Model Iteration
For some applications of decision nets (especially in areas like gambling, control or automated decision making), it is relatively easy to set up the tables to suitably reflect reality, and the decisions that the net makes are much better than decisions you would make unaided, because of Netica’s superior ability to reason precisely with probability, and to deal with many interactions in very complex situations.
However for other applications (especially in areas like business decision making or public policy), the decisions that Netica makes may not match your “gut feelings”, and you may suspect that Netica is wrong. In that case, usually the decision net doesn’t capture the decision problem well enough. The usual procedure is to go back to the decision net model, change it to more accurately reflect reality, and solve it again. Then repeat the process until you are satisfied with the results.
But, of what use is the decision net if all you do is try to get it to match the decisions you would make anyway? Professional decision analysts who use decision nets on a regular basis tell us that they are invaluable for disciplined decision making. The process of building the decision net clarifies the problem (often slowly changing your gut feeling as you proceed), and the iteration process forces you to re-examine assumptions you have made.
In a group decision making environment, creating a decision net is often of great benefit in fleshing out the differences in people’s beliefs and values, and allows people to discuss specifics rather than arguing in generalities.
Also, once the policy is formed, the decision net acts as a living document of the beliefs and values that lead to it. If those change over time, the decision net can be changed to generate a new policy. Or parts of it can be copied and pasted into new nets for new decision problems. If a policy results in a bad outcome, it is possible to go back and determine whether it was based on bad information, and if so, to change that information for the next decision making situation.