I wrote an essay about a speculative idea that I’m quite fascinated by, which is whether AI can be used to automate privacy. AI would be used as a privacy layer between raw sensor data and useful knowledge. Sensors like cameras, microphones, and screens would capture rich information, and an AI system would be used to redact and abstract information that shouldn’t be exposed, e.g. medical details.

The less dramatic version of what I explore here is something as simple as a livestream on Twitch: names, notifications, or user-specified redacted information. However, a more ambitious idea would be AI-mediated transparency for institutions and physical-world systems, such as hospitals, workplaces, etc.

I think this idea might be a bit jarring, but I always like to think: what if we could extend the same amount of objectivity that we use in, let’s say, a manufacturing context to design or maintain a manufacturing system, to something like an interpersonal relationship, a romantic relationship, or a friendship? This is not only the kind of aggregate thing that I think this can definitely help with, but also something that could improve privacy and make interpersonal relationships (with some mediator layer) a lot more objective while being reasonably private, which is a weird example that I just wanted to mention beyond the aggregate focus.

I’d love to know if others think it’s technically plausible, institutionally impossible, impractical, or just privacy theater.