What is Muse and Dots?
- They are recently released personal AI agents, Muse from Facebook owner Meta and Dots from Open AI, the company behind Chat GPT. They are meant to be used in a different way than regular chatbots, which mostly answer questions when you ask them. Agents that are always active and can perform tasks for you. Tell them to keep an eye on the job chat, note when something important is said, create a to-do list, and maybe even do some of the tasks themselves. They can read your files, your emails, use programs on your computer. If you allow it, that is.
What should they do?
- In companies' marketing, they function as ever-present personal assistants, for private and professional life. There, they help you with a powerpoint before a meeting, note that new figures have appeared and adjust the presentation on their own. Then they read in the email that a cake delivery for one's wedding has been cancelled, but quickly find another one and make the reservation. They can even make calls to one, with a robotic voice that never denies it's an AI agent, telling you it's representing you. It's only supposed to be this smooth in the world of commercials, but it shows what the aim is: AI agents that don't just react to your commands, but take their own initiative and perform tasks for you.
What is required for it to work?
- Now we get into the scary side of it all: In order to function well, the agent must get access to as much of your digital life as possible. In Muse's case, it revolves a lot around Meta's platforms, but it also has a built-in browser that it can use to access your online accounts. Dots can run on your computer and access programs and files on the hard drive, but also all conceivable web services. Meta and Open AI are careful to emphasize that you set the limits of what it can achieve and what the agent is allowed to do on its own. But on the other hand, it works best the more access it gets. Does it feel comfortable to allow an AI agent, made not only to talk to you but also to perform tasks, to have such access?
Why is it sensitive?
- It is more than just a matter of personal integrity. You probably already have a lot of private information on services belonging to the big tech companies. But here it's not just about storing data, but about giving an AI system the right to perform tasks for you. Not just read emails and chat messages, but send them. Not just search online stores, but make purchases. Not just running the company's internal chat, but talking in it ("Share the results with the team," says an unabashed user in Open AI's commercial without looking further into what her agent has come up with.)
Earlier this year we could read about a poor woman who had 200 emails deleted by an AI agent. She had asked it to go through the inbox, but given clear instructions to ask permission before doing anything. The woman's name is Summer Yue and her job is to make AI systems obey human intentions on … Meta. (In fairness, though, it wasn't Meta's agent, but a more hobby-developed predecessor called Openclaw.)
Can I test?
- Actually not, at least not if you live in Sweden. As is often the case, they will be launched first in the US, with an unclear date for Europe. That ties in with EU regulations on digital services and AI, which often delay technology launches here.
