🚀 go-pugleaf

I am writing this article because I feel very bitter about an experience
that occurred today.  I'll have to be very careful, because I don't want
to name any names or create bad feelings.

A couple of months ago, a certain large manufacturer of complex electronic
equipment placed an advertisement in a local newspaper inviting people and
organisations with experience in Artificial Intelligence to get in touch.
I did (not so much presenting myself as in part RMIT and in part a research
organisation I have links with).  They wanted more information, so I passed
the torch to someone at the research organisation.  Let's call the company
C, the organisation O, and the person E.  E is not an AI expert, although
he has a PhD in something.  He is the industry liason, and believes whatever
the experts in O tell him.

C first of all tried to deal with another organisation, but finally came
back to O.  There was a meeting with me, E, and a representative of C.
C make, as I said, a kind of complex electronic apparatus which is very
popular indeed these days.  They are US-based but present world-wide, and
have a deserved reputation for good equipment and good service.  The trouble
is that they keep on introducing new models (and not just new models; they
are about to introduce a digital version of what was formerly an analogue
system).  It is hard for the technicians to keep up with the new systems.
(They already make as much use as they know how of automatic test equipment.)
So C would like O to make them an Expert Diagnostic System which can
interact with a human technician and the automatic test equipment to diagnose
faults in this kind of complex electronic apparatus.

My reaction was that there is a lot of work to do this, but that it is doable.
Quite a lot has been published about diagnosis, much of it on electronic
equipment, and as for incorporating their existing test plans, the approach
SRI took for Space Shuttle diagnosis looks good.

It also seemed to me that this was an application to which Prolog was ideally
suited (have a declarative representation of the "anatomy" (500+ components),
"physiology", "behaviour" (these devices do have a lot of states, but not as
many as, say, a non-programmable pocket calculator), and known failure modes
of the components; represent existing test plans a la PRS; then have an
"inference engine" for doing reactive hypothesis-and-test).  I went further
in my mind:  I thought "LPA Prolog-386 would let them use cheap stock
hardware to do this".

Most of all, however, I thought we needed a lot more information from C
before making any detailed plans.

This morning I received E-mail saying
	[Because C want to take the program over eventually, so they can
	maintain it, not O]
	"we believe that C will be unlikely to seek a system written in
	Prolog - the language is rare in business in the US, and C,
	as you know, is a US based company."

There was other information in that and a preceding message going into a lot
of detail about what other experts at O have told E (I myself don't quite
see how to apply case-based reasoning to the diagnosis of new products for
which there _are_ no cases, which is the ultimate goal of the project, nor
no I quite understand why a vision expert would tell E that it was 
impossible to diagnose "new" faults).  It seemed to me that everyone (and
yes, I have to include myself in that, if I'm honest, although I did at
least consult some of the diagnosis literature) is proposing their pet
technique.  I decided that harmony was better than fame (although, let's
be honest here, I _am_ rather peeved that _every_ single suggestion made
by the person who brought the business in in the first place has been
rubbished, ignored, or misunderstood.) and said "include me out".

But I also described "Company C is a US-based company, and Prolog is
rare in US business, so C won't be interested in a Prolog-based product"
as an "outrageous lie" which won me an angry phone call from E demanding
that I produce evidence to support this claim.  I apologised for my
phrasing, but asked E to produce _his_ evidence.  It turns out that X,
who has a _lot_ of experience writing expert systems and is generally a
nice guy as well as knowledgable about AI in general, told E all this
about Prolog being rare in US business.  (When pressed, E claimed (truly)
that Prolog is not as common as COBOL or C.  This is news?)

Anyway, there is the substance of my plaint.
    X, who is certainly competent at producing expert systems, tells E
    that Prolog is rare in US business,
    E, who believes his experts and worries about business, accepts this
    and designs his presentation according, so
    C, who just want a product, will be told
	- we're going to use Expert System Shell <A>
	- and case-based-reasoning package <B>
	- and neural net package <D>
	- and hypertext package <F>
	- and graphics system <G>
	- but we WON'T use Prolog.

We're talking here about a program which will be installed in _thousands_
of locations world-wide, if they can write it.  We're talking about a loss
of a great opportunity for some Prolog company to get their toe in the door
of an electronics giant.

Heck, if it comes to that, company C _does_ know about Prolog; they paid
Quintus money to put Quintus Prolog on one of their machines, and they got
ALS to put ALS Prolog on another of their machines.

So, in context, was it extreme to call "an outrageous lie" the claim that
"Prolog is rare in US business so company C won't be interested".

-- 
You can lie with statistics ... but not to a statistician.