I think you’d like a blog post I wrote in November where I put forward an outline of what such a system would look like:

https://lukas.dev/posts/how-to-trust-again/

Digital signatures, media verification, authenticity and more are all covered!

AI researchers and companies need to starting asking themselves: just because we can does it mean we should?

The societal implications and risks of this technology are enormous. This could lead to an erosion of publc trust in audio and video evidence. Imagine someone creating something that gives dangerous medical advice using Fauci's likeness? Or giving the worst politicians an excuse to say that something they actually said/did was just AI generated.

In the life sciences there are limits to what researchers are allowed to do. Especially with genomics.

There are great opportunities for AI to do good things for humanity. We should not waste the time and effort on trivial things like this that have negative net benefits to us.

Edit: spelling

While I agree with you, this is wishful thinking. We need regulations before things get out of hand.

Yep. And that time is right now. Things are already getting out of hand, very quickly. Can we stop the ride? I'd like to get off.

> In the life sciences there are limits to what researchers are allowed to do.

And yet we're sitting here questioning whether covid was man made and there's serious speculation human genome experimentation is active in some parts of the world...

> AI researchers and companies need to starting asking themselves: just because we can does it mean we should?

This just results in the researchers/countries who take the moral high road being left behind thus giving those who continue to do research a competitive advantage.

I think the USA is far and away the most reckless here, and has the most funding. China is not having letting their AI labs release all the new models publicly, and in fact, our research is helping them develop it faster, since US companies are open sourcing everything. (Hey, let's develop tech more dangerous than nukes and share it!)

There is no arms race, we're all in a car driving straight toward a cliff, with everyone fighting to press the gas pedal harder. We need to stop, hard.

It's kinda way too late. The ramifications for society are going to be massive and probably pretty bad overall, not even counting the existential risk which is what I think is really going to do us in. I think there may be a path to not everyone dying, but it depends on the right people in government getting the message and taking drastic action a-la Yudkowsky's plan https://www.lesswrong.com/posts/oM9pEezyCb4dCsuKq/pausing-ai...

This exists, and it's a joint effort across a lot of tech companies: https://en.wikipedia.org/wiki/C2PA

As long as it is not owned by a centralized authority (government/corporation/etc) and is open source, I would be game. But this does not look like that.

I'd vouch for this.

Much like you can sign PDFs with your private key certificate[0], you should be able to sign any kind of file, including audio.

Even if my voice was AI-generated, I could endorse it by signing it.

All we need are CA's jumping into this bandwagon.

--

[0]: https://www.digicert.com/kb/document-signing/how-to-sign-a-p...

You mean like China's social credit system? Sounds dystopian

Ideally not centralized, and only for signing creative digital content made for distribution.

Not for signing (and potential blocking/tracking) financial transactions.

Power has a tendency to coagulate.

There are few to no paths to a non-dystopian outcome currently.

> There are few to no paths to a non-dystopian outcome currently.

Said Caveman 1 to Caveman 2 shortly after the discovery of fire. ;-)

Not exactly sure what you're getting at, but the idea that we can some kind of privacy when the capability to develop near omniscient technology exists seems like a great path for dystopia.

Why wouldn't the solution be just check the canonical source for the content?

IE go to joe rogan's spotify page and see if the content is linked there.

because mis/disinformation is the real problem here. We will be seeing a lot in the media and when 90% of consumers don't bother to check the source, it _will_ skew a lot of things.

I don't think the top concern is verifying the authenticity of Joe Rogan podcasts. This is just an example.

bitcoin fixes this.