Robotics Reference
In-Depth Information
asking whatever questions they do ask. Many people do not always ac-
cept the answers of a human expert without some form of justification;
for example, a doctor providing a diagnosis is expected to explain the rea-
soning behind his or her conclusions, partly so that the patient is aware
of any risks or possible alternative treatments.
Expert systems typically explain themselves by answering two types of
question: “How did you conclude that fact?” and “Why are you asking
me this question?” To answer the “how” question a backward chaining
expert system would keep note of all the rules that were proven true
en route
to reaching its conclusions, and then translate each rule into
understandable language for the benefit of the user. To answer the “why”
question, the system examines the rule it is currently trying to prove.
Suppose, for example, that the system has asked if it is raining. It might
answer the “why” question by saying: “I am asking you if it is raining
because I want to prove that you are going to get wet if you go outside,
and rule 23 says that you will get wet if you go outside and if it is raining
and if you do not have an umbrella handy.”
Confidence Estimates
Many of the rules in an expert system will not have definite conclusions,
rather they will carry an estimate of certainty that the conclusion of the
rule (the “
then
” part) will hold if the conditions (the “
if
” parts) are true.
This is analogous to human reasoning—when we reason we do not al-
ways arrive at conclusions with 100 percent confidence. In the earlier
example of a prediction that it “is likely to rain today” if the pressure does
not rise, the rule might state that “it will rain”, supported by a certainty
level of, say, 85 percent. Statistical techniques are used to determine these
certainties, based on data collected for the relevant problem domain (in
this case the previous 20 years' weather statistics).
MYCIN—A Typical Expert System
MYCIN was developed at Stanford University in the early-mid 1970s,
and was the first large expert system to perform at the level of a human
expert and to provide users with an explanation of its reasoning. The
system was intended to be used by a doctor, to provide advice when
treating a patient—advice about infections that involve bacteria in the
blood and advice about meningitis.
49
These infectious diseases can be
49
Infections that involve inflammation of the membranes that envelop the brain and spinal cord.
