My sense is that, for prediction markets to work, there needs to be some real knowledge/analysis/judgement spread across at least a material subset of participants. Simply aggregating random guesses is likely no better than any given random guess.

Put another way there needs to be SOME signal buried in all the noise.