An IF by any other name
by Markus Roberts

Columbus discovered America in 1492, which led to its exploration by the
British, Dutch, Portuguese, and Spanish. The fact that their lands had been
discovered (and subsequently needed to be explored) came as quite a surprise to
the Apache, Dakota, Hopi, Navaho, and Siouan....
                                             --Stephen Fife

We are becoming a community of dilettantes. Dabblers with more gee whiz then
judgment, more concerned with trends in "paradigm-based methodologies" than
with getting work done and using our increasingly uncommon sense. We're
becoming so gullible we believe our own press releases (or plan to, by 3rd
quarter '89). We'll believe anything. We'll buy anything.
    Understanding this fact is the key to understanding expert systems.
    Their genesis was well intentioned and conceptually sound. As a branch of
artificial intelligence research and cognitive psychology the study of experts
and how they reason makes a valuable contribution to science. But artificial
intelligence research got simplified into AI and expert reasoning models got
twisted into expert reasoning systems and these got reduced to expert systems
and finally (ugh) ``ES.'' At each step, new unfamiliar, or difficult concepts
were factored out and replaced with easier, more familiar concepts, until
nothing of value was left.
    The proverbial pig in a poke sans poke, expert systems as we now find them
are new only in the sense that a used car is new--it's the first time YOU'VE
owned it. Expert systems are, insofar as they are anything, matted tangles of
IF statements with no control structures. Instead they use "backtracking" and
"forward chaining" and "sideways guessing" to try all of the code in something
approaching random order.
    Forward chaining, by itself, is semantically equivalent to the normal order
of code execution in a procedural program. The one new control structure,
backtracking, isn't all that new and it isn't all that structured. In many
cases it isn't efficient and in most it isn't clear. Most expert systems are
astoundingly inefficient, and only run (or should I say meander?)
satisfactorily on the fastest of hardware.
    So why are they such a Big Deal? Because, like Columbus, the DP dilettantes
now known as knowledge engineers have rediscovered the lost continent of If and
the archipelago of Recursion. And these "discoverers" are hell bent on
exploring their new territory--never mind the fact that it is terra cognito,
well explored and heavily traveled. All of that can be erased by renaming
things: call assignment statements facts and IF statements rules, use
recursion, backtracking, statistics, and tarot cards to make sure the right
code gets executed eventually, and then junk the rest of computer science.
    Where systems analysts used to design programs based on interviews with
users, now knowledge engineers construct knowledge systems based on what they
learned during heuristic extraction from domain experts. Except for the number
of users willing to submit to a process called "heuristic extraction" there is
nothing new about this process.
    I have a simple test I use to tell if students have understood a topic, or
simply memorized it: I present the same material with arbitrary substitutions
for the nouns and verbs. If they understand they cry out "but this is exactly
the same as what we covered last week"; if they don't understand they take
notes.
    The computer industry has failed that test. We said "gee whiz" and cheered
"neat-o" when we should have yelled "what a gyp" and cried "con job!"
    Imagine how much money you could make re-selling the publishing industry
movable type, by calling it "modifiable letter forming technology." Imagine
being able to take credit for inventing the telephone. Imagine claiming to have
invented the internal combustion engine, just by renaming it the "liquid fuel
heat engine" and getting away with it. You could, if the rest of the world was
as gullible as the computer industry.
    Fortunately the rest of the world isn't that gullible; unfortunately we
are. And we suffer for it.
    There would be less harm in the expert systems charade and the ilk if not
for two insidious consequences. First is the decline in quality that invariably
follows; second is the increased chance the same sort of thing happening again.
    Expert systems are harder to debug and much harder to maintain because they
lack structure and the paradigm lacks the tools for imposing structure. This is
an inevitable consequence of the steal-part-of-something-and-rename-it
philosophy; you loose the infrastructure that held the old discipline together.
Progress is retarded because ground that had once been gained is needlessly
lost and must be painfully recaptured.
    Since ours is a rapidly growing community, many of the people exposed to a
fad are novices who know no better. They are easy prey at the onset, and often
become so enamored of the fad that they cling to it long after more experienced
fools have rushed to embrace new folly.
    Whenever something is as popular as expert systems have recently been,
vendors will flock to the field to make a buck. This is a fundamental feature
of capitalism. But when the field is a bog as ill-founded as the expert systems
paradigm, the vendors will be uncomfortable and very sensitive to criticism.
And these vendors are the backbone of the industry. The are the source of R&D
funds, they are the advertisers that keep magazines in print. When they're
sensitive to criticism then the industry isn't receptive to critical thinking
and another layer of gullibility settles over us all.
    This article should have been written and printed two years ago, but it
wasn't. No one wanted to hear these things, few people wanted to say them. It
spoils the fun of being a dabbler to have to think about things, make
comparisons, make value judgments. It's easier (and more enjoyable) to believe
what everyone else seems to believe.
    A few more years of this, and who knows what we'll fall for?