In practice, machine learning deals with supervised and unsupervised learning problems where one can compare, in some metric, the obtained results with real values of goal variable in test sample, or some set of benchmarks. Latent goal variables that are unobservable or unavailable for a researcher introduce an interesting challenge for data scientists. This problem may arise in several Artificial Intelligence domains, such as prediction of personal reaction on events, subjective assessments etc. There are two main challenges coming up in these cases: (a) to build an appropriate machine learning algorithm; (b) to provide reliable performance indices. Several approaches have been suggested to overcome the above challenges.