Don't use zero-knowledge proofs to fight fake news
This is an article I wrote in Feb 2023 when this was a hot topic, but I never published it. The initiative died, probably because it doesn’t make sense.
“Blockchain solves this”. It’s easy to jump onto new technologies and vastly overestimate the space they can be applied in. This makes sense. You find something new, so you throw it at everything and see what sticks. It’s famously hard to predict use cases. Microwaves were not intended for food, nor was dynamite intended for war. However, misguided research is doubly painful: it has a direct cost, and it detracts talent from more useful things. I would like to convince you that using zero-knowledge proofs in a fight against fake news is misguided.
I will do a case study on applying zero-knowledge proofs to solving a common fake news technique. A large set of fake news stories rely on photographic pseudo-evidence. These pictures were usually taken at a different time or place, or perhaps the photo was simply manipulated completely.
However, what if we could use mathematical proofs for preventing these kinds of manipulations? Imagine a stamp on every picture that proves that the photograph or its metadata (such as its location and timestamp) have not been tampered with.
This is relatively easy to do. Just sign the picture with a private key that is built into the camera, in some private tamper-free way. Private keys are stored securely on chips on phones and laptops already, so why not add cameras? In fact, this is what Sony proposed.
So far, so simple. However, as Trisha Datta and Dan Boneh noticed, it’s far from enough to help journalism. They note that raw photographs always get manipulated before they get printed. They need to be compressed from their 100MB RAW files into 200KB JPGs, they need to be cropped, contrast-adjusted etc. All of this would break the cryptographic link between the file and its signature. After all, the signature cannot tell whether you have used Photoshop to compress the file or to change the meaning or metadata.
That is why they propose to use ZK proofs to trace these transformations and ensure only they have been executed. For instance, JPG compression is algorithmic, so you could prove with a simple signature that only that permitted algorithm has been used on that image, and nothing else. It has been shown that these proofs take seconds, and verification of these proofs takes milliseconds, relying on ZKP’s succintness. Sounds like a great solution then!
I want to convince you that it is not. For several reasons:
It’s impossible to come up with a list of permitted transformation. For instance, cropping can be used manipulatively, whilst retouching can be legitimate.
Fake news is spread through social media and undergoes many media-specific and user-specific transformations. This happens through file conversions, screenshots, even taking pictures of screens.
It is not plausible that all social media tools will adapt the new standard. Such a standard would have to have a high critical mass of adoption.
However, I propose an easier solution: a searchable database with the original pictures and the proofs that metadata has not been tampered with. This approach relies on human intelligence using such a database as an additional information point for debunking or verifying stories.
This way, the standard can grow organically. It would start with cases of high importance, where the photographer is worried they might not be believed. For instance, as additional evidence in courts or documentation of high-profile cases, such as war crimes. Then, slowly, it could become more widely used.
One last worry with such proofs is that they might give viewers a false sense of security. After all, many fake news rely not on the picture but of the worded story describing the picture, or on showing only one part of the story.
Unfortunately, fake news are hacking the brain, not the data or cryptography.