

3. Relate intelligent systems to global trade.


WHAT’S IN FOR ME ?WHAT’S IN FOR ME ?

F O R T H E A C C O U N T I N G M A J O RIntelligent systems are used extensively in auditing to uncover irregularities. They arealso used to uncover and prevent fraud. Today’s CPAs use intelligent systems formany of their duties, ranging from risk analysis to cost control. Intelligent agents arealso used for several mundane tasks such as managing accounts in operating systemsor monitoring employees’ Internet usage.

F O R T H E F I N A N C E M A J O RPeople have been using computers for decades to solve financially oriented problems. In-novative applications exist in stock market decisions, bond refinancing, debt risk assess-ment, analysis of financial conditions, business failure prediction, financial forecasting,investment in global markets, and more. Intelligent systems were found to be superior toother computerized methods in many instances. Intelligent agents can facilitate the use ofspreadsheets and other computerized systems used in finance. Finally, intelligent systemscan help in reducing fraud in credit cards, stocks, and other financial services.

F O R T H E M A R K E T I N G M A J O RNew marketing approaches such as targeted marketing and marketing transactiondatabases are heavily dependent on IT in general and on intelligent systems in partic-ular. Intelligent systems are partially useful in mining customer databases and pre-dicting customer behavior. Successful applications are noted in almost any area ofmarketing and sales, from analyzing the success of one-to-one advertisement to sup-porting customer help desks. With the increased importance of customer service, theuse of intelligent agents is becoming critical for the provision of fast response.

ACC

FIN

MKT

�1 Describe artificial intelligence and compare it to conventional computing.The primary objective of AI is to build computers that will perform tasks that canbe characterized as intelligent. The major characteristics of AI are symbolic pro-cessing (in contrast with numerical processing in conventional systems), use ofheuristics (instead of algorithms), and the application of inference techniques.

�2 Identify the characteristics, structure, benefits, and limitations of expert systems.Expert systems technology attempts to transfer knowledge from experts and docu-mented sources to the computer’s knowledge base, in order to make the knowl-edge available to nonexperts for the purpose of solving problems quickly andeffectively. The inference engine, or thinking mechanism, is a program that usesthe knowledge base to solve problems. Expert systems can provide many benefits.The most important are improvement in productivity and/or quality, preservationof scarce expertise, enhancing other systems, coping with incomplete information,and providing training. Limitations of expert systems include the inability to learnfrom mistakes, the development cost, and legal/ethical issues in their applications.

�3 Describe the major characteristics of natural language processing and voice technologies.Natural language processing (NLP) provides an opportunity for a user to commu-nicate with a computer in day-to-day spoken language. Speech recognition enablespeople to communicate with computers by voice. Voice-synthesis technology en-ables computers to reply in a humanlike voice. Voice portals provide access to theInternet by using voice via telephone. There are many applications and benefits ofthese emerging technologies.

�4 Describe neural computing and its capabilities.Neural systems are organized and operated in a way similar to biological neuralnetworks. Artificial neurons receive, process, and deliver information. A group ofconnected neurons forms an artificial neural network that can be used to discoverpatterns in historical data and make predictions accordingly. Neural networks areespecially useful in data and Web mining in e-commerce.

SUMMARY

Summary 419

F O R T H E P R O D U C T I O N / O P E R A T I O N S M A N A G E M E N T M A J O RMany of the early expert systems were developed in the production/operations man-agement field for tasks ranging from diagnosis of machine failures and prescription ofrepairs to complex production scheduling and inventory control. Some companies,such as DuPont and Kodak, have deployed hundreds of expert systems in the plan-ning, organizing, and control of their operational systems.

F O R T H E H U M A N R E S O U R C E S M A N A G E M E N T M A J O RHRM departments use intelligent systems for many applications. For example, intelli-gent agents can find resumes of applicants posted on the Web and sort them to matchneeded skills. Expert systems are used in evaluating candidates (tests, interviews). In-telligent systems are used to facilitate training and to support self-management offringe benefits. Neural computing is used to predict employee performance on the jobas well as to predict labor needs. Voice recognition systems provide benefits informa-tion to employees.

POM

HRM

�5 Define intelligent agents and their role in IT.Intelligent agents are software entities that can sense the environment and carryout a set of operations with some degree of autonomy, using a knowledge base inthe process. They can perform many mundane tasks, saving a considerable amountof time and improving quality.

�6 Describe virtual reality.Virtual reality is a 3-D interactive computing environment that is beginning to sup-port business simulation applications.


I N T E R A C T I V E L E A R N I N G S E S S I O N

DISCUSSION QUESTIONS

1. Explain how an ES can distribute (or redistribute)the available knowledge in an organization.

2. What is the difference between voice recognitionand voice understanding?

3. Compare and contrast neural computing and con-ventional computing.

4. Compare and contrast conventional processing withartificial intelligence processing.

5. Review the various tasks that intelligent agents canperform. Do these tasks have anything in common?

6. Deep Blue of IBM defeated the world chess cham-pion, Gary Kasparov, in 1997. If computers cannotthink, how is such a defeat possible? Find some re-cent information about Deep Blue.

7. Why are neural computing and case-based reason-ing viewed as machine learning?

8. How is e-learning related to AI?

9. How is virtual reality related to AI?

PROBLEM-SOLVING ACTIVITIES

1. Lance Eliot made the following comment in AI Ex-pert (August 1994, p. 9): “When you log-on to thenetwork, a slew of agents might start watching. Ifyou download a file about plant life, a seed companyagent might submit your name for a company mail-ing. Besides sending junk mail, such spying agentscould pick up your habits and preferences and per-haps make assumptions about your private life. Itcould note what days you get onto the system, howlong you stay on, and what part of the country youlive in. Is this an invasion of your privacy? Shouldlegislation prevent such usage of intelligent agents?Perhaps a network police (more intelligent agents)could enforce proper network usage.”

a. Prepare arguments to support your perspectiveon this issue.

b. Prepare counterarguments on the same issue.

2. Access the U.S. Department of Labor Web site onsafety (osha.gov). Go to “OSHA etools.”

a. Examine the expert systems available.

b. Write a report on the capabilities of two of thesystems.

3. Airline gate assignment, the responsibility of gatecontrollers and their assistants, is a complex and de-manding task. At O’Hare Airport in Chicago, forexample, two gate controllers typically plan berthingfor about 400 flights a day at some 50 gates. Flightsarrive in clusters for the convenience of customerswho must transfer to connecting flights, so the con-trollers must sometimes accommodate a cluster of30 or 40 planes in 20 or 30 minutes. To complicatethe matter, each flight is scheduled to remain at itsgate a different length of time, depending on the

Real-World Case 421

INTERNET ACTIVITIES

1. Prepare a report on the use of ES in a help desk.Review products and customer cases at4helpdesk.com, exsys.com, and ebehelpdesk.com.

2. There is considerable interest in intelligent agents atMIT (ai.mit.edu). Find the latest activities on IA fromthat site. What projects deal with crime prevention?

3. Enter the site of Tower Records (towerrecords.com)and examine the use of virtual reality in 3-D presen-tations. Also, visit mindflyx.com/au, sgi.com, andliquidimage.com/ca and examine the VR products.Finally, join a VR newsgroup and try to locate newbusiness applications.

4. Visit sas.com, egz.com and hnc.com. Identify linksto real-world applications of neural computing in fi-

nance, manufacturing, health care, crime fighting,and transportation. Prepare a report on current applications.

5. Enter google.com and cio.com, and identify the lat-est managerial trends and issues related to appliedAI technologies.

6. Enter botspot.com/newsletter, pcai.com/pcai, andagentland.com, and identify electronic commerce in-telligent agents for helping shoppers. (Use the mostrecent 3 months.)

7. Enter worldpoint.com and check the capabilities oflanguage translation. Find a competing vendor andcompare.

TEAM ACTIVITIES AND ROLE PLAYING

1. Assign each team a major functional area (account-ing, finance, etc).

a. Using a literature search, material from vendors,or industry contacts, each team finds recent appli-cations (within the last year) of intelligent systems.

b. The team will submit a report about the applica-tions found in each functional area.

c. The class will conduct an analysis of the similari-ties and differences among the applicationsacross the functional areas. A possible arrange-ment is to look at the underlying technology.

(For example, compare the use of ANN in mar-keting versus finance or management.)

2. Each team composes a list of mundane tasks he orshe would like an intelligent software agent to exe-cute. Are intelligent agents available today to do thetasks on your lists? Consult botspot.com and agent-land.com.

3. Enter PCAI.com/pcai and find the AI informationcategories. Assign each team to one or two cate-gories. Search for recent (last 6 months) news itemsand interesting applications.

schedules of connecting flights and the amount ofservicing needed. Mix those problems with the needto juggle gates constantly because of flight delayscaused by weather and other factors, and you getsome idea of the challenges.

a. Based on what you learned in this chapter, what isthe most appropriate intelligent system that can beused to support the work of the controllers?

b. Why? How could the support be given?

R u l e s o f T h u m b S c h e d u l e T r a i n s i n P a r i s

REAL-WORLD CASE

The Business Problem One of France’s busiest trainstations is the Gare de l’Est in Paris. Trains are parked at30 platform tracks and then funneled onto six mainlinetracks. More than 1,100 trains come and go every day,including some that cruise at close to 200 miles per hour!That’s a train every 30 seconds during busy periods.

Scheduling local and long-distance trains at the 30platforms is a complex logistics problem. Traffic levelsare near the theoretical maximum. Each train must beassigned one of 640 possible routes into and out of thestation. Local and long-distance trains share the same

platform assignments. One single delayed train cancause a chain reaction that reverberates through theschedule for as long as four hours afterward. When atrack or platform must be taken out of service for re-pairs, as many as 250 trains each day may have to be di-verted. Only specialists at the Gare de l’Est have theskills to reroute these trains without creating major de-lays—skills they have derived from 10 to 15 years of ex-perience working at the station.

The number of possible solutions to such problemsis astronomical. As in chess, the moves and counter-

moves are so numerous that there is no satisfactory al-gorithmic solution to this problem. To enumerate allpossible solutions using even a powerful computer maytake days to execute. Yet, in the real world of railroad-ing, a solution is needed in minutes.

Dispatchers who have solved daily problems man-ually for scores of years handle the scheduling. Untilnow, the solution for these human experts has been touse rules of thumb—heuristics. These rules are enunci-ated as constraints that say what may and may not bedone in terms of train routes; they also consider the ef-fects of any change on all the others routes. A basicrule in railroading, for example, is that two trains maynot occupy the same track at the same time. The firstcorollary is that, on a single track, no train may pass an-other from either direction. These are rules that mustnever be violated. Other rules can be relaxed to solvepressing problems, such as “Don’t assign a train to aplatform until the previous debarkees have fullycleared the platform.”

The deficiencies of a manual system are the following:

1. When a dispatcher is out sick, an extreme amountof pressure is placed on the remaining dispatchers.

2. Due to time constraints, dispatchers can run throughonly a limited number of possible arrangements forboth planning and rerouting, so the best ones maybe missed.

3. On holidays, when extra trains are needed, it is nec-essary to relax some rules. Working under timepressure, dispatchers do not always relax the mostappropriate rules.

4. Employees fill out paper documents manually, atime-consuming process.

5. Less experienced dispatchers make mistakes, whichmay cause significant delays.

6. Daily planning and preparation of the semiannualtimetable takes a great deal of time.

7. The scheduling issue creates a limit on the flow oftraffic that is much lower than the physical limit.

8. Unnecessary delays may develop when the dispatchers cannot work fast enough to handleemergencies.


The IT Solution Adding more dispatchers is an ex-pensive solution that does not resolve all of the abovedeficiencies. In late 1988, it became clear to manage-ment that computerized support was needed. One pro-posal suggested the use of an expert system.

Such a system was built in 1989 and has run success-fully since. Today, it is interactive, runs on an intranet,and works with a combination of rules and object-oriented programming. When a problem develops, theES divides it into subproblems. Possible routes arequickly examined. If a potential conflict between trainsis indicated, appropriate recommendations for its reso-lution are automatically provided by the ES. The pro-gram applies its rule base automatically, listing anysituation it cannot solve. Then the dispatcher, in an in-teractive mode, attempts to solve the problem with theaid of the computer.

The Results The system is especially useful for theless experienced dispatchers, but even experienced dis-patchers use it to save time. Overall, the ES increasedthe productivity of the dispatchers by up to 100 per-cent, reduced errors, and solved most of the deficien-cies listed above.

Source: Based on information provided by Texas Instruments, thebuilder of the system.

Questions

1. A preliminary study concluded that DSS or MISwould not be correct solution approaches. Why?

2. The possibility of using neural computing was exam-ined but quickly discarded. Why?

3. Which of the deficiencies listed earlier cannot be re-moved by an expert system? Why?

4. Can this system be transferred to train stations inother countries? Why or why not?

5. Explain how the improvements are achieved. Whatis the role of dispatchers now? Are dispatchersneeded at all?

6. The ES output was designed so that the electronicforms and information flow would look exactly thesame as those of the manual system. What is thelogic of such a design?

Virtual Company Assignment 423

E x t r e m e D e s c e n t S n o w b o a r d s

As you walk down the hallway to JacobMarch’s office, you wonder what your nextassignment will entail. You knock on thedoor, and Jacob greets you with a smile andoffers you a seat across from his desk.

After a few minutes of pleasantries andsmall talk, Jacob gets to the point of yourmeeting. “When customers purchase snow-boards from our Web site, they enter infor-mation about themselves using an onlineform. Using this information, we custom-build a board to exact specifications. Somecustomers know exactly what they wantwhen they choose a board; however, manymay not be sure of what they want or need.They end up buying a board that doesn’tmatch their exact needs or riding require-ments. To help these customers, we wouldlike to create an expert system that wouldguide these customers through the selectionprocess by asking them a series of questionsand then recommend a snowboard bestsuited for them. Eventually, this expert sys-tem could be embedded in our Internetsite.”

Jacob explains, “For your next assign-ment, I need you to prepare a report on howan expert system could support the cus-tomer when choosing a snowboard.” Jacobhands you a sheet of paper that outlinesyour assignment, and you head back to yourdesk. Looking at the paper Jacob gave you,you read the following assignment.

• Can embedding an expert system in ourInternet site provide a competitive ad-vantage over our competitors? Can thisadvantage be sustainable?

• Should a customer have an option ofchoosing his or her own snowboard orusing the expert system to choose asnowboard? What are the advantagesand disadvantages of having the expertsystem choose a snowboard for everycustomer?

• Can you provide an example where anexpert system could support another areaor department within Extreme DescentSnowboards? How could the cost of de-veloping this expert system be justified tomanagement?

VIRTUAL COMPANY ASSIGNMENT