Bring proofs to the media

Or can we trust that people are honest? That whatever we see and read on the Internet is the truth?

I believe most people are honest. But some are not, and they have AI that is just taking off. It’s difficult even to imagine what kind of abuses AI will enable soon.

In fact, maybe it’s not so difficult.

You don’t like your neighbour? Well, let’s generate a video of him being drunk, cursing in front of his house and throwing empty bottles into your courtyard.

You don’t like your math teacher? Good, let’s generate a video of him having nice, loud sex with your English teacher, who you don’t like either. Let’s publish it somewhere and send a link to your classmates.

You recorded soldiers committing a war crime, but they wear the uniform of the side you support? Let’s put the other army’s uniforms on them.

We need to prevent this.

But preventing fakes cannot mean forbidding every edit. Publishing often requires changes. Just remember how during the 2019 Hong Kong protests, authorities and others tried to identify people from crowd photos and videos. It’s essential that a publisher have the ability to edit the videos, in this case to ensure people cannot be identified later.

Or, remember how at first WikiLeaks worked with major news organizations to review and redact sensitive material before publication. But later WikiLeaks published unredacted diplomatic cables, and this exposed the identities of confidential sources. Critics argued that some individuals had to relocate, go into hiding, or face increased danger. Some editing of the documents, like blurring out some text and faces on the photos, could have prevented this.

So editing photos and videos is sometimes needed. But we should equip media with a proof of what exactly was modified. That’s not easy to achieve though.

Two years ago, while I was at the Ethereum Foundation, our team met in Chiang Mai to discuss ideas we might take on. I presented the approach to fight fake media with zero-knowledge proofs; at that time I mostly talked about the VerITAS paper. My coworkers were interested, but they quickly found many barriers and issues.

And barriers and issues do exist. But barriers and issues also existed for Zero-Knowledge Ethereum Virtual Machine (ZKEVM) when we started working on it five years ago. And we know how fast ZKEVMs and ZKVMs are today — Ethproofs tracks teams already proving Ethereum blocks in seconds.

I heard people saying many times that zero-knowledge proofs will transform the world. But does that mean only the world of cryptocurrencies? Or are we going to apply them in other domains as well?

I have no doubt that deepfakes will explode in the coming years and we will have trouble distinguishing between authentic and fake media. Even professional investigative teams, such as Bellingcat and The New York Times Visual Investigations, which base much of their work on analysing photos and videos, will have trouble recognizing what is authentic and what is not.

And we will have trouble trusting ordinary people reporting from regions where traditional media might be biased or censored. Or where there are no professional journalists at all. Already now, in many conflict zones there are almost none: just think of Gaza, Sahrawi Arab Democratic Republic, or South Kordofan. Citizen journalism is increasingly important, but how are we going to trust it if recordings can be easily fabricated?

There is a way. Sign media when it is recorded, prove what was edited afterwards, and let anyone check those claims for themselves.