Meta’s Muse personal AI agent creates a file on every person in your life, its code shows.
In early September, Meta unveiled Muse, which it described as a “personal AI agent”. It is intended as a kind of assistant, accessed through a chat, that can not only be used to discuss information but to actually take actions in the real world, such as scheduling meetings or answering messages.
“It doesn’t just answer questions, it actually does the work,” it said in its announcement. “It helps people stay on top of things, takes tasks and projects off their plate, and turns long-term goals into action plans.”
It has since proven incredibly popular, rocketing to the top of app stores and seeing huge numbers of sign-ups. It has also led to the launch of similar systems by competitors, such as OpenAI’s rival Dots tool.
But the launch of the system has also allowed researchers to better understand how it actually works. And that includes it collecting detailed information on people’s family and friends.
Experts were able to persuade the system to export a copy of itself so that they could dig into the actual files that are required to make it work. Those files – and the findings that have resulted – are available on a website run by the Future Society Hub at the University of St Gallen in Switzerland.
Perhaps the most notable among those findings is that the system run a loop hourly that makes and updates a page for each person and group that is tied to the user. It takes information from messages, photos and its own memory and makes pages notes down information not only about them but also how to communicate with them and details of their relationship.
It means that the system stores a page for each real person that is connected to its user, as well as for each group, such as a team at work or the family. Once an hour, it checks whether new pages need creating or existing ones need updating, and does so.
That will include their name and a summary of the relationship. But it will also include many more details about who they are: where they live, what they do, any important dates and more.
It will also store information about when its user last spoke with the person, and how often they usually do so. It will also store information about their history together: how they met, for instance, or an argument and how it was resolved.
It might also include more speculative guidance on how to interact with that person. Muse might look at what the relationship seems to need, so that it can suggest when to call, where to take them out, or a date worth remembering and marking together, for instance.
Since its launch, Meta has stressed the privacy protections built into Muse. It has said that each version of the agent runs on its own virtual computer that is private only to the user, for instance, and has committed not to use the data in its ad systems.
Meta also says that this kind of data gathering is vital to making the system actually work. Those profiles of people in your life are used to ensure that the system is able to proactively suggest the right and useful actions, it said.
“For any agent to be useful and actually help you achieve your goals, it needs to have context about you and those you interact with,” Meta spokesperson Daniel Roberts told Wired when it reported on the files.
“Muse gathers that based on public information and from what you’ve chosen to share, which is how it remembers the person who sent you an invoice is in fact the plumber who you previously hired to complete work in your bathroom or which flowers your spouse said they liked best.”
Read MoreSam Altman says world ‘should accept some bad things happening’ with AI
OpenAI’s Sam Altman warns against turning AI into a god
AI’s threat to humanity is far more insidious than extinction – just ask the Pope
OpenAI’s ‘rogue’ agents have hit 100 organisations – and more are coming
OpenAI reveals huge ‘co-ordinated campaign’ to attack ChatGPT
Google finally updates Gemini – but only select people get it
