AI and Consent

I didn’t expect to be blogging about AI while in Miami¹, but here we are. We were eating dinner in the Hard Rock Cafe when a woman came over and asked if she could take our photo. We agreed, as we knew the drill: she would come back and try to sell it to us in a frame, a memento of the trip. What I didn’t expect was for her to return with images that had been run through generative AI to place us in a rock band and similar scenes. If we had known our image would be used this way, would we have agreed?

We had no option to opt out of the AI processing. Worse, has she signed those holiday snaps up to be used in future training runs? Would she even know if I asked? The T&Cs for these kinds of system are particularly murky. In this case I have potentially agreed to terms, without ever being asked, from:

• The photographer²
• The image tool the photographer uses
• The model that image tool uses

We are seeing this lack of transparency and consent across multiple fields. This post describes an academic who submitted papers that were then reviewed by AI. The frustration is fair: novel, as-yet-unpublished research was fed into a model that could train on it, potentially letting others query that research from future iterations of the model before the original is even published.

In the UK we have started to see laws appear in very niche cases, the big one being the banning of nudification apps. This follows earlier laws stating they could not be used on photos of an individual without consent. While not directly about AI, the organisation Chayn argues that targeting nudes alone is not enough. Context matters: sharing a photo of someone “exposing her bare shoulders and wearing Western clothing”³ can be equally damaging in cultures that require women to cover up. It is not a leap to see AI used to build a false narrative, in the same way the images in that story were cropped to create one.

AI-enabled smart glasses make the whole problem worse. At least I knew a camera was pointed at me! There have already been stories of owners surprised at how their footage was being used, with human annotators reviewing extremely private footage. Wired recently uncovered that Meta has been adding biometric identification code for its smart glasses to the app, while publicly saying it is still a feature under consideration. Consent is hard when there is so little transparency about what AI can do and how it decides. France has considerably stronger laws for minors: explicit consent must be given before the image of a minor is taken⁴. The problem is that AI glasses could only realise they had broken this law after taking and processing the photo. How the glasses and the law will settle in France is yet to be seen, but I hope it lands on the side of privacy and consent.

AI moves fast, and legislation struggles to keep up. Hopefully, when it catches up, it lands on the side of openness and transparency for civilians, not the AI labs.

Maybe I will try to make a Camulos badge that stops generative AI from working, something like the glasses that prevent facial recognition. Reach out if you would be interested in helping build such a thing, or just want to give it a go.

Footnotes

1. I started this post in April and never got around to finishing it. It is still relevant, so I thought I would get it done. ↩

2. We were asked about having the photo taken, but not about the post-processing. ↩

3. Quoted directly from the BBC News article linked above. ↩

4. This consent rests with the parent or guardian until the age of 15. ↩