Robotics Reference
In-Depth Information
simplest pathways—those involving the fewest substances and steps—
before going on to examine more complex ones. The known constraints
help the program to eliminate certain pathways and to help in identify-
ing any products that are created at intermediate stages in the reaction.
The program's output is a set of the simplest pathways consistent with
the background knowledge, that conform to the constraints and explain
the experimental evidence.
Because the pathways are generated from scratch rather than being
taken from a database of common reactions, MECHEM will quite
often suggest hypotheses that are original. In addition, the program's
systematic and exhaustive approach to the task is often responsible for
unearthing hypotheses that have been overlooked by human scientists,
who simply lack the time and resources to conduct such an exhaustive
search of the problem space. Not only are the pathways proposed by
MECHEM sometimes original, they are often of scientific interest be-
cause of their simplicity. By starting the search on the basis of “simplest
first”, the program tends to reveal pathways that minimize the number
of steps in the chemical reaction and the number of chemical substances
that are assumed to be created at intermediate stages in the reaction.
A more interactive approach is used in the ARROWSMITH pro-
gram, developed by Don Swanson at the University of Chicago, which
helps the user to interrogate a database of medical publications called
MEDLINE, in order to spot new relationships between causes and ef-
fects in the world of medicine and to form and assess novel scientific
hypotheses. The underlying philosophy behind the program is that in-
formation developed in one area of research can be of value in another
area without anyone being aware of the fact. The user might suspect that
some connection exists between, say, eating a lot of bananas and con-
tracting gout.
37
So the user begins with a direct question concerning the
connection between bananas and gout, but a conventional search of the
database will provide no answer if there are no publications containing
both words “bananas” and “gout”, and some sort of link between them.
(Thus far the search process described here has been rather like using
Google.) But then the user can ask ARROWSMITH: “Can bananas, or
a deficit of bananas, influence the cause or risk of gout?” Now it is pos-
sible that bananas might influence some hitherto unconsidered factor,
37
This is a purely hypothetical example born of the author's total ignorance of both gout and
compulsive banana eating.
