02.03.2013 Views

Thinking and Deciding

Thinking and Deciding

Thinking and Deciding

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

134 NORMATIVE THEORY OF PROBABILITY<br />

With a tree like this, it is easy to calculate the probability of a given set of symptoms,<br />

as long as each line (branch) on the tree has been assigned a probability <strong>and</strong> as<br />

long as we know the prior probability of the disease. To get some idea of how this<br />

is done, consider only the probabilities labeled a, b, <strong>and</strong> c in the tree diagram; these<br />

are, respectively, p(P1|D1), p(S1|P1), <strong>and</strong> p(S2|P1). Considering only this part of<br />

the whole tree, the probability of S1 <strong>and</strong> S2 together, given D1, would be a · b · c.<br />

This is because the probability of the pathstate P1 given D1 is a, <strong>and</strong> if the pathstate<br />

is present, the probabilities of the symptoms are b <strong>and</strong> c, respectively. Because the<br />

symptoms are independent, given the pathstate, we can multiply their probabilities.<br />

(For more details, see Duda, Hart, <strong>and</strong> Nilsson, 1976; Kelly <strong>and</strong> Barclay, 1973; Pearl,<br />

1982; <strong>and</strong> Schwartz, Baron, <strong>and</strong> Clarke, 1988.)<br />

Computers make complex calculations easy, <strong>and</strong> it may seem as though all we<br />

need to do, in order to construct an expert system, is develop the best possible normative<br />

theory <strong>and</strong> then program a computer with it. We must remember, however,<br />

that expert systems are truly systems. They are not only computer programs: rather,<br />

they are computer programs, plus their justifications, plus (most important) procedures<br />

for eliciting probabilities from people (experts). Experts must be given the<br />

best opportunity to use their knowledge, <strong>and</strong>, if necessary, they must be helped to<br />

think about it in the most useful way. Thus, expert systems are prescriptive solutions<br />

to practical problems. They are not necessarily just realizations of a normative<br />

model. In particular, we need to do more descriptive research, to find out what kinds<br />

of structures are most suited for elicitation of personal probabilities from experts <strong>and</strong><br />

to determine the extent to which expert opinion <strong>and</strong> frequency data can each be relied<br />

upon. The design of expert systems depends as much on the psychology of experts<br />

as on the mathematical possibilities.<br />

Conclusion<br />

The theory of probability describes the ideal way of making quantitative judgments<br />

about belief. Even when we simply judge which of two beliefs is stronger, without<br />

assigning numbers to either belief, the one we judge to be stronger should be the<br />

one that is more probable according to the rules of probability (as applied to our<br />

other beliefs). Let us now turn to the psychological research in this area to find out<br />

how people ordinarily make judgments of these types. The difference between the<br />

normative theory (described in this chapter) <strong>and</strong> the descriptive theory (Chapter 6)<br />

will suggest some prescriptive improvements.<br />

Answers to exercises:<br />

1. p(neg|ca) =.208, p(neg|ben) =.904, p(ca) =.01<br />

p(ca|neg) =<br />

p(neg|ca)p(ca)<br />

p(neg|ca)p(ca)+p(neg|ben)p(ben)<br />

=<br />

(.208)(.01)<br />

= .0023<br />

(.208)(.01) + (.904)(.99)<br />

The lesson here is that negative results can be reassuring.

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

Saved successfully!

Ooh no, something went wrong!