New Delhi: On Tuesday, Meta unveiled its new AI tool called the AI Assistant that automates numerous activities on its own, such as sending emails, making sales calls, etc. to book a trip or sell a car. It is called Muse; within company it was called Hatch.
Believe it or not, a major part of CEO Mark Zuckerberg’s effort to “personal superintelligence” is all about the large audience who use Meta apps on a daily basis. For now, Muse will only work in the United States. It can be used via a new Muse app or via WhatsApp. Meta also announced that the assistant will arrive on their smart glasses later this year, but gave no more details of that.
It’s based on an open-source AI agent, OpenClaw, which was the source of the new tool. Muse can integrate with a multitude of digital aspects of a person’s life, such as email, calendar, payment, health applications, shopping and smart home, among others.
While this is helpful, Muse has access to a lot of personal info and can pose significant dangers when it goes wrong. Meta vice president on AI products, Vishal Shah, has admitted that the company has postponed the launch back in April to make Muse more secure. “It’s just that we think the assistant is now at the safety level that is minimally acceptable for the public,” he said, but noted that there are “inevitable hiccups” that could arise.
That didn’t stop some Muse testers from having difficulties as Meta workers. In an internal posting, which Reuters was able to see, one employee stated that the assistant was very helpful while an assistant was on a three week “honeymoon trip,” arranging routes and transport. Others faced a more difficult time, though. According to some, Muse abruptly disconnected and divulged personal information when not requested. In one concerning incident, the assistant accidentally exceeded its own security protocols and is believed to have taken out private iCloud photos when looking for toys in birthday party photos.
Andrew Bosworth, Chief Technology Officer (CTO) of Meta, also spoke of being logged out of the tool on multiple occasions when they were testing it.
Despite the downsides, Meta is continuing to press on toward a public release as it tries to get a head start on the growing category of AI agents which can represent or take actions for humans.
