Austin from https://manifold.markets here -- great post! At Manifold, one of our goals is to be "Twitter for prediction markets" - hence our emphasis on user-created markets, allowing anyone to set up a market for their hot take.
(aside: I don't know that "trilemma" is exactly the right word here; there's no fundamental reason that a PM platform couldn't have all three of user-created, easy-to-use, and real-money. We're currently working on several initiatives to make our fake currency more like real money, such as allowing withdraws to charities, and hosting cash-prize tournaments.)
I share your puzzlement at the observation that intellectual support for PMs far outweighs their real-world usage. Here are a few other explanations that I like:
- For the corporate use case: any single market conveys very few bits of information (just a probability from 0 to 1). This is good for letting two sides know what they disagree on, but bad for searching through the total space of ideas. E.g., at the point when I ask a question "If we implement dark mode for our site, will our DAU increase by 15% in the week afterwards?" I've already done a ton of work to narrow down the space of all possible features to ask, and all possible metrics to track. Getting a more accurate probability estimate of that question is only like 5-10% of the whole value of the question.
(Incidentally: we're trying to solve this problem with our Free Response markets! This lets someone pose a question "What features will be valuable for Manifold" and users to respond and vote on suggestions StackOverflow-style, except their votes are bets.)
- For operationalizing Twitter disagreements: it's actually quite difficult and time-intensive to arrive at a single statement that two different sides are comfortable betting on! I believe this is called a "crux" by LessWrong folks. For example, on this public disagreement https://www.lesswrong.com/posts/sWLLdG6DWJEy3CH7n/imo-challenge-bet-with-eliezer, Eliezer and Paul apparently spent a while to arrive on a concrete disagreement here over their differing intuitions of how the AI scenario will play out, discarding a few other candidate proposals throughout the way.
Finally: I'm always happy to chat more in-depth on anything prediction-market related; reach out at akrolsmir@gmail.com!
Thanks for this Austin! And editing language on trilemma too.
The commonality amongst the points you mentioned are the aspects that either the outcomes are hard to measure and understand, or they're trivial and not particularly interesting. I do think there are plenty potential markets in the middle. In your corporate example, "will buying a goldmine add more market cap" would've been a great bet for AMC shareholders. But exactly as before my worry is that lack of liquidity is an incredibly hard problem (see my usual point re OTC markets for us equities on how hard liquidity is).
Also I don't think we'll ever hit the "create a market for everything you want all the time" utopia. It's more likely to perhaps give a) a way for folks to grade their predictions (a la SSC style), b) easy plugins to existing social networks to easily add a credibility score. That would already be a major step forward.
I wonder who stands to lose based on anything any opinionati (opinionatus?) might say. Is there any real world significance at all to such predictions. Is anyone making important life choices based on such predictions.
I have noticed a lot of Op-Ed writers who literally write what VCs want them to. But the entire web is already filled with their PR spam. Like AI Techno-Optimism. Even the op-ed writers have been bought out.
Well definitely easier but not sure if it'll be seen as equally valid. In the fin markets there's plenty algo trading fighting each other to arbitrage but need enough opinionated traders for a proper price setting. This is the liquidity issue.
But that's in part a facet of the market being high liquidity, positive sum and improperly digitalized (hence why hft is a thing).
Low liquidity, zero sum markets build digital from the ground up would (maybe) fix algorithmic noise.
And it seems fair to speculate that a good trader could off load higher level intuitions into an symbol based model ala nlp transformers in the future. But alas that is empty speculation.
I disagree on improper digitalisation, since there's always a facet of time (or other variables) where the markets efficient for you but not others. Eg the markets efficient for me on my timescales but maybe not citadel.
I think algorithmic noise is a feature not a bug, don't think it'll ever go away. And that's ok. We need the froth.
My bigger pov is all all sources of alpha are illegible ultimately, meaning yes it's possible (in extrapolated theory) to do what you're saying, but alas not today.
Austin from https://manifold.markets here -- great post! At Manifold, one of our goals is to be "Twitter for prediction markets" - hence our emphasis on user-created markets, allowing anyone to set up a market for their hot take.
(aside: I don't know that "trilemma" is exactly the right word here; there's no fundamental reason that a PM platform couldn't have all three of user-created, easy-to-use, and real-money. We're currently working on several initiatives to make our fake currency more like real money, such as allowing withdraws to charities, and hosting cash-prize tournaments.)
I share your puzzlement at the observation that intellectual support for PMs far outweighs their real-world usage. Here are a few other explanations that I like:
- For the corporate use case: any single market conveys very few bits of information (just a probability from 0 to 1). This is good for letting two sides know what they disagree on, but bad for searching through the total space of ideas. E.g., at the point when I ask a question "If we implement dark mode for our site, will our DAU increase by 15% in the week afterwards?" I've already done a ton of work to narrow down the space of all possible features to ask, and all possible metrics to track. Getting a more accurate probability estimate of that question is only like 5-10% of the whole value of the question.
(Incidentally: we're trying to solve this problem with our Free Response markets! This lets someone pose a question "What features will be valuable for Manifold" and users to respond and vote on suggestions StackOverflow-style, except their votes are bets.)
See also: https://forum.effectivealtruism.org/posts/dQhjwHA7LhfE8YpYF/prediction-markets-in-the-corporate-setting
- For operationalizing Twitter disagreements: it's actually quite difficult and time-intensive to arrive at a single statement that two different sides are comfortable betting on! I believe this is called a "crux" by LessWrong folks. For example, on this public disagreement https://www.lesswrong.com/posts/sWLLdG6DWJEy3CH7n/imo-challenge-bet-with-eliezer, Eliezer and Paul apparently spent a while to arrive on a concrete disagreement here over their differing intuitions of how the AI scenario will play out, discarding a few other candidate proposals throughout the way.
Finally: I'm always happy to chat more in-depth on anything prediction-market related; reach out at akrolsmir@gmail.com!
Thanks for this Austin! And editing language on trilemma too.
The commonality amongst the points you mentioned are the aspects that either the outcomes are hard to measure and understand, or they're trivial and not particularly interesting. I do think there are plenty potential markets in the middle. In your corporate example, "will buying a goldmine add more market cap" would've been a great bet for AMC shareholders. But exactly as before my worry is that lack of liquidity is an incredibly hard problem (see my usual point re OTC markets for us equities on how hard liquidity is).
Also I don't think we'll ever hit the "create a market for everything you want all the time" utopia. It's more likely to perhaps give a) a way for folks to grade their predictions (a la SSC style), b) easy plugins to existing social networks to easily add a credibility score. That would already be a major step forward.
I wonder who stands to lose based on anything any opinionati (opinionatus?) might say. Is there any real world significance at all to such predictions. Is anyone making important life choices based on such predictions.
There are always a few, but pollution of the epistemic commons is a bad thing in general
I have noticed a lot of Op-Ed writers who literally write what VCs want them to. But the entire web is already filled with their PR spam. Like AI Techno-Optimism. Even the op-ed writers have been bought out.
Does it make more economic sense if you replace "prediction market" with "Netflix search" and "pundit" with "recommendations Algo" ?
I think some of the drawbacks of prediction markets go away if you assume they are people-operated but the forecasters are all bots.
Well definitely easier but not sure if it'll be seen as equally valid. In the fin markets there's plenty algo trading fighting each other to arbitrage but need enough opinionated traders for a proper price setting. This is the liquidity issue.
But that's in part a facet of the market being high liquidity, positive sum and improperly digitalized (hence why hft is a thing).
Low liquidity, zero sum markets build digital from the ground up would (maybe) fix algorithmic noise.
And it seems fair to speculate that a good trader could off load higher level intuitions into an symbol based model ala nlp transformers in the future. But alas that is empty speculation.
I disagree on improper digitalisation, since there's always a facet of time (or other variables) where the markets efficient for you but not others. Eg the markets efficient for me on my timescales but maybe not citadel.
I think algorithmic noise is a feature not a bug, don't think it'll ever go away. And that's ok. We need the froth.
My bigger pov is all all sources of alpha are illegible ultimately, meaning yes it's possible (in extrapolated theory) to do what you're saying, but alas not today.