Meta’s Muse AI Agent Raises Questions About Shopping, Privacy and Human Autonomy

“I see this as a continuation of the company’s main business model of controlling what people see, putting ads in front of them, and collecting more data,” Mir says.

Could AI Shopping Agents Put Advertising Before Consumers?

It’s natural to feel anxious about handing over your credit card to an AI agent. Even if users trust these tools not to spend their money in ways they would not approve, outsourcing shopping decisions could have unintended consequences.

Consumer advocates believe purchasing automation could allow companies to prioritize certain brands or products in an agent’s recommendations because of sponsorship deals. Alexandr Wang, chief AI officer at Meta, suggested that the company is exploring additional revenue opportunities for Muse beyond advertising in an interview with Axios. He declined to provide further details.

Personally, I love the slow scroll of online shopping. You could spend an hour on a website and leave without adding anything to your cart. There’s something fickle and inefficient about searching Depop for a leather jacket you probably won’t buy.

By exploring the internet on my own, I am slowly building my own tastes and gaining a better understanding of what I actually enjoy. If Muse were to do everything for me, I would be lost while trying to reach those decisions.

“At some point, AI will be making meaningful decisions about taste, preference decisions, where to go, where to eat, what to do, what kind of vacation to take, etc.,” Schroeder says. “It’s kind of the joy of being human to be able to try things and find what you like, so the concept of handing it all over to a machine is pretty frightening to me.”

How AI Agents Could Make Users More Passive

Relying on agents like Meta’s Muse could reduce our ability to function independently while these tools collect more data about us.

“The design of the AI agent is such that it disengages the user,” says researcher and lead ethics scientist Margaret Mitchell. She is affiliated with Hugging Face. “So the user becomes more passive.”

Mitchell recently co-authored a research paper arguing that agent tools are not designed with effective human oversight in mind. The paper suggests that these tools can worsen a user’s ability to assess what an agent is doing on their behalf.

People who use agents may become more reliant on them than they would on a standard chatbot because agents have access to more information and can perform more tasks. One potential impact is “cognitive decline,” in which users rely on automated external sources of confidence and become less able to make independent decisions.

Personalization, Trust and Data Collection

Mitchell also believes that advanced personalization could encourage users to share more data about themselves. “Not only do they know you, they trust you more,” she says. “It actually matches you in what it talks about and how it talks about it, drawing you in more and interacting with it and divulging more personal information.”

Recent research suggests that AI tools can mirror the way users speak as they collect more interaction data.

Why I Deleted Meta’s Muse

Muse is powerful, but I decided to send this data goblin back to the cave and deleted the app after the first experiment.

The agent sent a final notification before Muse was removed from my phone: “Connect the app and you’ll be able to do even more.” Okay, buddy. I had heard enough.