Build the normative asteroid Bayesian network of Figure 5.19 and put in some likely prior probabilities and CPTs. Make a copy of this network, but increase the prior probability for P4; this will play the role of the user model for this problem. Operate both networks by observing P1, P2 and P3, reporting the posterior belief in P4 and G.

This is not the way that NAG operates. Explain how it differs. In your work has

“explaining away” played a role? Does it in the way NAG dealt with this example in the text? Explain your answer.