Hacker News Reader

Comment by repelsteeltje | Hacker News Reader

Can you explain how blockchain is saving us from AI shitting over the internet? I can think of ways how it might have, but I think that horse has long left the stable and while the crypto tools and infrastructure technically predate the LLM stuff they haven't been put to work in any meaningful way.

I think the current boom in bitcoin is rooted in plain greed, as it has been before. And outside of tokens no significant stuff is happening in cryptoland AFAIK.

View on HN

Satoshi and others spoke about this on the mailing list but one of the original ideas or use cases behind Adam Back's hashcash and other Proof-of-work methods involved or was centered around research to stop the proliferation of automated spam.

> Can you explain how blockchain is saving us from AI shitting over the internet?

One of the original use cases was the defense against Sybil Attacks: https://en.wikipedia.org/wiki/Sybil_attack To make it prohibitively expensive to attack the network.

Also: https://en.wikipedia.org/wiki/Electoral_fraud#Ballot_stuffin... https://en.wikipedia.org/wiki/Sybil_attack#Identity_validati...

Easy example: blockchain can be used to ensure that camera manufacturers implement a way to ensure that a photograph is "real" and not a deepfake/generated AI one. I think some variant of this will happen naturally in the next few years as a way to verify official journalistic images – although I'm not sure it'll be marketed as blockchain.

This of course has nothing to do with random new cryptocoins, that are indeed mostly scams.

Camera manufactures can come up with a cryptographic watermark for the pictures their camera takes or not, but blockchain really adds nothing to equation.

Would photos signed in that way need to be immutable or could they be changed? (White balance, contrast, saturation, etc.)

If they need to be immutable, isn’t it easier to use ordinary asymmetric cryptography? Given that the Intel Management Engine is still a black box to us, perhaps the private key could be obfuscated well enough in the camera that it cannot be recovered by adversaries.

If the concern is that the private key could be reverse engineered by an adversary, then how does a blockchain-based method circumvent such an attack?

I am imagining something like this:

- Camera manufacturers all agree to implement blockchain record on photos produced with their devices. This includes smartphones (iPhones, etc.) as well. They publish this to a public blockchain.

- Image editing tools like Photoshop develop a way to maintain the blockchain record of edits made and publish to the same public record. This is not substantially different from the way document verification works already.

- Social media sites use the same record to flag or otherwise mark images that are not verified.

This seems complex, I know, but if the alternative is that anyone, anywhere can create fake images at the push of a button, a verification solution like the one I've proposed will be necessary, whether it utilizes blockchain or not. It may turn out that we just end up trusting corporations to handle this, without requiring any sort of public records like blockchains.

Security issues at the camera level might be a problem, but realistically the vast majority of camera devices (including phones) are made by a handful of companies and I don't foresee this being that much of an issue.

Much more likely that if camera manufacturers decided to do this, they would each offer a their own proprietary and incompatible image verification scheme, forcing implementations to integrate with many different services. Eventually recognizing the problem, they would form a Trusted Imagery Working Group™ to develop a patent-encumbered chip that would need to be licensed for use (available only to invited member organizations), and present in all devices that will create or validate image signatures. Keys from that chip will eventually be extracted, rendering it useless, but industry will be widely reliant on it, with too much inertia to change.

Very true and likely, but it doesn't change the fact that blockchain would be a potential solution to this problem.

A blockchain can't verify official journalistic images. Someone has to do the verifying, and that person can still lie to the blockchain. It's the same situation as NFTs.

Yes, and that someone would be the collection of camera manufacturers that agree to implement a blockchain record on all photographs and then publish that to a public blockchain.

So then what's the point of the blockchain if you're relying on exclusive attestation by camera manufacturers? They could just use plain old asymmetric encryption to sign the images. Has the bonus of not needing an internet connection.

You would still need a way for image editing programs to allow for edits in a way that doesn't compromise the validation.

Yes, but while a blockchain could be useful there, the very act of editing defeats the point of the signature. Eg editing someone into or out of the photo. Validation by requiring the original unedited image to also be accessible doesn't work either since often the edits have a valid journalistic reason (eg to preserve a victim's anonymity). Allowing exceptions will just result in those exceptions dominating, just like how "anonymous sources tell us" has become a popular way to write clickbaiting hit pieces without having to actually back anything up with evidence.

Personally I really don't buy the idea that we can technologically eliminate something as vague as fake images. The solution is a less gullible public, and an acceptance that the short interval where communication was easy and forgery was comparatively difficult, was an exceptional circumstance. We're just returning back to the norm.

I’m guessing they’re taking a hand wavey position that as AI spews forth it degrades provenance as you start to struggle to tell things apart.

Immutable “crypto” ledgers offer the promise of restoring confidence in provenance.

Reply on news.ycombinator.com