Robotics Reference
In-Depth Information
These programs behave like human experts in various useful ways: they
operate efficiently and effectively in a narrow problem area such as spe-
cialist medical diagnosis; they solve problems that are difficult enough to
need a significant level of human expertise for their solution; they can
offer intelligent advice; and they can explain the reasoning behind their
decisions in such a way that a user of the system can understand their
“thinking”.
The domains in which expert systems have been used cover a wide
field, including not only medicine but also mathematics, engineering,
geology, crime, computer science, business, law, politics and defence.
Examples of the type of decisions that can be made by expert systems
are “From which of three possible thyroid complaints is this patient suf-
fering?”; “Why is my car not starting?”; “How likely is it that the stock
market will crash in the next month?”; “Is there anywhere on this parcel
of land where there might be oil deposits?”; and “Is the judge more likely
to find me guilty or not guilty?”
Expert systems usually represent their expertise and knowledge as
rules within the program, often supplemented by data of various types.
In a weather forecasting system, for example, the rules might include “If
the pressure does not rise then it is likely to rain today”, while the data
in the program might include the rainfall and air pressure statistics for
the relevant area over the previous 20 years. In general, each rule in an
expert system represents a chunk of expert knowledge, and most systems
contain hundreds of rules. For example, the MYCIN medical diagnosis
system employed about 450 rules while the PROSPECTOR system for
locating mineral deposits had more than 1,600. These rules are usually
obtained by a process called knowledge engineering, based on interview-
ing human experts for periods of weeks or longer. The person who carries
out these interviews and converts the resulting information into rules is
called a knowledge engineer.
The philosophy of expert systems design is rather different to that
of programming most other types of task. Computer programs that em-
ploy conventional decision-making logic are usually driven by algorithms
(methods) designed to solve a specific problem. Such algorithms are em-
bedded as part of the program and whenever the algorithm needs to be
modified, so will the program. But knowledge-based systems work in a
different way. Their methodology remains constant but what can and
usually does change with time is the set of rules incorporating the sys-
tem's knowledge. Rules are stored as data that can be added, removed
