https://imaginatorium.org/stuff/nose.htm
Basically, making an average of a large number of estimates of an unknown value will (of course) fail if most/none of the estimators have any idea of the actual value being estimated.
It's the same with this topic. You won't get an answer to the question "Is this particular paper true or not?" but you'll get an answer to the question "Are papers submitted under these circumstances making claims like this likely to be true?". The crowd will only answer the question they can answer. I think that's from "Thinking fast and slow".
The point of the story when I originally heard it is that no one has SEEN the Emperor's nose. So any statistical function (like averaging) of estimates is totally useless as they are all guesses.
No one has seen 'papers submitted under these circumstances' so no amount of 'crowd wisdom' will make any difference.
Also, as an aside the idea that 'the crowd will only answer the question they can answer' is ... bizarre. People will answer anything you ask them, and you have no way to know if they are just making it up.
A subtle distinction is who is allowed to participate in a prediction market.
"Everyone with $1" is a terrible answer, and produces the bad results people are pointing to.
Financial markets avoid this because of their scale, where there's enough smart money to (usually) punish stupid money.
Absent that scale, it's just stupid money muddling the decisions of smart money.
Prediction markets with a knowledge barrier to entry would produce better results.
Also, suggesting that there is such a thing as 'smart' money - presumably due to having more of it? - is amusing. As pointed out elsewhere in this discussion, there has been a lot of smart money acting particularly dumb over the last few years.
And the relevant question isn't whether 'smart' money does dumb things: it's whether 'smart' money does dumb things less frequently than a random sample of money.
No one is an oracle, and there are absolutely outlier events that specifically confound experts, but I can't believe that increased expertise is uncorrelated with increased accuracy.