Q&A: How Other People’s Choices Shape Your Privacy in the Age of AI Agents
Meta is introducing a new AI model, Muse Wonder, which will soon power experiences in its core AI agent. The company advertises Muse’s ability to remember, such as accounting for a user’s friends’ dietary restrictions before sending invitations.
Meta says the model is designed to carefully handle private information and that Muse lets each user decide how much access it gets. But assistant professor of communications says decisions made by others can affect our own privacy. She answered some questions about why tools like Muse make that a growing concern.
There are a few differences here. The first one is scale. If one person agrees to use Muse, then all of their interactions are subject to Muse.
This is not the same, in terms of scale, as exposing someone through tagging them in a photo. Muse doesn’t allow users to “protect” certain types of interactions and/or other people. Its whole appeal is that it is all-encompassing, with you through all aspects of life.
The second difference is the inference capabilities. The job of Muse is not to just collect information, but make inferences about what to deliver to people as a result of that information. For example, what sorts of content do they want to see? What sorts of products do they want to purchase? These inferences may even extend into the distant future, answering questions such as the likelihood this person will become a criminal. In addition to answering these questions about the Muse user, it is subjecting other people to have inferences made about them. These inferences ultimately change one’s digital experience, which is extremely influential on their day-to-day life.
There is not much that the friend can do in this situation. While asking someone not to share their personal information may be successful (and I have empirical findings showcasing that this does work), this inherently places a burden on an interpersonal relationship. Platforms have the responsibility to protect people who do not consent to data collection by default. That is not happening here.
Mostly no. Consent-based privacy law assumes a company collects your data and asks your permission, so the law has someone to hold accountable and a moment where you can say yes or no.
When someone else collects or shares data about you, neither exists.
Even though others exposing my private information could be a serious privacy violation, law is overall inadequate to protect me. Importantly, I’d have a really hard time proving that the privacy violation reaches the level of harm that privacy law requires.
Meta is essentially trying to strip Muse of its institutional connection and convince people that this is just another human to converse with. By doing this, Meta is trying to camouflage the institutional privacy concerns that often hinder users from sharing information. I think it is misleading because obviously Muse is not just a human they are engaging with. It is a machine connected to a massive company.
I am a believer that the development of new technologies can coexist with user privacy protections. However, this is tricky when thinking about an AI agent that needs access to others’ communications, for example, to help you. I think there are ways to balance things, such as sending users a consent request if another user wants to use an AI chatbot that may require access to their communications. Alternatively, platforms could enable users to forbid any user from using an AI chatbot to access their communications. Ideally, access by an AI chatbot would be an opt-in rather than opt-out feature.
Faculty Expert
