Hacker News Reader

Comment by __MatrixMan__ | Hacker News Reader

I can make something and produce a digital signature for it. AI can't fake the signature. We meet in real life and exchange keys. You pass my key to your browser so when it encounters content with my signature it marks it as having come from somebody whose humanity you have verified. We build a network of such relationships. Maybe it's enough to just see an annotation, a sort of glow around the content that has verified human signatures. Or maybe we go full dark-forest and configure the browsers to only render content that has a trusted signature.

View on HN

The problem is that the majority of the time, the one ultimately trying to dupe you is not the AI model, but the person who handles it.

In other words, signatures don't address the fact that a human can ask an AI model for content and then sign it to pass it as their own.

Isn't that an enormous risk though? As soon as such abuse becomes known, that individual will never be trusted again.

unfortunately this would lead to more innocents getting coerced into various schemes, or simply their signature stolen.

https://en.wikipedia.org/wiki/There%27s_a_sucker_born_every_...

https://xkcd.com/538/

That makes more sense for decryption keys than it does for signing keys.

For any of this to work we have to already be gossiping metadata about each other's content. "I think this key was compromised on Nov 1, 2026" is exactly that kind of metadata. That sort of thing rarely happened in PGP because key revocation was necessarily a side channel that you were often ignoring. But if the only way to avoid being inundated by slop is to assemble the page from the gossip of trusted peers, then the untrusted nature of the recently compromised key is gonna show up right next to its dubiously trustworthy signatures, and your browser can then interpret it without you having to do any sleuthing on your own.

How hard are you really gonna work to compromise a key that you can only use for a few days? I mean, the going rate for placing an ad on a page is like, fractions of a cent per view. Maybe it'll be more valuable in a post-ad world, but I have a hard time imagining that getting your content in front of some eyeballs for a day or two will ever be valuable enough to be worth hitting somebody with a wrench over. What you have to gain by doing so just isn't valuable enough to be worth the risk.

Maybe for high-stakes things, serious journalism or closer interactions, but most of the time in the Internet we just consume content by unknown people and we don't stop to check their trustworthiness. In the OP's context (game wikis) you're probably not going to know the people signing the wiki's content.

You probably know somebody who knows somebody who knows them, we're just not presently engineering for that fact to matter.

The most famous attempt at a web of trust was PGP, but it just attempted to bind a public key to a name. That's too generic. Only privacy nerds are actually going to have a key signing party if that's all they get out of it.

I think that it'll go better if the we're attesting about common interests. Like, go to the event because you're enthusiastic about games, end up with some public keys as a side effect.

It'll be hard if you're the only person you know with a particular interest and it's impractical to travel to wherever people with that interest gather. Then you'll have to take a gamble on some internet handle you've never met. But really that's just falling back to the dubiously trustworthy web we currently have.

Right, for that we need different domains of trust. Like maybe my commentary on current events is trash but I can be trusted to give good plumbing advice. So people who know this about me can trust some of my content but not the rest, which causes it to be seen/hidden by the people who only trust me transitively through them.

The architects of the original web gave up on the trustworthiness problem pretty early. All we got is : "is this server lying about its name?". We're going to have to pick that problem back up and bake solutions to it into our browsers. Except instead of attestations about server names, I'll be about the content. Was it made by somebody who is known to be sloppy with AI? Or maybe they're the sort that would put out content that doesn't disclose that its secretly an ad.

People have been gossiping this sort of stuff about each other for a very long time. If were gonna have a useful web again, we'll need to find a way to scale that capability.

[bell ring] What is PGP?

Reply on news.ycombinator.com