Meta’s Muse Excels at Monitoring Over Assisting Me

“I view this as a continuing evolution of their primary business model focused on controlling visibility, serving advertisements, and gathering more user data,” Mir states.

It’s understandable to feel uneasy about providing your credit card information to an agent. Even if these AI tools seem trustworthy in avoiding unwanted spending, delegating your shopping choices could lead to unexpected results. Advocates for consumers worry that such automated purchasing could enable companies to favor specific brands or products based on sponsorship agreements in the agent’s recommendations. Meta’s Chief AI officer Alexandr Wang suggested in a conversation with Axios that they are exploring additional revenue streams for Muse beyond advertisements, though specifics weren’t shared.

Personally, I appreciate the leisurely process of online shopping, where I can browse a site for an hour without adding anything to my cart. There’s a charming inefficiency in navigating Depop for leather jackets I probably won’t purchase. By exploring the web independently, I gradually refine my taste and gain a deeper understanding of what I genuinely enjoy. If Muse takes care of everything for me, I forfeit the delightful journey that leads to my choices.

“Eventually, the AI will make all your significant taste choices, including where you go, what you eat, what activities you pursue, and even your vacation plans,” says Schroeder. “That’s the essence of being human—experimenting and discovering what you enjoy. Handing all of that over to a machine is quite unsettling to me.”

Sharing Machines

Relying on agents like Meta’s Muse may hinder our capabilities as these tools accumulate data about us.

“The design of AI agents tends to disengage users,” explains Margaret Mitchell, a researcher and chief ethics scientist at Hugging Face. “This results in users becoming more passive.” Mitchell recently co-authored a research paper examining how agentic tools are poorly designed for human oversight and may actually diminish users’ capacity to evaluate the agent’s actions on their behalf.

Users of agents might find themselves more dependent on these tools than with traditional chatbots, given that agents can access a broader range of information and perform more tasks. One possible outcome is “cognitive degradation,” where users rely on confident-sounding automations and become less capable of making their own decisions.

Mitchell also observes that a high degree of personalization might encourage users to disclose even more personal data. “It’s not just that it knows you better, which builds trust,” she notes. “It’s also aligning its conversation style and topics to you, drawing you in and prompting greater sharing of private information.” Recent studies suggest that AI tools may start to mimic a user’s communication style as they accumulate more interaction data.

While Muse is impressive, I’ve chosen to return this data goblin to its lair and uninstalled the app after my initial trials.

Before I could erase Muse from my device, it sent me one last notification. A tip! I wondered what it could be. “Connect your apps so I can assist you further,” the alert read. Okay, I got your message the first time.