320
The formal ontologies and languages developed in the first two parts of the
book are based on deductive reasoning and deterministic algorithms. There
is no room for uncertainty. Reality, unfortunately, is quite different, and any
endeavor that attempts to model reality must deal with this fact. This part
of the book compares and contrasts deductive and inductive reasoning, and
then proposes how they can be reconciled.
The first chapter compares deductive reasoning with inductive reasoning,
taking a high level point of view. There are several approaches to reasoning
about uncertainty, and these are surveyed. The most successful approach to
uncertainty is known as Bayesian analysis, and the rest of the book takes this
point of view. Bayesian networks are a popular and effective mechanism
for expressing complex joint probability distributions and for performing
probabilistic inference. The second chapter covers Bayesian networks and
stochastic inference.
Combining information from different sources is an important activity in
many professions, and it is especially important in the life sciences. One can
give a precise mathematical formulation of the process whereby probabilistic
information is combined. This process is known as “meta-analysis,” and it is
a large subject in its own right. The third chapter gives a brief introduction
to this subject.
The book ends by proposing a means by which inductive reasoning can
be supported by the World Wide Web. Because Bayesian networks express
reasoning with uncertainty, we refer to the inductive layer of the web as the
Bayesian Web.Although this proposal is speculative, it is realistic. It has the
advantage of allowing uncertainty to be formally represented in a web-based
form. It offers the prospect of assisting scientists in some important tasks
such as propagating uncertainty through a chain of reasoning, performing
stochastic inference based on observations, and combining information from
different sources.
