Knowledge Management and Specialized Information Systems - Principles of Information Systems - page 455

Neural Networks
An increasingly important aspect of AI involves neural networks, also called neural nets. A
neural network is a computer system that can act like or simulate the functioning of a human
brain. 51 The systems use massively parallel processors in an architecture that is based on the
human brain's own mesh-like structure. In addition, neural network software simulates a
neural network using standard computers. Neural networks can process many pieces of data
at the same time and learn to recognize patterns. A chemical company, for example, can use
neural network software to analyze a large amount of data to control chemical reactors. Neural
network analysis has also helped some medical clinics diagnose cardiovascular disease. 52 Some
oil and gas exploration companies use a program called the Rate of Penetration based on
neural networks to monitor and control drilling operations. 53 The neural network program
helps engineers slow or speed drilling operations to help increase drilling accuracy and reduce
costs. Some of the specific abilities of neural networks include the following:
• Retrieving information even if some of the neural nodes fail
• Quickly modifying stored data as a result of new information
• Discovering relationships and trends in large databases
• Solving complex problems for which all the information is not present
A particular skill of neural nets is analyzing detailed trends. 54 Large amusement parks and
banks use neural networks to determine staffing needs based on customer traffic—a task
that requires precise analysis, down to the half-hour. Increasingly, businesses are firing up
neural nets to help them navigate ever-thicker forests of data and make sense of a myriad
of customer traits and buying habits. Computer Associates has developed Neugents
( www.neugents.com ), neural intelligence agents that “learn” patterns and behaviors and pre-
dict what will happen next. For example, Neugents can track the habits of insurance
customers and predict which ones will not renew an automobile policy, for example. They
can then suggest to an insurance agent what changes to make in the policy to persuade the
consumer to renew it. The technology also can track individual users at e-commerce sites
and their online preferences so that they don't have to enter the same information each time
they log on—their purchasing history and other data is recalled each time they access a
Web site.
AI Trilogy, available from the Ward Systems Group ( www.wardsystems.com ), is a neural
network software program that can run on a standard PC. The software can make predictions
with NeuroShell Predictor and classify information with NeuroShell Classifier. See Figure
11.7. The software package also contains GeneHunter, which uses a special type of algorithm
called a genetic algorithm to get the best result from the neural network system. (Genetic
algorithms are discussed later in this chapter.) Some pattern-recognition software uses neural
networks to analyze hundreds of millions of bank, brokerage, and insurance accounts in-
volving a trillion dollars to uncover money laundering and other suspicious money transfers.

neural network
A computer system that can
simulate the functioning of a human
brain.

Other Artificial Intelligence Applications
A few other artificial intelligence applications exist in addition to those just discussed. A
genetic algorithm , also called a genetic program, is an approach to solving large, complex
problems in which many repeated operations or models change and evolve until the best one
emerges. The approach is based on the theory of evolution that requires (1) variation and (2)
natural selection. The first step is to change or vary competing solutions to the problem. This
can be done by changing the parts of a program or by combining different program segments
into a new program, mimicking the evolution of species, in which the genetic makeup of a
plant or animal mutates or changes over time. The second step is to select only the best models
or algorithms, which continue to evolve. Programs or program segments that are not as good
as others are discarded, similar to natural selection or “survival of the fittest,” in which only
the best species survive and continue to evolve. This process of variation and natural selection
continues until the genetic algorithm yields the best possible solution to the original problem.
For example, some investment firms use genetic algorithms to help select the best stocks or
bonds. Genetic algorithms can help companies control inventory levels and get the best usage
of warehouse space. 55 Genetic algorithms are also being used to monitor patient health. 56

genetic algorithm
An approach to solving large,
complex problems in which a
number of related operations or
models change and evolve until the
best one emerges.