While, as noted by others, pseudolikelihood is unlikely to be what humans actually use, I think it is interesting to ask whether some cognitive biases come from some sort of approximate inference. Designing an experiment to test this conclusively would be quite challenging, but very useful.