Il piu' certo modo di celare agli altri i confini del proprio sapere, é di non trapassarli. Giacomo Leopardi, Pensieri, LXXXVI

In this and in the following chapter we present some relatively more-advanced techniques that can help in the construction of the Bayesian nets. We focus on the elicitation problem in this chapter, and on a miscellany of additional tools in the next, which can broadly be grouped under the rubric of facilitating the analysis of the quantities of interest (e.g., obtaining joint probabilities for a subset of variables, deriving implied rather than assigned conditional probabilities, etc.), or of enabling the handling of large Bayesian nets.

We said ‘relatively more advanced’ because, by the standards of sophistication required to deal with hundreds or thousands of variables (see the discussion at the beginning of Chapter 13), these remain entry-level techniques. Since, however, we want to keep our approach as transparent and non-black-box as possible, this is about as far as we are comfortable to go in terms of algorithmic complexity.

The underlying issues are deep and fascinating, pertaining as they do to the topics of causation and inference, and we therefore provide throughout this chapter some bibliographic pointers for the interested reader. We are aware that, once causation is brought into play, matters can quickly become very complex, from both the mathematical and philosophical point of view. Hence the quote that opens this chapter.