The article discusses how to deal with uncertainty when choosing a Bayesian network for calculations. One way is to put a distribution over all possible networks and parameters, then use this to find the most likely outcome. This can be difficult due to the large number of possible networks. Different methods like Monte Carlo algorithms or Bayesian model averaging can help simplify the process. Scoring and searching for good networks can be complicated by equivalent models and hidden variables. By searching within classes of networks instead of all possible networks, the process can be made easier.