[mod.ai] economics of expert systems - summary of replies

ross@psych.uq.OZ.AU.UUCP (05/24/86)

A while back I put out a request for information on the economics of the
development and deployment of expert systems. This is a summary of the replies
I have received.

I received around ten replies, most of which were of the 'please let me know'
variety. Some of these went to some length to indicate that they felt this
was an important area. It does seem that there is a need for this information
and it either doesn't exist or somebody is not sharing it.

There were three substantive replies which told of:

1	A company which attempted to develop three expert systems.

	One took twice as long to develop as the FORTRAN program it replaced,
	the second was too slow to be usable, and the other was abandoned for
	lack of an expert.

2	A successful family of expert systems that are widely used in-house.

	The point made here was that the development cost was an insignificant
	fraction of the cost of packaging the product for deployment and the
	continuing cost of training the users.

3	A pointer to the November 1985 IEEE Transactions on Software
	Engineering which was a special issue on "Artificial intelligence and
	software engineering".

	I found the articles by Doyle, Bobrow, Balzer, and Neches et al to be
	the most relevant to my needs. Doyle argues that the productivity
	advantage of the artificial intelligence approach comes from the tools
	and techniques used to construct the product, not from the ultimate
	form of the product itself. The other papers do not explicitly address
	the modelling of costs. However, an implicit model is discernible from
	the areas they choose to emphasize.

I will send a request to the software engineering list and see if I can get any
joy there. If not it looks like I might be forced to do some work for myself.
What I would like is a predictive model which will give me the costs to
implement and deploy an expert system or conventional system as functions of
various features of the problem, the tools available, and the development and
deployment environments. As I do not have any empirical data the best I can
aim for is a set of statements on the qualitative shapes of the cost curves for
various factors. Using these curves backwards would allow me to say what
problem characteristics are a lot more conducive to an expert system solution
being cheaper than a conventional solution. I will probably start with the cost
models in Tom de Marco's book, "Controlling software projects" and try to
identify expert systems analogues of the cost factors he identifies for
conventional systems.

If I manage to get anywhere with this I will let you know.

Ross Gayler			| ACSnet:	ross@psych.uq.oz
Division of Research & Planning	| ARPA:		ross%psych.uq.oz@seismo.css.gov
Queensland Department of Health	| CSNET:	ross@psych.uq.oz
GPO Box 48			| JANET:	psych.uq.oz!ross@ukc
Brisbane	4001		| UUCP:	     ..!seismo!munnari!psych.uq.oz!ross
AUSTRALIA			| Phone:	+61 7 227 7060