I keep seeing people talk about prediction markets as though they're some magic way to gain knowledge that isn't publically available. (e.g. #9 here.) As I explained in a response to that post, this seems overly optimistic to me. Markets are very good about updating on public information once it's known, but can they reliably do so beforehand?
See for example all the election markets that just track the polls until vote counts start coming in and then update based on tweets. What prediction value are those providing exactly?
Prediction markets are great at aggregating public information, but I'm not convinced they're actually that useful for prediction of future events, especially when those events are highly contingent upon a small group of people or tightly-guarded information.
This market resolves to YES if someone gives me an example of a market (doesn't have to be on Manifold) about some notable real-world event that had a mysteriously high or low price for quite a while, followed by the reason for that price later becoming known.
For example, if this market had been at upwards of 20% before any of the news broke anywhere else, that would convince me that prediction markets might actually work for fraud detection.
Otherwise it resolves to NO at the beginning of 2025.
Related questions
@xander The market: https://manifold.markets/quinesweeper/will-there-be-another-wellrecognize. And a summary of what happened: https://news.manifold.markets/p/manifold-predicted-the-ai-extinction
The transmission mechanism of this is insiders buying "Yes", or making the market in the first place, as a kind of whistleblowing-lite.
It's the prediction market version of paralepsis: building a market on "did @Aella have sex with a scorpion" causes people to consider the idea, and ask why anyone would ask. A subsequent large "yes" bet, with or without comment, would be another important indicator.
This is the way prediction markets can (and will) reveal previously unknown events.
Any arguments that this shouldn't resolve YES based on the AI letter market?
@IsaacKing except as far as I know firstuser is saying he didn't have any inside info, and that he just pieced it together somehow using public information
@levifinkelstein Others have confirmed that they had inside knowledge: https://manifold.markets/JosephNoonan/does-anyone-have-genuine-inside-kno
Also, I think firtuserhere did have inside knowledge, he just didn't have any evidence he could provide for it, probably because that would have required him to reveal other details that he doesn't want to/isn't allowed to reveal.
@JosephNoonan the question is about whether the market elicits the knowledge, not whether someone has insider knowledge, so if it's not expressed in the market it doesn't count.
@levifinkelstein The people who were confirmed to have had inside knowledge also bet heavily on YES. And even if firstuserhere didn't have inside knowledge (I'm pretty sure he did), he was betting based on the behavior of people who did.
the question is about whether the market elicits the knowledge, not whether someone has insider knowledge, so if it's not expressed in the market it doesn't count
No, that's not true. If dmayhem and mkalquiera have inside information and don't bet on it, and then firstuserhere has a vague intuition that it'll happen based on mkalquera's discord messages and bets the market up to 50%, that is prediction markets working. They've successfully revealed information that nobody explicitly wanted to reveal!
However, from discord -
"From my PoV, I saw the letter, list of signees, and release date early and told @mkualquiera and @dmayhem about it on the 23rd. We all bet YES heavily, then the market seemingly exploded in popularity after the people betting it back down put it on the front page."
So it was expressed in the market. Firstuserhere piling on after is just him following mk and dmayhem
@IsaacKing I just read manifold's telling of the story and it doesn't sound like the elicitation was clear, even to manifold users, let alone to the broader world. CAIS were worried it would be, but it sounds like it didn't happen.
@IsaacKing Seems pretty clear the answer is yes based on https://news.manifold.markets/p/manifold-predicted-the-ai-extinction. CAIS wanted it deleted because it leaked information. It clearly leaked information according to manifold itself, the organization whose info was leaked, some of the leakers, etc.
It even satisfies the condition of the market having a "mysteriously high" probability, according to manifold itself and many traders on that market.
@makoyass The resolution criteria for this market are that some market has to have a mysteriously high probability, which very clearly did happen. So even if you don't think it clearly elicited knowledge, it would still resolve YES based on that criterion. But regardless of that, it think it's clear that it did elicit knowledge. Aside from the people with inside knowledge, no one had any reason to expect that the letter would happen, and we probably would all have put the probability at around 1% if it wasn't for the market's strange behavior. Some people bet YES solely because they suspected inside knowledge, and even the NO bettors were nervous about it. Even if the market only increased our credence from, say, 1% to 10%, the market still gave us the knowledge that a letter might happen, which we would never have thought before.
The press release also gives at least two examples of people who clearly elicited knowledge from the market. Firstuserhere suspected the existence of a letter, but he wasn't sure about the release date until others with inside knowledge bet in the market. Jack had no inside information, but said he was 95% sure about the existence of the letter based on others' bets (only predicting NO because he deduced that the release day was uncertain).
So I think the elicitation of knowledge is clear. As for whether it's a major event, the press release makes the case for that as well.
@JosephNoonan I think the point about increasing credence is a good one, as even though the final point probability might have been unclear, the rest of the information, volatility and comments did provide information. However, given that it is about the market "clearly" providing knowledge and that it was being manipulated, the burden of clarity rests upon the beholder, as many people in the market were manipulated, so on this point I am not sure this passes. Specifically, clearly eliciting knowledge to whom? The majority a minority or one or all.
@parhizj Well, the description states that the FTX fraud market having been at just 20% would be enough to meet the resolution criteria. So it doesn't have to convince a majority of people, it just has to be high enough to be suspicious. This market was hovering around 50% for the whole time and specifically led to speculation of insider trading, with a lot of people being convinced that that's what was going on. So the AI letter is well above the threshold for what counts as clear elicitation in the description.
@JosephNoonan ok good, so by that view "clearly" is satisfied. then whether it was an important event or not criteria is met is resolved by Isaac's judgment ultimately.
I really appreciate the discussion in this market. The question says "Important" and the description says "notable" event. I think to resolve what qualifies you could ask a number of authoritative sources at resolution time that would differ by profession, individual/group size, beliefs and preferences (culture, time preference, etc) and get different answers depending on the event. I think this relies too much on the the psychology of a single resolver rather than some less subjective aggregate measure, and instead this type of market should rely on some neutral / independent (to manifold) and aggregate sources.
I think that letter is incredibly and facially notable and important, especially given the broader rationalist community manifold's adjacent to
I believe the market creator isn't active any more so we might need one of the mods to take a call on this.
This clearly resolves YES from the market below.
Irrelevant because Hyperion on discord said he provided insider information to dmayhem and mkualqueira Two arguments against: maybe the big YES holders didn't, themselves, have inside information. They were just making what they thought were positive EV based on inferences from other peoples' behavior. But the description doesn't mention insider info - the title is "eliciting knowledge", and the description says "that had a mysteriously high or low price for quite a while, followed by the reason for that price later becoming known". Mysteriously high price was satisfied - most manifold users were confused by the high price.
Maybe this event isn't "important" or "notable", it's just a few sentences, and open letters are released all the time. Notability is trivially satisfied - the statement was signed by executives of most big AI labs - DeepMind, Anthropic, OpenAI. And it's not just a restatement of something obvious like many open letters, AI companies openly acknowledging the risk of human extinction is quite dramatic and, to many, unexpected. This should satisfy 'importance'.
The description also mentions "especially when those events are highly contingent upon a small group of people or tightly-guarded information" - and that does describe this letter's release date.
@jacksonpolack It wasn't that tightly guarded, apparently it was at one point posted on Twitter and then deleted.
tightly guarded is not part of the resolution criteria. it's under the motivation for posting the question, labeled with an 'especially when'.
Related questions
