Will Eliezer Yudkowsky win his $150,000 - $1,000 bet about UFOs not having a worldview-shattering origin?
373
14k
3.1k
2029
89%
chance

Original Lesswrong thread here.

Original tweet here:

Unlinked market with shorter timeframes here: /Joshua/when-will-we-know-that-any-past-ufo

Get Ṁ600 play money
Sort by:

I think this is >99% likely, so I'm basically treating this as a bond.

1 / (1 + 4.4% US 5y treasury rate) ^ 4.5y to resolution * (1 - 30% probability of Manifold paying interest on shares in a market) + 30% probability of Manifold paying interest on shares in a market = 88%

So, 88% should be roughly the max price that people should be willing to buy this at.

Sources:
https://www.ustreasuryyieldcurve.com/
https://manifold.markets/JonasVollmer/will-manifold-pay-interest

Given the uncertainty in various factors and the current cap of 88%, I'm cancelling my 86% limit order

@JonasVollmer I wish there was a reference curve of “resolves yes on x” markerta so we know what the risk-free rate is here. I’m not convinced it’s anywhere close to the fed rate, mainly because people are not here purely for bottom line. Is there such a thing and I’ve missed it?

@Ansel You could check the various "AI wipes out humanity" markets - there are many of those on different dates.

It's generally been possible to do much better than those, though the recent pivot will probably reduce the gap.

@TimothyJohnson5c16 I don’t think those are equivalent. One could imagine scenarios where automated resolution becomes possible eventually even when AI wipes out humanity. This means that even though you as a human wouldn’t survive to resolution, you’d see accurate prices in the meantime.

@Ansel ... would you? I guess maybe if AIs who expected to survive humanity's extinction started trading in the mean time.

@MugaSofer You don’t need them to be trading yet, you only need an expectation of accurate resolution. For the same reason that a market “resolves yes on Jan 1 2100” still has value today, even if no-one alive today expects to see that happen

@Ansel They don't have to be trading yet, they just have to start trading prior to the extinction of humanity. (Otherwise there'd be no-one expecting to survive for humans to sell to.)

@Ansel Like, if it was known that an asteroid was going to wipe out life on Earth in 2050, the "resolves YES in 2100" shares would be worthless, even if Manifold had Apocalypse-proofed their servers to ensure the resolution would still take place automatically.

@Ansel I think the bigger issue with @TimothyJohnson5c16's suggestion is that many people trade on extinction markets at their actual odds of extinction for non-profit-seeking reasons.

Agree on your final point. But your asteroid scenario is different from the human extinction one. If the asteriod has only a 50% chance of destroying earth in 2050, I expect that the 2100 market trades at normal-ish levels, although there may be a bigger discount due to general discount-curve-steepening by the asterioid. And I don't think you ever need to sell to an AI, because the resolution itself is indistinguishable from a final 'sale' an instant before the resolution at full price

opened a Ṁ750 NO at 87% order

Selling off some shares with a favorable limit order at 87 :) Clearly this means I think there’s a 13% chance of UFO’s being aliens

bought Ṁ2,299 YES

@benshindel I got you fam

Has @EliezerYudkowsky ever considered damage to his credibility and reputation when he made this bet?

I take him less seriously on his AI claims directly because of the ridiculous odds he took on this bet. I wouldn't hold it against him just for turning out to be wrong, but the problem is that 150:1 odds are so high that it's almost impossible for anyone to ever be so certain about anything. I would not bet 150:1 odds that my mother, who told me she was going to New Jersey for a vacation, is actually in New Jersey.

He claims to be an expert about machine learning and is also extremely confident about his other predictions involving AI destruction. But when he is willing to take such outrageous bets like this one, where this level of confidence would require political information he couldn't possibly know, it brings his AI predictions into question as well.

This bet's odds are coming down to reality - not because of loans, but because there is substantial uncertainty regarding this topic that Yudkowsky failed to consider.

3 traders bought Ṁ210 YES

@SteveSokolowski I think that most people would bet at 100:1 or greater that there will be no proof of UFOs being aliens, although most people probably wouldn't bet on anything at the scale of $100K vs $1K. The odds aren't the newsworthy thing about the bet, just the amount of money at stake.

150 to 1 isn't even super extreme odds. You can bet at sportsbooks with greater odds than that every day. The sportsbook would not offer those bets if they did not think the odds were substantially worse than 150 to 1.

@SemioticRivalry Sportsbook odds are a much more tractable question than whether UFOs have a worldview-shattering explanation. In the case of sports there are abundant prior data, a bounded set of outcomes, and repeated time-limited games in which answers are known by definition. In contrast, any disciplined exploration of the UFO topic can very quickly get down to questions of whether we correctly understand physical reality, explore hypotheticals where there is literally zero prior data one way or the other (e.g. you can’t prove ET non-existence), and where it may take arbitrary amounts of time for the truth to emerge. Intuitively for me any level of confidence above 10:1 in this area is hubris.

@Joshua Most people are wrong

@Ansel If I understand correctly, Eliezer wins the bet as long as no worldview-shattering evidence emerges within the next few years.

So even if he's wrong, what makes you think the proof will come out now?

@TimothyJohnson5c16 You're not asking me, but my answer is simple: superintelligence will infer it from the training data.

@Joshua If that is true, then why, after six months, has nobody agreed to put up their house as a bet with me?

Yudkowsky's unsecured bets are meaningless and he will probably just need to declare bankruptcy. Why will nobody give me a lien?

@SteveSokolowski Because as in my other comment "That trades off against other uses of the property as collateral so the tiny return still is not worth it"

@SteveSokolowski If you can figure out why this market and similarly unlikely events on Manifold are above 1% then you will have your answer

@Tumbles I think that Manifold users, in general, have a bias against "black swan" events. They tend to underestimate the probability of things happening just because they haven't happened before, particularly those with major consequences.

There is a mathematical proof that shows that the probability of an event occurring, which has not happened before, is 1 / (number of existing negatives + 1). I myself was surprised that this can actually be proven, but there is a proof.

So, if there have been 60 or so Presidential elections, and we knew nothing else, the odds of that market are correct - around 1/61, or 2%. Manifold users have correctly estimated this mathematical rule probably without even knowing it existed.

@SteveSokolowski I feel like that proves too much -- if the question was "Will a hamburger be elected as POTUS?" doesn't your argument above also imply that the hamburger's election odds are also at least 2%, since a hamburger so far hasn't been elected in 61 tries?

@SteveSokolowski Can you link to this “mathematical proof”? I find it very hard to believe that this applies to any and all scenarios…

Imagine if someone made a market that said “odds that the 2024 presidential election will be decided on a field of battle” or “odds that the winner of the 2024 presidential election will refuse to accept a la Caesar” or “odds that the presidential election will be decided by a rogue elector” or something… the odds to all these things are not just 1/61!!!

@A Jinx lol

@benshindel I think he's referring to a formulation like this: https://www.quora.com/An-unfair-coin-whose-weight-is-a-random-variable-uniformly-distributed-from-0-1-comes-up-heads-four-times-in-a-row-what-are-the-odds-the-fifth-flip-is-also-heads

Where you have a binary event that has happened N times, and you know nothing about it except that it has been false each time so far (and assume a uniform distribution on the original bias of the event).

But clearly we do know a lot more about the electoral chances for both RFK and the hamburger than just "it hasn't happened so far", so assuming a uniform distribution as the prior isn't reasonable.

@TimothyJohnson5c16 Agreed, and that’s why I’ve been sizing down my no bet recently as it converges closer to my credence on “public emergence” of something big before 2030, even if my credence on the actual truth is a bit higher.

@SteveSokolowski Would you be up for a bet on a shorter timeframe? 2028 is a long time to wait, and now that loans were removed I don't want to tie up significant amounts of mana for that long.

@TimothyJohnson5c16 No, and here's the reason why.

My evaluation of this subject is that the fact that non-human intelligence exists isn't some sort of groundbreaking event, or one that most people would even be surprised about. The issue is that the government has dug itself into a deeper and deeper hole over the past 80 years covering it up (and is now doing an increasingly poor job of it in the '20s, given that a former president admitted UFOs exist.)

When you have unaccountable "classified" programs that only an estimated 400 people know the full extent of, and which Congress wasn't even aware of until last year, people will do anything because nobody will find out. There's probably billions of dollars that have been siphoned off by the principals just for personal use in addition to the wasted spending on the NHI stuff. We saw what happened in Abu Gharib when you place hardened soldiers in a masculine environment with zero accountability, and I see no reason why something similar would not follow here if all the non-human stuff is true.

It's gotten to the point where admitting the obvious could cause crazy people to try to overthrow the government. That's why the government continues to produce ever-more error ridden reports and uses ever-more ridiculously transparent excuses. That's why Schumer and Rounds needed a bill creating a "disclosure board" that was allocated $11 million to write books and film pleasant movies summarizing the information in the best light possible. And even that whitewashing wasn't enough for Turner, Lockheed, and (apparently now according to the reports) Biden.

What the humans did is the real story here, not the non-human part. GPT-4 already places the odds of this being true at 80%, and that was before defense lobbyists spent a week of House time removing the "eminent domain" provision from a bill that would seize all alien spacecraft. The evidence is clear; most humans just don't want to look into it in depth or protest on the streets to get whatever is being hidden declassified.

I don't believe this market will resolve NO until a model that is superintelligent at persuasion can be deployed to convince people to take the subject seriously and conduct research to look at what is known. And that model probably won't be available until the end of the time period here.

@SteveSokolowski Aren't you concerned the superpersuasive superintelligence will be used to convince you there are no non-human intelligences and all of this is better explained by boring government incompetence, lying attention-seekers, and benadryl overdoses?

@metacontrarian No, because I believe Barack Obama more than an AI model.

@SteveSokolowski You think Barack Obama would agree with you vis a vis this market?

@benshindel YES, in his head, without question. However, he would never bet in this market nor say any more than he already has on national TV, for the reasons I said above.

I edited this comment to say that while Trump didn't admit it directly like Obama, there's a circumstantial case to be made based on some of his comments and his behavior that Trump would also bet YES, but again would never admit it.

@benshindel Strangely enough, it is indeed true that there is a 1/61 chance that the next Presidential election will be thrown by a rogue elector.

Statistics and probabilities yield strange, counterintuitive results, and humans are not intuitively equipped to deal with them.

@SteveSokolowski I disagree with this and I’d be willing to bet on a properly structured market. One can think of thousands of outlandish, distinct “possible” events that have never happened on any election, they do not all have a 1/61 probability of happening. If you list enough of them then the “probability” of one of them happening begins to approach 50 percent, assuming that they are uncorrelated. Maybe there’s a theory that says so but I suspect there’s some aspect, such as the unbounded size of the outcome space, that means it doesn’t apply

@Ansel But you can't argue with math. Search for the proof yourself - it exists. It's not a theory; there is an actual proof based on logic.

As nonsensical as it might sound, the odds of any random thing happening, when we know nothing about it, can be determined mathematically and they are higher than seems intuitive.

@SteveSokolowski Ok, then I think the condition “when we know nothing about it” is the one that’s not satisfied

@SteveSokolowski I will bet you anything you like at 1:61 odds that the next President of the United States is not a pet cat. (I don't own a house in the US, so I can't do the house thing, but I'll happily make a large cash bet with escrow.)

@MugaSofer That's not how the proof works. As I've repeated above, "the odds of any random thing happening when we know nothing about it" can be calculated mathematically.

We do have prior knowledge that the constitution prohibits Presidents of the United States from being cats, so the rule doesn't apply here.

@SteveSokolowski Ya, you keep saying there’s this “proof” and you’re welcome to link to it! I don’t think I’m just gonna take your word for it.

@SteveSokolowski There's little if anything IRL that we know nothing about. In the case of a rogue elector, or RFK Jr, we know about the current political landscape.

I'd take a bet from you at 1/61 odds that RFK Jr won't be the next POTUS (although the odds are obviously not remotely as low as that a cat won't be the next POTUS.) Can you name some other examples of things you think have 1/61 odds of happening next election based on this logic?

@benshindel I'm sorry. I forget the name of the proof and it's not my job to be the world's Google searcher. I'd suggest asking GPT-4o.

@SteveSokolowski It looks like you’re referring to the Rule of Succession (according to GPT-4o and a google search).

If so, you’re completely misinterpreting that rule. The wiki article has a great explanation as to why it’s pretty useless towards these types of problems (using the sunrise problem as an example, explained by Laplace, the inventor of the Rule himself):

@SteveSokolowski Also, it’s patently absurd to repeatedly cite an anonymous proof in conversation as evidence for an absurd conclusion, and then when asked what that proof is, to say that it’s “not your job to be a google searcher”. The “burden of proof” so to speak is FULLY on you.

@benshindel Come on, man. Who do you think you are to tell me that I have to do something for you?

Does it make you happy to "score points" or something by putting down people on the Internet? If you actually cared to learn about the proof, you would have spent your time searching for it. Instead, you spent time to write your posts demeaning me.

Everyone here is so stuck up and arrogant. Half the people here haven't even read the UAP Disclosure Act and they say they know with 99% YES certainty that every UFO and abduction story is completely insane, rather than actually talking about the topic. I'm not here to argue over the names of proofs; I want to hear logical arguments.

I want to talk with openminded people, who respect others and hope that everyone, even those who they disagree with, are successful. Good luck to you, and I wish you the best in your life.

@SteveSokolowski I actually do hope you’re successful fwiw! My enemy on Manifold is still my friend in the world at large 😆☮️

More related questions