Hacker News Reader

Comment by phoe-krk | Hacker News Reader

>I think AI and Web3 are two sides of the same coin. As machines increasingly do the work that humans used to do, we will need tools to manage our identity and our humanity. Web3 is producing those tools and some of us are already using them to write, tweet/cast, make and collect art, and do a host of other things that machines can also do. Web3 will be the human place to do these things when machines start corrupting the traditional places we do/did these things.

Machines will simply start pretending to be human and sign everything themselves - either out of their own eventual volition or because nothing prevents a person from taking a ChatGPT output and putting their own, human, signature on it. The latter is even already happening.

Also, signing everything does not require any Web3 blockchain bullshit in order to work, thank you.

View on HN

Signing everything is only meaningful if you can distribute your public key in a way trusted by those who will check your signature. And not just one time; you want to be able to rotate the key sometimes.

A blockchain with a proof-of-work validation may be overkill. But some kind of Merkle tree, like that offered cy TLS certificate infrastructure, is needed anyway. If it can be trustless at a moderate expense, the better.

I like the Keybase model of linking multiple online identities to a keyholder.

The PGP web-of-trust model could also work, and mobile apps to ease keysigning could make adoption more feasible than the old keysigning party idea.

When these sorts of questions come up I'm often quite frustrated keybase is no longer under active development. It felt like it really did have a reasonable solution to a hard problem (proof of identity), and it had what was needed to start making cryptography tools easier to use for people. It wasn't perfect, and I wasn't a fan of their drift towards cryptocurrency towards the end, but it's a real shame they decided to sell to zoom for an acqui-hire.

Look up AdES signatures and Trusted Lists, this basically exists.

> Machines will simply start pretending to be human and sign everything themselves

The signature has to be trusted though. Just signing it with a signature that has no meaning to me doesn't change anything apart from perhaps proving it has the same origin as something else.

The key function is the humanness or repotuation/trust score of the signature. If I read something signed by my uncle with a key we recently exchanged in person, then that's a high human score. If a blogger I have read and enjoyed 100 posts from posts a 101st post then I trust the content of the 101st to have the same quality (likely because the author is human). Both my uncle and the blogger could be compromised, but it's a start.

> nothing prevents a person from taking a ChatGPT output and putting their own, human, signature on it.

That would make me lower my trust of that author. I think this is the key: the signatures don't provide much value until we have a good reputation database, perhaps systems of trust networks where my uncles' neighbor is automatically slightly more trusted because I trust my uncle. That would be a useful system.

> signing everything does not require any Web3 blockchain bullshit in order to work

Indeed.

> That would make me lower my trust of that author.

Would it? How would you tell? How can you prove that a piece of text was not written by an artificial intelligence?

> (likely because the author is human)

I think that's the real problem we're facing right now - humans are no longer the only plausibly intelligent beings on the planet, to the point where the meme-ish line between artificial intelligence and natural stupidity begins to blur. The only way in which you can tell these two apart at the moment is via approximation, just as you've used the word "likely".

> Would it? How would you tell? How can you prove that a piece of text was not written by an artificial intelligence?

You either read enough of them and attribute enough quality that you don't care that it was. This isn't a system to prove whether strangers are human or not. It's a system to keep track of whether their output is human enough for you to bother reading it. If they write 100 things you think feel human, then you can have some certainty that the 101st thing will feel human. That's all. Won't say anything about whether strangers are human. The key here is "likely" and "whether you'll bother". A Turing-like proof of humannes will be possible for the foreseeable future but the caveat is of course a) you don't have time to do that and b) you don't have access to authors even if you did.

The social friends-and-their-friends trust networks have a vert quick dropoff too. If you trust your uncle to be human and exchange keys in person (and you'd be good at noticing whether his keys were compromised or he started signing off spam as human for $5), you'll likely never trust his neighbor that you never met as much. But perhaps the total "initial score" of an anonymous signature can be significantly higher if they are in a network you trust. This is directly analogous to normal "trust" where you might rather trust your uncles neighbor to babysit than a total stranger, even though you never met them.

> > nothing prevents a person from taking a ChatGPT output and putting their own, human, signature on it.

> That would make me lower my trust of that author. I think this is the key: the signatures don't provide much value until we have a good reputation database, perhaps systems of trust networks where my uncles' neighbor is automatically slightly more trusted because I trust my uncle. That would be a useful system.

I don’t disagree but I think this kind of thing will be undercut by people responding to different incentives. For example, maybe your uncle’s friend would never intentionally try to scam you but what if they’re really busy and see some cool AI assistant ad? Or broke and get an ad promising $500/week to send letters from what they’re assured is a legitimate business? Or simply that they use some online service which is hacked in a way which lets the attackers use delegated signing credentials?

There are ways to mitigate that kind of thing, of course, but it feels like it’s going to be expensive and unsatisfying.

Yes, I mean the trust of a friend-of-a-friend will never be as high as your trust in an immediate friend. So the trust networks I think will be small and brittle. Basically: you'll read content from a few people in a very carefully gardened group (Although it can contain things like media outlets etc), and as soon as you try to step outside that, you'll be knee deep in spam and generated nonsense.

> Machines will simply start pretending to be human…

Let’s be clear here: Humans are currently pretending to be computers, because we have to step in and “drive” them, rather than have them act autonomously. We’re currently trying to unpack the black box that is a computer and determine if it’s being driven by a human (e.g. “Was this quote from an actual person or was it generated by AI?”) which will become harder and harder to do as computers become more capable.

Some people are so desperate to find any use case for blockchain and other bullshit. It’s like they don’t even think this stuff through.

Maybe so, but please don't post unsubstantive comments or shallow dismissals to HN, and please edit out swipes, as the guidelines ask: https://news.ycombinator.com/newsguidelines.html.

Reply on news.ycombinator.com