If another statement on AI risk is released by May 31, 2023, who will be the Turing Award winner that signs it?
32
resolved May 30
Geoffrey Hinton
Chosen 33%
33%
Yoshua Bengio
Chosen 33%
33%
Martin Hellman
Chosen 33%
33%
Mira avatarGeoffrey Hinton
33%
Mira avatarYoshua Bengio
33%
Fedor avatarMartin Hellman
33%

Derivative market of https://manifold.markets/quinesweeper/will-there-be-another-wellrecognize

Will there be another well-recognized letter/statement on AI risk by May 31, 2023?
Will there be another well-recognized letter/statement on AI risk by May 31, 2023?
50% chance. Resolves YES if there is a similar letter as the Pause Letter released by the Future of Life Institute by end of May 2023. Resolves NO otherwise. We’ll call it well-recognized if it gets signed by at least 10 big public figureheads in AI, and at least one Turing award winner. It may ad…

That market requires a Turing Award winner to sign it to resolve YES. This market resolves to an equal split of every such person. Resolves "No such letter will be released" if it doesn't apply.

If a letter is published, any new entries added after publication are disregarded. The special entry "Someone not listed here." accumulates the weight due to any persons without an entry.

REGARDING HASHES: A hash of a Turing Award winner is acceptable if the name isn't already added as an entry. If somebody adds the plaintext name to the list, weight due to the person will be split across them by mana invested as of the publication time of the letter.

🏅 Top traders

#NameTotal profit
1Ṁ948
2Ṁ269
3Ṁ129
4Ṁ62
5Ṁ57
Sort by:
Mira avatar
Mira

My market description is not accurate: Martin Hellman should've been excluded unless he was one of the hashes, since people added the name after the statement was published. I see a Twitter announcement at 2:17 AM, and the entry added at 2:22 AM.

But people are trading based on the name being listed, and I have no way to resolve to a time in the past to exclude those trades. So I'm going to resolve including Martin, calculate the positions and size of the pot as of publication, and offer manalinks to anyone with a "Somebody not listed here." position which should've gotten the payout.

firstuserhere answered
22b5bb65b14af58d9cfe2d1516e953e9
firstuserhere avatar
firstuserhere (edited)

@firstuserhere my hash was 'ian goodfellow' - who did sign the letter but isnt a turing award winner i think, so, thats excluded

firstuserhere avatar
firstuserhere

@firstuserhere echo 'ian goodfellow' | md5

22b5bb65b14af58d9cfe2d1516e953e9

MayMeta avatar
MayMetabought Ṁ40 of Martin Hellman

@Mira I believe this should resolve to: Geoffrey Hinton, Yoshua Bengio, and Martin Hellman.
Based on: https://www.safe.ai/statement-on-ai-risk#signatories

MayMeta avatar
MayMeta (edited)

Posted a bunch of hashed Turing Award winners that were not yet listed but still might sign such a letter. If they will be publicly listed before this market closes, I'll sell my position in the hashed versions

firstuserhere avatar
firstuserhere (edited)

MayMeta avatar
MayMeta
Mira avatar
Mirabought Ṁ10 of 22b5bb65b14af58d9cfe...

I'll accept this hash if you later reveal the plaintext, but if somebody posts a plaintext answer that matches I'll split the share across them both.

So if there are 5 Turing Award winners, and your hash matches one, then I'll choose 6. 4 will get 20%, and 2 will get 10%.

RobertCousineau avatar
Robert Cousineausold Ṁ5 of No such letter will ...

@Mira that seems unfair to existing holder of plaintext responses - I could just make a hash of each and buy really cheap shares of them.

firstuserhere avatar
firstuserhere

@RobertCousineau my hash isn't of an existing answer

Mira avatar
Mira

@RobertCousineau It has to be a new answer to count. So if you add every Turing Award winner known to the list and have some minimum stake in each, it would prevent new hashes from being accepted.

Imuli avatar
Imuli

@firstuserhere If it's MD5 (rather than something truncated) and salted I would vote for dismissing it preemptively - it's fairly easy to generate MD5 hash collisions for salted text strings.

MayMeta avatar
MayMeta (edited)

@Imuli I've just finished reading the readme of the repo you've linked, and my conclusion is that this type of attack on an md5 hash is highly impractical: it requires the insertion of huge random-looking blocks of bytes. I say, that if the salt is reasonable, e.g. "John Smith, salt: 5af017d4", or even simpler "John Smith 5af0", then there's absolutely no reason to disqualify such a string.

If your understanding of those attacks is different - please do enlighten me.

I propose the following policy for hashes in the Free response markets:
If I think that Alice and Bob will be chosen, and Alice is already listed, I am allowed to list Hashed_Bob. When the market resolves, the creator is allowed to split the resolution equally between Alice and Revealed_Bob. But if Bob is listed by somebody else before the market resolution, I should transfer all of my shares from the Hashed_Bob to Bob, because his candidacy became public. In no case should the weight of the resolved market be equally split between the correct answer and the hashed duplicate of the same answer.

Mira avatar
Mira

@MayMeta The main risk is if they announce a letter and I'm not the first person to hear about it, someone might add an entry and bid it up to it's natural percentage before I can close the market. I don't want to pay for people to rush to copy news into my market; I do want to pay for a prediction a week in advance like firstuserhere has done. And in your proposal, the predictors lose 100% of their investment if they're not the first ones to trade on news.

I think rejecting entries added after the letter is published, and otherwise splitting them by mana invested before publication is better.

Imuli avatar
Imuli

@MayMeta I should have been more specific about the formats of salts that are easy to generate. Anything under 64 bytes long seems unlikely to be a generated collision (there are 64 byte collisions, and they can be generated shorter with the automatic padding, but I think that's not something you generally do on your own computer). My interest is more in building secure systems than breaking them, so I'm not terribly confident in that difficulty assessment.

That said, the most suspicious would be a name with a long salt after it, especially if the total comes to 128 bytes (or slightly less). The actual answer is to just use something other than SHA256 or Blake3 and truncate the hash if you want something shorter.

RobertCousineau avatar
Robert Cousineaubought Ṁ5 of No such letter will ...

To confirm: if there isn't a letter released that fulfill the criteria to resolve yes in the parent market, you will not resolve this market N/A and will instead resolve it 100% to the option titled "No such letter will be released"?

Mira avatar
Mirabought Ṁ1 of No such letter will ...

Related markets

In what year will an AI be named Time Person of the Year?
Will anyone very famous claim to have made an important life decision because an AI suggested it by the end of 2023?46%
Will an AI become TIME's Person of The Year by 2030?78%
Will AI be a Time Person of the Year in 2023?26%
Will an AI become TIME's Person of The Year by 2025?45%
Will Donald Trump claim to have a solution for the existential threat of AI (but not say what it is) by the end of 2023?20%
Before 2030, will certain prominent figures publicly adopt unequivocal stances against AI racing? (Yann LeCun)37%
Before 2030, will certain prominent figures publicly adopt unequivocal stances against AI racing? (Eric Schmidt)59%
Will a leading AI organization in the United States be the target of an anti-AI attack or protest by the end of 2023?31%
Will Yann LeCun change his mind about AI risk before 2025?15%
Will an AI system, product, company, concept, or affiliated person be TIME's Person of The Year for 2023?37%
Before 2030, will certain prominent figures publicly adopt unequivocal stances against AI racing? (Demis Hassabis)81%
Will Donald Trump propose a solution for the existential threat of AI by the end of 2023?27%
Before 2030, will certain prominent figures publicly adopt unequivocal stances against AI racing? (Sam Altman)81%
In 2023, will a major financial crime occur at the behest of an open-source A.I. algorithm?8%
Will Robin Hanson debate Eliezer Yudkowsky on AI risk in 2023?24%
Will Tyler Cowen agree that an 'actual mathematical model' for AI X-Risk has been developed by October 15, 2023?14%
Will Donald Trump propose a solution for the existential threat of AI by the end of 2023?19%
Will Donald Trump propose a solution for the existential threat of AI by the end of 2023?22%
Before 2030, will certain prominent figures publicly adopt unequivocal stances against AI racing? (Jeff Dean)48%