How we will trust art made by humans
This is a story of one possible future in which AI works are indistinguishable from human ones.
You can’t tell whether an image was AI generated. If you disagree, I urge you to play a game to test that hypothesis. If you think you’re still pretty good at spotting AI media when you’re scrolling online, I am going to quietly link you the Wikipedia article on survivorship bias.
Anyways, I’m not here to talk about visually identifying AI works. Even if you’re the top 1% of AI discerners, I would like to ask you to consider what the other 99% believes.
Today, I want to discuss trust in a world where we are unable to tell computer– and human–generated works apart. If you still don’t believe this world is our present world, maybe you will consider our future world instead.
Returning to meatspace
Perhaps the only way for you specifically to know for certain that a work was human-produced is to… watch a human produce it. If you watched your friend take a photo, write a story, or draw a picture, barring sleight-of-hand and other deception, you can believe in the integrity of that media.
This is going to become the only way to verify the humanity of a work. It’s possible for this to be scaled up. Perhaps an exam proctor can verify the integrity of every one of their students’ submissions. But it is unavoidable that this future will involve human oversight, including the time it takes to watch someone produce a work.
If you watch someone make something, you might be able to vouch for that work. Others may be able to take your word for it—or if not your word alone, the word of 3+ people who all witnessed its creation. How do we know whose word to trust, though?
Trust networks
Trust in real life looks like a web. I may trust someone new if two of my closest friends can vouch for them. Depending on who I am, I may also trust someone if fifty of my acquaintances can vouch for them.
If you can’t be there in person to see the creation of a work, your only bet is to hope that someone in your network was there instead: that is, someone to attest that the work is authentic. However, in this new world, trusting someone doesn’t simply mean that you believe that they have no intention to deceive. To trust someone, you also have to make an evaluation on their ability to be deceived; the word of someone tech illiterate isn’t useful when the goal is to tell whether something was made with AI.
Unlike online webs of trust, these networks need to remain small. Crucially, they are limited to meatspace interactions—there is no way to verify the integrity of a work if you’re online. Or even worse, the agent that you could be building trust towards may be an AI chatbot.
Luckily, we have tools to make this process of trust and attestation robust.
The cryptography backing trust
In this possible world, we might start using cryptography to solidify these trusted relationships. Keyrings in protocols like PGP make keeping track of trust networks easy.
As for attestation, this might take the form of public/private key cryptography employing signatures. After I sit down and write a blog post, I would ask my two witnesses (whom still need to be physically present) to sign the digital representation of my work with their public keys, effectively declaring that they watched me produce it.
Attestation can also show up as a matter of public record. This is the sole feat I believe blockchain to be useful for. If my two witnesses post the hash of my written work on a blockchain, they’ve forever and irrevocably attested to its authenticity.
Conclusion
In my humble opinion, all possible futures of our world are ones in which AI-generated works are indistinguishable by human-generated works, whether by human inspection or automated programs. Here I presented the mechanism we will use to verify humanity in one possible future.