Original Lesswrong thread here.
Original tweet here:
Unlinked market with shorter timeframes here: /Joshua/when-will-we-know-that-any-past-ufo
As far as aliens are concerned, I generally vibe with the work done by Bostrom, Sandberg, Armstrong, Hanson, and the like. I suspect we're alone in our galaxy, but not necessarily in the universe, and the speed of light means that anything we'll see will be very long ago and far away. I expect the most likely observation to be some sort of megastrucure and the most likely interaction we would have to be getting absolutely annihilated at a distance by gamma radiation with no prior warning.
So my general outlook for this market: I expect it to resolve YES in 2028. I think we should apply a discount rate of 5% to it, and so in the meantime I will be scooting limit orders forward on both sides of that price up until then, which I think is the most sensible approach in the face of the regular flow of bets that are insensitive to time or price.
On the 23rd of each month, I've been publishing polls and markets like this that track AI progress: /SteveSokolowski/will-manifold-users-declare-that-we
This month, I'm going to start a similar poll series to track progress on people accepting NHI. How should I define the term "non-human intelligence" so that all the polls are as accurate as possible and also vague enough to withstand future revelations?
I want to leave it undefined ideally, but in the past that has led to people misunderstanding the question and answering YES thinking it referred to whales or dolphins. I also don't want to restrict to the word "alien," because there is no difference between "God" and a superior intelligence.
Given that there were no comments, I went ahead and defined the question:
I finally finished Imminent, after its release on August 20, and wrote this post so I can post it on multiple sites. Overall, I would say that I'm more convinced towards NO. Here are the key bullet points:
I see no reason to believe that what Elizondo writes about UFOs is false. Everything he writes is in agreement with the other whistleblowers.
Before Elizondo published the book, some users below were stating that all of this was secondhand. Even if one discounts the DoD evidence that he actually worked on his program, we now know the name and position of a firsthand witness making specific claims. It can no longer be stated that only secondhand claims exist.
Elizondo's credibility is boosted because he ignores some of the more far out topics that I think have little evidence to support them, like crop circles, alien races, the "prison planet," and others. I believe these topics are created by people who take things supported by strong evidence and then extrapolate too far. He also correctly criticizes people who claim that UFOs are "angels" or "demons" by labeling them religious zealots. "God" is equivalent to an alien using different terms, if it exists, and he believes those religious people harm scientific research on the topic.
His work is diluted by the chapter where he discusses "remote viewing." It's publicly known that the government did have a program on the topic that was cancelled, as he states. I also recall there is some evidence that things can be inferred with slightly greater probability than chance, and there are actually mobile apps trying to show that. But even if science knew what remote viewing was and even if it could somehow be used to spy, I don't see why it is relevant to the UFO topic, and even he doesn't make a good case why it has to do with UFOs. He even admits that he wasn't part of any programs involving remote viewing. The chapter should just have been omitted.
It's difficult to explain why Elizondo would have left government work if he was lying about working on the UFO program. He didn't have a job lined up, had to move to California, never lived in the house he bought because his salary was cut, and ended up in a small trailer in a small town. This book was published after seven years, so he clearly didn't have a plan to sell books and make money.
He provides new information about how Harry Reid played a much larger role than was previously known - Reid worked for years to get people to take this seriously, and made it his dying effort. He corroborates Grusch's statements that Grusch met with Reid and that Reid was unwavering that the UFO data needed to get out.
The series of events that led Schumer and others to take the subject so seriously is discussed at length and provides new information that there are significantly more Representatives and Senators involved in the legislation. He also goes into the 40 whistleblowers (without identifying them) and discusses Marco Rubio's involvement.
All in all, the book presents a consistent and credible story that avoids speculation on wild claims like Danny Sheehan is making. People making such claims don't stop to think about whether the government would actually be able to prove them.
Everything presented in the book is something the government actually can and would do, and the idea that the existence of non-human intelligence would be kept secret is not only plausible, but likely, given how TS/SCI clearances work and how there are other systems that have been secret for 50+ years that have not been exposed.
I think that we now have a credible outline supported by multiple independent firsthand sources of what is going on:
Elizondo and others worked on a program that tracked UFOs and investigated them, enabled by imaging technology that is not publicly available
Congress was not aware of this program and the reverse engineering efforts
The people in the DoD who committed fraud by misapproprating funds and deceiving Congress retaliated against Grusch with the false news and made threats against others
Lockheed currently possesses materials from craft of unknown origin
Lockheed and the DoD are lobbying against the UAP Disclosure Acts every time they come up due to legal liability, and Mike Turner, who represents the right district, blocks the bills
There are biological samples where the chain of custody has been lost, so they would have no scientific value
There are various known characteristics of what the craft do, such as being drawn to nuclear sites and water
The government knows more about UFO flight characteristics and frequency than is publicly known, but they have no knowledge of intent and don't understand their origin
Nobody can figure out how the technology works because the programs are too classified to get the best scientists working on them
@metacontrarian Thx, although I'm not in a place right now to be able to listen to audio. I definitely will look at this tomorrow morning when I am.
Here's another article that goes a little further than what I said above - the author, who previously published an article investigating the facts in part 1, basically agreed with most of my conclusions but also believes that the fact that UFOs are real is intentionally being used by adversarial countries to start investigations and watch how enemies' command structures work:
https://www.splinter.com/a-theory-on-what-senate-majority-leader-chuck-schumer-may-know-about-ufos
If he is correct about point #4, that might explain why "remote viewing" is repeatedly brought up in conjunction with this topic, including by Elizondo, even though the evidence for RV is so obviously weak compared to the strength of evidence for non-human intelligence.
Here's my thoughts on discount rates and how they affect prediction markets:
Markets with long duration attract large amounts of smart money only when the smart money believes they are mispriced enough relative to the duration (or, to be more precise, it's about how long they expect until the price moves). The current interest rate on USD is ~5%, and so if a market resolves in 2 years, you have to look at how much you think it's mispriced and see what return on investment (ROI) your bet would give you and compare that to 10%. If the ROI is too low, then little smart money will be interested in trading.
This means that it's better to view the market not as a point prediction where the market's average (money-weighted) prediction is equal to the price, but rather to look at it more like a confidence interval - the market's average prediction is in a range around the current price. For a market that resolves in 1 year may have a confidence interval of about 5% in either direction (this is a relative 5%, not absolute percentage points).
Additionally, this range is asymmetric - it's not centered at the market price. In a market where YES is favored, buying the higher-priced YES shares requires putting up much more capital, and the market has to move more to yield the same ROI. If one person thinks it's underpriced the market price needs to be farther from their prediction to be worth the same ROI as for someone who thinks it's overpriced. And because of this, if you think about a distribution of traders with different views, the market price will end up closer to 50% than the average prediction - and the higher the discount rate / the longer the duration, the bigger this effect is. See this comment for more detail.
I think all of this explains the behavior we see pretty well. On markets with significant disagreement in trader beliefs and with a price is far from 50% (e.g. this one, or whether Biden will resign, or whether an underdog will win, for example), the bigger the discount rate * duration, the more the market price tends to be closer to 50% than the average prediction (which you can see on e.g. Metaculus forecasts).
On the other hand, on markets where there's little disagreement, e.g. markets that everyone agrees are basically certain, or markets where everyone knows how to calculate the correct probability, a single trader betting a few cents can set the price to close enough to the true probability, so you don't need a lot of smart money to get a good prediction. That's why the "resolves YES in 5 years" markets don't reflect actual discount rates.
I think a fun research project would be to compare prediction market and non-market forecasts (e.g. Metaculus) across different durations and measure the effect of discount rates - I bet we would find them. (I'd do it if I had a lot more free time than I actually do.)
You could also experiment with prediction markets on questions that actually are interesting questions with substantial disagreement with different durations. One way you could do this is to look at forecasts on "Will <rare event that's expected to be uniform over time> happen in December 2024?" vs "in December 2025?" vs "in November or December 2025?" If the market is priced correctly without discount rates, you'd expect the prices of the first two to be the same and the price of a two-month period to be double the price of a one-month period. In reality, with discount rates I would definitely not expect it to end up priced like that, if there is both substantial disagreement and substantial trading interest in the questions - the farther the expiration date the closer it'll be priced to 50%, and furthermore the two markets that both expire in December 2025 will not be priced "correctly" where the two-month period has double the probability of the one-month period.
Also, we know how to greatly reduce discount rates on prediction markets. The simplest solution is for positions to accrue interest at market rates (IBKR's prediction market does exactly this). You could also have positions be denominated in a market index like the S&P 500. See my comment on that (I and many others have been describing this solution for years, it's nothing new.)
It's not that hard, and although it does cost Manifold money compared to keeping the interest for themselves, you get an increase in trading volumes and more useful markets.
I think you have some fair points here. I guess the crux of what you’re saying is that the level of uncertainty / disagreement interacts with the long resolution time to depress the price. Possible.
Maybe this is a straw man, but what I saw here was people saying “this has zero chance of resolving NO, the only reason it’s not trading at 99% is discounting” and that’s what I disagree with. There’s uncertainty. Sure, the price is also affected by discounting but it’s also not a sure bet that it resolves YES.
I don't think anyone is actually saying it's literally zero chance. When people say bond rates, they're referring to the risk-free interest rate and how market prices are often mispriced by that much. And when people talk about markets being bonds, it's a shorthand for saying the prediction market overprices the unlikely outcome, it doesn't mean it's guaranteed.
But there are many people who think it would be priced at 99% without discount rates, and I think that's not far off. https://www.metaculus.com/questions/7384/alien-tech-in-solar-system-before-2030/ for example is at 1% chance that a consensus for aliens develops by 2030.
Agreed! @jack I think I may have been the first to use this term on Manifold a while back (mostly because I don't know much about finance and forgot the probably more relevant term "discount rate" lol) and I think I used it in the context of something like:
"In real-money prediction markets, you'd never put your money into a market earning less than the annual rate of return of a gov bond, say, because that's almost entirely risk-free whereas prediction markets are inherently risky, even if you're almost certain that you're correct, because of crypto trading rates, inaccurate resolutions, disagreements, etc."
@jack 100% agreed. Relevant market: https://manifold.markets/JonasVollmer/will-manifold-pay-interest
If anyone is interested, I'd happily bet my 1000 NVDA shares to your 20 NVDA shares that this market will resolve YES.
@Joshua , this poll is meaningless. I had bets open for six months on this page, when it was trading at 97%, offering $1000 or $2000 in exchange for a recorded lien against a house in the United States in any but two states, and not a single person took this offer up.
You can review down below - the idea was that if they won, they keep the money; otherwise, I can foreclose on the house and evict. Given that people were claiming this market was a farce, they should not have been worried about eviction and it was free money.
There certainly has not been any data published that suggests the odds are lower now than earlier in the year; if anything, Schumer's speeches and the Congressional actions and the content in Imminent make it more likely.
The people answering this poll with odds "1000:1 or greater" are lying because they did not accept the offer when it was valid (which is not anymore, because I spent the money on nVidia stock.)
Because who wants to do all that for a bet where they stand to gain 3% times 2k = 60 dollars? It makes no sense.
People did make over 100k USD of real money bets on the page linked in the description.
No, the poll is assuming or handwaving away the practical difficulties in arranging such a bet, and therefore trying to get at your real beliefs, rather than your willingness to put up with counterparties and paperwork. You cite the lack of real-money trades, but people also offered to bet real money on YES for months as well, and nobody took them either. Because arranging ad-hoc trades outside the market is extremely annoying and inefficient - that's why we have the markets.
Also:
people did make the bet on this market at 97% when you had loans to unlock the mana. To discount that, you have to argue that predictions with mana aren't valid
people did make the real-money bet on the linked lesswrong post. To discount that, you have to argue that the difference is they didn't put down collateral and don't intend to pay up
people did offer both YES and NO bets here in real money, but nobody ever took them (AFAIK) because making bets like that is a horrible hassle. To discount that, you have to argue that the offers from one side were fake.
To discount all three of those, you have to believe all sorts of different reasons they don't count. Isn't the explanation that people are just stating their honest beliefs more likely?
I think you have some good points for many of these issues and you hit upon the key issue in your second point. That one is the core of the matter.
Unsecured debt is worthless. As I said in another comment on Manifold somewhere, most people don't know that you can rack up $100,000 in credit card debt, and refuse to pay the bills. Half the companies will simply write it off (a 66% discount after taxes) and the others will sue and then offer a settlement for 40% (a 40% discount after taxes.) You can basically earn $50,000 and your credit score starts to recover after a year, and it's all gone after seven. So, with interest you could do this repeatedly your entire life and earn a million dollars.
Whoever owns Yudkowsky's debt doesn't even have a signed contract. He doesn't live in the same jurisdiction, and depending on how it is set up he might even be legally required to hire a lawyer to sue. With no collateral, even in the miracle he does win, Yudkowsky will just declare bankruptcy and discharge the debt.
@jack , there is a massive difference between a secured loan with a contract and these "real money bets" being made on lesswrong.com - so much so that if the market resolves NO I would make a secured bet on the derivative that there would not be a single bet over $100,000 that is paid in full.
That's the real reason why your statements about the markets being inefficient is true - it's because the legal system, at least in the United States, is ineffective and relies on fear. The only people who fear prison or lawsuits are those who were "trained" to think that way in elementary school.
You missed my point. I said that there are three types of bets, and I said you could raise plausible arguments to discount them individually, but that you'd need to believe separate and different arguments to discount all three. And in your comment you are just saying the same thing I said regarding point 2. But then what about the other two types of bets? People did bet in the high 90s right here in this market, when we had loans to avoid locking up our mana.
@Joshua Thanks for making this, I consider myself sceptical but looks like Manifold at large is even more so lol
@Joshua It's just occurred to me, would we be able to achieve the same end result as the hypothetical in the poll by resolving yes now and re-resolving if it turns out to be wrong later?