Expert Networks vs. Expert Systems
by Harry Stevens

Ever since I worked in the 1950's on computer-aided language
translation, I have been skeptical about the potential of
so-called Artificial Intelligence (AI). Much more promising
is another type of AI, Augmented Intelligence. I like to
think of expert networks as being a form of Augmented
Intelligence, which has more immediate potential than the
expert systems that are currently getting much attention within
the field of Artificial Intelligence.

Both types of AI certainly have potential. However, I believe
the exaggerated view that intelligence can ever become
artificial tends to diminish more promising efforts to augment
intelligence.

Expert networks represent a new dimension, an electronic
societal and organizational dimension, that did not really
exist prior to the development of Computer Aided Communication.
While expert networks can be used by traditional organizations
to strengthen their efforts to produce and provide products
and services, expert networks also seem to represent almost a
new form of organization.

An older form of organization, namely associations, may be
transformed by expert networks. Prior to the development of
expert networks, associations have not been as important forms
of organizations as, for example, private corporations and
public agencies of government. Associations -- whether formed
for professional, religious or other reasons -- may be
transformed by expert networks into much stronger
organizations. The economic opportunity as well as the
electronic challenge for associations may be greater than some
of them will be able to handle. If so, then some new
networking form of organization not identified directly with
what we now know as associations may emerge.

An example of expert networks was developed in the late 1970's
when science advisors to state legislatures joined together
with technical professional societies, federal labs, and
public interest research groups to form the Legitech Network.
In 1979, approximately 1000 inquiries were initiated online as
topics, to which another few thousand notes were then added as
responses to those inquiries. These inquiries were on a wide
range of legislative issues that had scientific or technical
content -- in such areas as energy, health, environment,
economics and communications.

For example, one frost-belt state posed the question: "What
are alternatives to road salt for dealing with icy highways
without polluting water supplies?" Another state, having
recently dealt with that problem, responded, as did
associations and labs that knew of relevant research on that
topic. Other frost-belt states joined the topic to get the
benefit of inquiry responses that might help their states as
well as the inquiring state.

Sun-belt states did not join and therefore were not subjected
to information overload about a topic of no importance in
their climate. An important feature of expert networks is
this screening to avoid information overload.

There were generally four types of responses for these kinds
of inquiries:

(1) BACKGROUND responses usually provided
immediately by inquirers, in order to
place their pointed inquiries in
broader contexts of (a) what they
already knew and (b) the situation
that led to the inquiry being made

(2) SUBSTANTIVE responses from online
expert peers or professionals

(3) LEADS to off-line experts, giving their
qualifications and how to contact them

(4) REFERENCES to published information or
even copies of references found online
by searching public domain databases

The order of the above types of inquiry responses is
consistent with what I have been told is the God Given Rule of
Research, "First, ask someone who knows. If that fails, ask
someone who knows someone who knows. If that also fails, then
look it up."

Actually, the kinds of inquiries that are asked online are
sometimes about issues for which there is little or no
published information. Yet there may still be much knowledge
in the minds of experts who can be reached through an expert
network.

I recall that, when the Three-Mile Island Nuclear Power
Incident occurred, numerous inquiries arose within the
Legitech Network. These ranged from concern about the effects
of nuclear radiation to how to de-commission nuclear power
plants. Inquiries such as those, for which there may not have
been quick answers were nevertheless good indications of what
different states were then considering.

Inquiries within expert networks usually have half-lives of
about two weeks when they are receiving responses.
Occasionally, an inquiry might be fuzzy or interesting enough
to become a mini-conference that could be written to for two
months or longer. But the half-lives for inquiries and their
responses needing to be referenced or read online by
latecomers may be two years or longer, depending upon how
obsolete that knowledge may become due to later research.

The searching, indexing, archiving, and editing tools
incorporated in the software supporting expert networks allow
the communications flow of inquiry networking to yield
valuable knowledge that remains online. This knowledge is by
no means as well organized as the rule-based knowledgebases of
expert systems. Yet the by-product knowledgebases of expert
networks may serve as bridges to the development of the
rule-based knowledgebases of expert systems.

Furthermore, expert networks may be even more promising than
those systems that focus upon capturing the expertise of single
experts. Expert networks incorporate within them ongoing peer
review, a process fundamental to the advancement of knowledge.