If you ask Meta's new AI agent, Muse AI, to call a hair salon, the person on the other end of the line might not be an AI. According to an exclusive report from Reuters, citing internal company postings, Meta has been testing a 'human concierge' for Muse, with contractors handling some of the calls made through the agent.
See also: Meta Muse: Amazon blocks AI shopping agent
Meta informed its staff about the trial, having previously launched the Muse calling feature. For half of Meta's staff, the "human agent calls" option was enabled, with an opt-out option for those who did not wish to participate. This move demonstrates Meta's intention to examine user reactions and collect data on the effectiveness and safety of this new feature.
“Muse is now capable of handing off requests to a trained agent, who places the call and completes it,” the statement said, as reported by Reuters. Adding a human factor to calls can improve the accuracy and efficiency of the services offered by Muse AI, especially in cases where the AI may not be able to fully understand the user’s needs or the specifics of a conversation.
However, the posts showed that some employees had warned that sensitive information could end up with call center contractors, even though no one intended it to. This concern is especially important in an era where privacy and personal data protection are critical for users and businesses. Meta, which has faced criticism over privacy issues in the past, appears to be taking these concerns seriously.
See also: Meta invests in AI agent Muse to make up for lost ground in the artificial intelligence race
Daniel Roberts, a Meta spokesman, told Reuters that employee feedback was “overwhelmingly positive” and that the purpose of the test was to help shape security and privacy protections ahead of a public launch. He added that Meta will “only release it when it’s ready and with appropriate disclosures,” though he did not specify what those disclosures would be or who would receive them. The statement suggests that Meta is being cautious in its approach and wants to ensure that its product is safe and reliable for the public.
Muse launched in the United States on September 8 as an agent that sends emails, makes purchases and books trips on behalf of the user. The calling feature forwards requests to businesses over the phone so they can book haircuts, check inventory or get quotes, and then provides transcription. The app has reached 2.5 million downloads, according to Sensor Tower, indicating its growing popularity and acceptance among the public.
At launch, Meta placed a strong emphasis on privacy, stating that each agent runs in its own virtual machine. While these statements refer to where the software is running, they say nothing about a person using the headset. Meta appears to recognize the need for transparency and security as users become increasingly privacy-conscious about their data.
The company has been in this situation before. Back in 2015, Facebook began testing M, an assistant in Messenger that was supported by human “trainers.” It remained in limited beta and was shut down in 2018, never achieving more than about 30% automation. That past experience could provide valuable lessons for Meta as it tries to improve Muse AI and make it more efficient and reliable.
See also: Meta patches Muse zero-day vulnerability
Meta offers Muse as an agent who works on your behalf 24/7 and has been assisted by someone on some calls. Adding a human element can enhance the user experience, providing a more personal and responsive service. However, the success of this approach will depend on Meta’s ability to effectively manage privacy and security concerns while continuing to innovate and improve its services.
