02.03.2013 Views

Downloadable - About University

Downloadable - About University

Downloadable - About University

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Expert systems 443<br />

described above represent complete replacements for underwriters, who<br />

are seen as invaluable since they are still required to deal with the most<br />

difficult of cases and to handle changes in risk assessment circumstances,<br />

with which the systems themselves are unable to deal. For example,<br />

updates are required to the rule bases when formerly peaceable countries<br />

become war zones that a proposed insured plans to visit.<br />

Of prime interest here, however, is the nature of the underwriting<br />

domain which has made it amenable for the implementation of viable<br />

expert systems.<br />

Ifoneconsidersthenatureofthistaskmoreclosely,thenitisapparent<br />

that it:<br />

(1) Involves repetitive components;<br />

(2) Is a clearly structured logical process;<br />

(3) Is a high volume task;<br />

(4) Involves the utilization of knowledge that is fairly static (otherwise<br />

the system would require vast amounts of updating and reprogramming);<br />

(5) Involves variation in human performance which could (beneficially)<br />

be made more consistent.<br />

Furthermore, and importantly, the underwriting domain complies with<br />

the two criteria set out by Wright and Ayton for commercial viability,<br />

namely, most of the rules for underwriting tend to be formalized<br />

in manuals (in the case of life underwriting these are produced by<br />

underwriters in collaboration with actuaries and chief medical officers),<br />

and the domain is also verbally communicable in a step-by-step fashion<br />

to novices. Indeed, before the expert system was developed, this<br />

is how novices were trained. These factors ensure that the knowledge<br />

elicitation and engineering aspects of devising the expert system are<br />

achievable – aspects whose difficulty and pertinence to eventual system<br />

success are often underestimated. Thus underwriting differs considerably<br />

from the bulk of those marketing applications for which systems<br />

have been built – few of which (if any) are based on rules or knowledge<br />

which are stated explicitly in reference manuals, or which can<br />

be coherently and reliably elicited from the supposed experts. In our<br />

view, Rangaswamy et al.’s second and third checklist items for selecting<br />

problems ‘suitable’ for expert systems have, in themselves, prompted<br />

others to engage in overambitious projects. Indeed, one recent survey 31<br />

found that expert systems in use in the UK were not designed to cross<br />

functional and organizational boundaries but were instead utilized to<br />

automate relatively simple problem domains.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!