Here on Manifold we have a storied and sordid history of looking for ways to resolve a prediction market without any sort of operator intervention.
The hope is that this will let us create markets about questions for which there's no way to get an externally-verifiable answer.
Previous attempts have fared poorly:
(meme courtesy of @SneakySly)
In the interests of completeness I am testing this latest model. Cancelling a market upon completion removes the profit incentive at the heart of prediction markets, but at the same time, a poll weighted by users' past betting performance seems like it would have a good shot of ourperforming typical polls, and may be worth exploring as a model.
Related questions
This is not rocket science.
Sqrt(n) people vote, in secret.
80% to override = override
>50% = N/A
<20% disagree, resolve.
Votes revealed, later votes weighted by percent agreement with other honest actors.
As much as some of you hate crypto, this is a kindergarten level challenge to generate a consensus algorithm
(Hint: not through betting)
@Gigacasting Yeah I've been saying for a while that polls clearly work better than resolve-to-mkt. Still not perfect though and potentially exploitable by a determined attacker who can coordinate enough people. And it seems especially tricky to do a poll in a crypto setting - how do you protect against sybil attacks?
I think this is the same idea as https://manifold.markets/IsaacKing/does-this-kind-of-pay-to-vote-poll. I still am skeptical of it.
@jack I didn't see that one.
Hm... Jack is totally disinterested in this market... We're off to a promising start!
@dreev Right. Well, we'll see how it goes.
And yeah, if you try to take this answer and use it somewhere else, like as the basis for resolving a different market, rather than for pure knowledge, the manipulation incentives would return, even if this market itself gets N/Aed
Ha, fascinating. It's like a poll where you can weight your vote by how much capital you're willing to tie up, but you're guaranteed to get that money back. Sounds especially relevant to the question of whether Manifold should include a poll feature.
My conjecture is that this will work much better when used for real questions than when done as an experiment like this. I've noticed some people (probably including the creator of that meme!) seem pretty annoyed with the whole concept of auto-resolving markets.
For newcomers, here's my recap of why auto-resolving markets have been worth experimenting with:
Some predictions by nature don't have ground-truth resolution criteria (or maybe the market creator wanders off). In that case, can you go with the consensus of the market participants to pick the fairest resolution? Seems reasonable! But how do you aggregate their opinions? Well, that's exactly what prediction markets do. So maybe the prediction market can be to predict the market's eventual prediction? The stock market can be like that. Keynes famously introduced the metaphor of a beauty contest where you judge who's most beautiful by defining "most beautiful" as "whoever is judged most beautiful". Ie, it's totally circular.
(there are checks and balances on it for stocks: if the company gets bought or goes bankrupt or pays dividends, those all anchor the stock market's prediction to reality. just that in between any of those things happening, the stock market can go arbitrarily far up its own butt. see gamestop and meme stocks, etc.)
Anyway, a big problem with auto-resolving markets -- where the market resolves to whatever probability it ends up at -- is intentional manipulation. You get a group of wealthy people to drive the price to, I don't know, 69% and then that becomes by definition correct and they make a lot of money. The point of the prediction market -- answering some real-world question -- gets spoiled. We've experimented with a lot of ways to prevent that. My latest idea was that if the market doesn't resolve until the price fully stabilizes then could that solve the problem? Manipulators try to move the price but the non-manipulators keep moving it back to what's actually true (in these experiments we've picked questions where everyone agrees on what's actually true). You might think the manipulators can't win unless they have much more wealth than everyone else. But maybe sometimes they do? And just the possibility that they might can distort the market price.
Another thing I've learned is that auto-resolving definitely doesn't work if the real probability is close to 0% or 100%. This was kind of an epiphany for me (even though it should've been obvious from the market math all along) and explains previously mystifying things like how even some real-money, non-auto-resolving prediction markets said that Trump might still win in 2020 even after he'd already lost. A handful of zealots or manipulators can have a drastically outsized influence on the market price when the price is very near 0% or 100%.
