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.

454 Alternative decision-support systems<br />

these rules into an optimal set or sequence for execution when the system<br />

is used.<br />

In common with bootstrapping models, expert systems are most useful<br />

in repetitive decision or advice-giving situations. The reason for this is<br />

simple: if the conditions that lead to a decision or piece of advice<br />

are rarely encountered then the knowledge engineering time needed<br />

to model that ‘leg’ of the decision tree may not be cost effective. In<br />

many commercial applications it is far better if the need for extrasystem<br />

human expertise is recognized by the system and a complex<br />

problem is handed over to an expert for resolution. For example, in<br />

the life underwriting system outlined above, infrequent combinations<br />

of medical conditions and medical treatments that could, potentially,<br />

indicate that a proposal concerns a poor life risk are dealt with by the<br />

chief underwriter.<br />

Overall, there are several differences in the domain of applicability of<br />

decision analysis, scenario planning, bootstrapping models and expert<br />

systems. One commonality to all is the primacy of human judgment. As<br />

we saw in Chapter 9, human judgment is likely to be good when practice<br />

and useful feedback provide conditions for the quality of judgment to<br />

be evaluated. In decision analysis practice the decision analyst working<br />

on (what is usually) a unique decision can only check the reliability<br />

of the decision maker’s inputs of probability and utility. Questions<br />

to do with the validity (e.g. calibration) of the assessments are much<br />

more difficult to evaluate for one-off assessments given by non-practiced<br />

assessors. Fortunately, sensitivity analysis provides a fallback that at least<br />

allows identification of critical inputs. Decision conferencing techniques<br />

allow further analysis and discussion of these inputs. Clearly, in the<br />

absence of ‘the truth’ an achievable alternative of a group consensus<br />

or, at least, knowledge of the variability in the groups’ estimates is<br />

useful knowledge.<br />

We have argued previously that decisions, once made, are often<br />

‘made’ to work. For this reason, questions to do with the validity of<br />

decision analyses are often raised but seldom answered. Most often, the<br />

question of validity is sidestepped and questions concerning the ‘valuation’<br />

of decision analysis are raised instead, as we saw in Chapter 12<br />

when considering decision conferencing. Questions on the validity of<br />

linear modeling are more easily answered, since the method is most<br />

useful under conditions of repetitive decision making. As we have<br />

seen, this method has shown evidence of incremental validity over<br />

the holistic judgments/predictions of the judge on whom the model<br />

was based.

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

Saved successfully!

Ooh no, something went wrong!