Meta tests ‘human concierge’ for Muse AI assistant amid privacy concerns and mixes human effort with tech promises
AI calling feature
Muse is designed to autonomously handle tasks like sending emails and booking travel. Users can instruct Muse to call U.S. businesses to perform errands such as booking appointments or obtaining quotes.
Concerns about privacy have arisen, with some employees worried about sensitive data being shared during these calls. Internal discussions revealed apprehensions about the potential for information leaks due to human involvement in handling calls.
One employee cited a troubling incident where a contractor made a racist comment during a negotiation call.
A Meta executive acknowledged that not properly informing users about contractor involvement was a mistake and mentioned the feature has been paused.
The testing showed that human calls could lead to a success rate of 95% to 98%, far higher than AI calling. An apology was issued to the employee who reported the racial comment, ensuring the contractor would no longer work with Meta.
Human intervention insights
The Muse project highlights how tech firms often depend on human labour to fulfil AI tasks amidst pushback against robocalls.
Historical context indicates similar reliance, as seen in a past initiative involving Facebook’s digital assistant, “M,” where human effort accounted for a majority of tasks.
A Meta spokesperson stated feedback from employees has been largely positive, focusing on enhancing safety and privacy features.
Despite challenges, Muse has garnered over 2.5 million downloads. Meta’s stock has increased significantly since the launch, reflecting positive market reactions.
Concerns about the human concierge feature compromising security have been expressed, as initial safety claims could be undermined.
Meta tested this calling capability with employees before its broader rollout along with user feedback mechanisms for opting out.
