Category : Files from Magazines
Archive   : AINOV86.ZIP
Filename : BRAIN.NOV

Brain Waves
"Viewpoint"
November 1986 AI EXPERT

by Alex Jacobson, President,
Inference Corporation



Expert systems technology enables computer to use human
expertise, judgments and knowledge to solve business problems in
an emulation of the way human experts do. There is considerable
evidence to suggest that this technology, when applied to a well-
focused, sufficiently well-defined domain of interest (e.g.,
authorization of a specific type of credit card, diagnosing of
faults in a specific piece of equipment, scheduling of a specific
fleets of vehicles in a specific geographic area or configuring
of a specific set of machines on a specific factory floor) can
provide human workers who operate in the targeted domain with
computer support at levels of performance equal to or better than
the best human experts in the domain.

The benefit of this capability is to enable computers to
formulate decisions, to draw conclusions and to propose actions
in response to the wide variety of unstructured or poorly
structured problems with which only humans could contend
heretofore. As a result, this technology makes it possible for
computers to do the same sorts of tasks that professionals and
white collar workers presently do in the work force --a necessary
accomplishment if these workers are to receive automation. The
significance of these capabilities is more far reaching than the
technical content, per se, implies. The reason is that expert
systems technology has matured at a time when the computer
industry as a whole is moving through a major transition. The
computer industry has, over the past 30 years, fulfilled much of
its promise in automating clerical level functions (typing
drafting, bookkeeping, inventory management, listings, records
keeping etc.). Business and industry is now focusing attention on
strategic uses of computers in mission-critical applications.
These applications, a prime example of which is the American
Airlines Saber System, can provide a major competitive edge to
companies able to conceive and to implement them. White collar
workers implement business strategies, hence it is this segment
of the work force that will be targeted for computer automation
as strategic uses of computers are undertaken in business. Expert
systems technology is a critical component for delivering this
automation to the professional, technical and administrative
workers who implement mission-critical applications in business
and industry. This propitious timing between a new capability
(i.e., expert systems) and a new requirement (i.e., mission-
critical applications of computers) explains the unusual sense of
importance that is attributed to expert systems technology
throughout the world.

Expert systems technology is primarily targeted for use in
applications software and in software tools that support the
development and operation of applications and systems software.
The fundamental difference between an expert system and a
traditional application program is that such an expert system is
rich in knowledge about the solution of problems in the
application domain in which the expert system operates; whereas
traditional applications are rich in the procedural knowledge
that instruct the computer how to process data to solve the
problems in the domain in question. It is this richness in
knowledge that makes expert systems an enabling technology for
the use of computers in mission-critical applications.
Nevertheless, expert systems contain procedural knowledge with
which to instruct the computer and traditional applications
contain knowledge about the problem solving. It is the higher
density and the greater extent of knowledge about problem solving
that distinguishes expert systems from traditional applications
programs, and provides them with their unusual functional power.

This fundamental difference leads to all of the basic differences
in the underlying tools, technology and programming methodologies
(i.e., knowledge engineering) that set the practice of expert
systems apart from that of conventional software engineering. In
order to elicit deep and extensive knowledge about problem
solving in any but most straightforward industrial task areas, it
is necessary for the software engineer to develop the expert
system by means of an iterative or evolutionary development
process. The reason is that humans cannot divulge the deep and
subtle levels of knowledge about their problem solving expertise
that industrial class expert systems require, and are able to use
effectively in a straightforward debriefing process. Rather, it
is necessary that the software engineering methodology be capable
of supporting a development regimen that permits knowledge
obtained by debriefing to be built into an operating partial
application so that areas of mission knowledge (i.e., knowledge
not accessible by straightforward interview) can be identified
and then added to the partial application to create a more
complete, yet, perhaps still partial application, which can then
be used to find still less accessible areas of germane knowledge
which in turn can be added to the system, and so on. This method
of evolving the expert system into existence is called "bottom-
up-discovery", and is the distinguishing feature of knowledge
engineering.

Expert systems tools contain the AI technology required to
support the process of knowledge engineering for building expert
systems. They contain the structures required to store a variety
of different types of knowledge paradigms, an inferencing engine
that permits this knowledge to be used as the system evolves even
though the knowledge is added to the system incrementally,
systems software that allows the knowledge engineering to browse,
modify, add, delete, understand or otherwise manipulate the
knowledge in the evolving knowledge base, and tools to assist the
knowledge engineer to build the expert system including the user
interface of the resulting expert systems. These tools serve the
purpose of accelerating the pace with which this new technology
can be effectively applied.

Expert systems technology is basically a software technology.
While it has almost exclusively been developed in Lisp, and, in
recent years, Lisp machines, like all other software technologies
it is intrinsically portable to other languages and to other
classes of computers. This is of vital importance. To realize
their full potential, expert systems must fulfill their role in
mission-critical applications. This requires that expert systems
operate effectively and efficiently in conjunction with existing
computer environments. Hardly any of these existing environments
support Lisp or incorporate Lisp machines. Since expert systems
technology is portable, it is clear that it must be ported to
mainstream computers and connected to mainstream software at the
levels of traditional languages, systems software and
applications programs. This requirement cannot be evaded -- nor
need it be.

Finally, there is the question of culture. Expert systems are
computer applications that arise from a technology culture that
is substantively different from the culture that has created
traditional computer applications. Cross culture communication is
always difficult. It will be no different in this instance. It
promises to be one of the more formidable obstacles to
commercialization of expert systems. Not only does the
applications programming community face the challenge of
assimilating this new technology, but business operations
management as well as end-users also must become both familiar
and comfortable with expert systems and their implications as
these systems move into the front office. Management faces the
challenge of managing business practices in which the underlying
logic of the practice has been made explicit for the first time
and for which accountability of performance is documented with
the scrupulousness of which only computers are capable. End-users
who have never before used computers must become comfortable with
these new mechanical assistants -- no simple task given the
anxiety often incurred by computers in people who have no
predilection for machines.

Although these obstacles are formidable, they can and will be
transcended. The benefits of industrial scale expert systems to
the businesses that employ them promise to be too great for these
transitional burdens to be anything but passing challenges.
DP/MIS workers, end-user computing programmers, applications
software vendors, all will benefit from their efforts to adopt
this new technology. Therefore, expert systems software will
inevitably lose its singular name and become "just another"
commercial software technology as the computer industry continues
to support the growth of business throughout the world.

commercial software technology as the computer industry continues
to support the growth of business