What Prior Probability Do You Assign to the Hypothesis that Ghosts Exist?

     In the "Of Miracles" section of the Enquiry, Hume makes a proto-Bayesian argument to the effect that we should assign a very low probability to the hypothesis that a miracle has occurred, owing to his account of the laws of nature, which such a miracle would, by definition, contradict. He takes a law of nature to be a regularity observed to be universally consistent across all available relevant evidence, such that if evidence in favor of a miracle that contradicts that law is presented, we must weigh all the prior evidence for the law against the new evidence for the miracle. 

     The existence of a ghost would contradict basic principles of our modern scientific understanding of life, the mind, and the physical fabric of reality. According to Hume's rough model, all of the vast quantity of the mundane observations we have made that have confirmed these principles would weigh against whatever evidence that could be found in favor of the existence of a ghost. 

     Now, it seems that Karl Popper's account, taken simpliciter, would be at a loss in this situation. This is because the corollary to the principle that a universal statement, i.e. a law, can be deductively falsified by a single contrary piece of evidence is that an existential statement, i.e. the statement that there exists at least one instance of a given phenomenon, can be deductively verified by a single affirming piece of evidence. However, this means that the claim that a ghost exists would be deductively verified by a single piece of evidence that, on parity, would be sufficient in form and content to verify the existence of more mundane phenomena. For example, one good-quality photograph in the right context is sufficient to prove that a bird species thought to be extinct still lives; this occurred in 2015 with the blue-eyed ground dove. So, in the right context, one good-quality photograph of a ghost should be enough to prove the existence of a ghost. In practice, however, we find that this is not the case. Despite the long history of doctored photos and other demonstrated fakes alleging to represent evidence of ghosts, a number of fairly good photographs of supposed spectral events exist that have yet found no definite worldly explanation, and scientists and the public are very far from concluding that ghosts exist as a result. So, Popper's theory apparently needs something more here.

     However, I'm not sure that Bayesian Confirmation Theory can fully account for what is happening here either. The problem is that the prior probabilities we assign to P(e) and P(h) will depend not only on our prior beliefs about ghosts-- and this is a serious problem due to the fact that the scarcity of evidence doesn't allow for a ready "washing out" of priors-- but also on what we take each piece of evidence to be. The likelihood of a photograph containing an optical artifact due to lens flare or digital processing might be many times greater than the likelihood of a photograph containing the manifestation of a paranormal phenomenon, but the same image might stand as being either depending on whom we ask. This problem appears to go beyond the scope of prior probabilities, such that even if the inherent prior probability problem of BCT could be resolved, there would still be a larger question as to weighing evidence and determining what it actually represents in the grand scheme of our model of reality.