The obvious ways of getting a distribution out of it involve multiple queries. It is unclear how you would even represent a distribution without using exponential compute.
Of course, the strategy described here would still take exponential compute, but the hope is for an algorithm that is similar in what its trying to do and faster in practice.
Answering with a point estimate seems rather silly. Shouldn’t it answer with a distribution? Then one question would be enough.
Logical induction is the algorithm described at https://intelligence.org/files/LogicalInduction.pdf.
The obvious ways of getting a distribution out of it involve multiple queries. It is unclear how you would even represent a distribution without using exponential compute.
Of course, the strategy described here would still take exponential compute, but the hope is for an algorithm that is similar in what its trying to do and faster in practice.