Non-deterministically prefer hypotheses which have higher truth expectation

[patham9]

This concerns:

• Resource allocation (do not always consider all hypotheses but a subset)
• Exploration (what if some causal knowledge is wrong and the next step does not lead to the necessary feedback to correct it? avoids trying the same stuff again and again)