Think Like a Village

Today I am doing a poster presentation at AI: Hype, Hope, and Humanity with my PhD candidate Guangwei Teng.¹ This is our 3 minute pitch.

The Withdrawal of Visual Warrant: Deepfakes and the Fragility of Social Legibility

GT: We used to take photographs as evidence of what really happened. In everyday life, we could reasonably rely on them without first checking their origin. This is what we mean by social legibility. However, AI-generated deepfakes undermine that default trust.

Can technology give us that trust back? It can help us check a particular image. The difficulty is that proof of origin does not necessarily travel with the picture. For example, a screenshot can keep the picture but leave behind the original file’s proof of origin.

Regulation can require images to carry that proof. But this does not restore social legibility because checking the proof is an extra epistemic step that needs to be done for each image.

So we cannot restore everyday trust through either technology or regulation alone. If we can no longer rely simply on how an image looks, where should we place our trust?

TS: The social legibility of realistic images allowed people to truth their eyes. We need to shift everyday trust to sources.

Who vouches for the images? If we can trust the voucher for the truth of the image, then we can use the image as evidence of what the thing that happened looked like.

The mechanism of image production becomes irrelevant. Court sketches, strict photojournalism and carefully selected AI images are on a par informationally when vouched for by a trusted source.

Vouching is not transitive, but warrant transmits across chains of vouching.

We shouldn't think like a state and assume trusted sources have to be institutions (or an impossible rules-based order). That can be self-defeating. They can also be informal social structures and respected individuals. Ask: who vouches for this? Who will take the sanction if it isn't true?

Think like a village.