Just a wild guess, but I wonder if there could be a nice connection between probabilistic programming and logic languages like prolog, if only due to the fact that both Bayes graphs and unification both have this "run forwards and backwards" characteristic that is so unlike imperative programming
The Probability monad allows for computing exact probabilities, as stated, but that more-or-less misses the point of newer probabilistic programming approaches: avoid the exponential blow-up in complexity of inference due to marginalization by generalizing from evaluation to sampling rather than from evaluation to probability assignment.
OT of course, but I think that's what our outsourcers did: keep adding monkeys then the probability of completing a programming task slowly converges on 1...