COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Bacteriology as an example AI knowledge representation problem
C00012 ENDMK
C⊗;
Bacteriology as an example AI knowledge representation problem

Current AI expert systems and the AI languages in which they
are written provide for certain kinds of knowledge and certain
ways of using it.  A somewhat elderly example is MYCIN which is
concerned with the diagnosis and treatment of bacterial infections
of the blood.  MYCIN asks its user questions about the patient
and about his symptoms and about the results of the diagnostic
tests that have been performed.  On the basis of this information,
MYCIN offers a diagnosis and recommendations for treatment.

MYCIN's information is represented as a collection of rules.  While
these rules contain much expert knowledge and MYCIN does as
well as experts in bacterial diseases at inferring a diagnosis
from a given set of facts, MYCIN doesn't have the basic knowledge
of bacteriology that is taught in junior high school.  Namely,
MYCIN doesn't know any of the following:

1. Bacteria are material objects that change in time.  In fact
MYCIN knows nothing of time and events that occur in time.
An immediate consequence of this is that MYCIN doesn't make
a prognosis; it can't answer questions about the expected
outcome of the treatment it recommends or the consequences
of not treating the patient.  It also doesn't do planning.
For example, it can't reason that a certain anti-biotic
is contra-indicated when a patient has a fever, and therefore
it is necessary to reduce the fever in order to use the
anti-biotic.  Of course, the experts who provided the information
embodied in MYCIN's rules knew a lot about bacteria, but
the knowledge engineers who extracted the rules from
the experts had to persuade them to ``compile'' their
fundamental knowledge into MYCIN rules.

It is not immediately clear how much a doctor treating
a particular patient uses fundamental knowledge of
bacteriology.  We will say more about this later.

It is clear, however, that fundamental bacteriological
knowledge is possible only for systems that have the
even more basic ability to consider processes occurring
in time.  Notice that understanding is required not
merely of the present time and the immediate future
as determined by the actions that are taken.  One
also needs to be able to reason abstractly about
processes such as the growth and reproduction of
bacteria.

2. Junior high school students are taught that bacteria
is a plural word, i.e. it is a question of many
individual bacteria.  They are taught that bacteria
grow in suitable media and reproduce.  They learn
that bacteria do not arise spontaneously but are
descendants of earlier bacteria.  They learn that
bacteria cannot penetrate many containers and cannot
even readily penetrate skin.  Therefore, if bacteria
are found somewhere it makes sense to ask how they
got there.

3. Bacteria are killed by sufficient heat, but the
required heat is usually too much for a person.

4. Bacteria don't grow, or grow very slowly, when
it is cold.

5. There are many kinds of bacteria.
MYCIN knows about many kinds of bacteria,
but it may not know that there are bacteria it doesn't know about.

6. Some kinds of bacteria can grow in some kinds of food
if the food is at the right temperature.  If the bacteria
grow enough the food is spoiled.  It may taste bad,
and even without tasting bad, it may make people sick
who eat it.

7. Some kinds of bacteria can grow in some parts of
human bodies.  Whether the bacteria can grow depends
on whether the person is immune to that kind of
bacteria.  Some bacteria produce substances that
are harmful to humans and make them sick.

Probably this is not a complete summary of junior
high school bacteriology.  Although it has been
taught at least since the turn of the century,
we should remember that these facts represent
major discoveries of 19th century science and
were matters of controversy.

Summarizing the facts as scientific discoveries,
we have

1. The existence of bacteria as reproducing
organisms sometimes visible under the microscope
and sometimes producing visible colonies in
suitable culture media.

2. Bacteria do not spontaneously arise but descend
from previous bacteria.

3. Specific diseases are caused by specific bacteria,

e.g. tuberculosis is caused by the tuberculosis bacteria
in the lungs.

4. The bacteria causing a specific illness can often
be isolated from the blood, urine or sputum of the
patient, then cultured outside the body and identified.

We would like to consider how to represent this knowledge
in an AI system and consider how it might be used.

For the present we are interested in epistemology rather
than heuristics.  Thus we are satisfied if the program
can follow the rationale of an application of the knowledge,
accepting reasonable applications and rejecting incorrectly
reasoned applications.  This is easier than making
programs that can solve problems and can serve as
a necessary condition for testing whether the
knowledge is represented in the computer.

Here are some examples of the application of junior
high school bacteriology.  These applications may
require the presence of even more basic common sense
knowledge.

1. Rationalization of canning.  Food is protected
from spoilage by bacteria by keeping it in a
can or jar.  It is heated to kill the bacteria
and sealed while hot.  The program should accept
the rationale of canning in jars and cans, but should reject th canning
in baskets.
For this we probably need the additional facts
that bacteria sometimes travel through the air
and that air penetrates baskets.