There are a whole slew of very good reasons that I should feel nervous about giving Meta access to my health data. Or my credit cards. Or read/write access to my personal and work emails and calendar.
I think we can all reasonably agree that over the last 20 years, Meta has given reasonable people some very reasonable reasons to think that granting it that much access to personal, private data is very unreasonable.
But I am not a reasonable person.
Which is why I stopped worrying about AI killing us all and learned to love Muse, Meta’s new AI agent. Well, mostly.
Muse launched earlier this month and has received a surprisingly warm reaction. I guess I say surprising because Meta has struggled to launch an AI app that has caught on with consumers. Remember “Vibes,” the video generating Sora competitor? Me neither.
Muse’s strength is that it’s an easy-to-use and understand AI agent. You use it to send it to-do tasks like “organize my inbox” or “Find me a good travel stroller for my upcoming trip” instead of the more common things you use LLMs like ChatGPT for (such as text generation or in place of Google searching).
I started by testing it with some obvious, straightforward tasks. First, I asked it to look through my email and send me a daily digest of “important” emails (my inbox gets quickly clogged with PR pitches and newsletters). It did this admirably well. I also asked it to look at my credit and debit cards and tell me what subscription I could cancel — it suggested Flickr (for irrational reasons only understandable to my human soul, I’m keeping it).
What more could I do? If I gave it more and more access to my life beyond my email and credit cards (!!) would it help me even more?
I gave Muse two main buckets of tasks.
1. Be my kids’ secretary
School-age kids generate a tremendous amount of paperwork, form-filling-out, calendar balancing, and various online sign-ups and email, and keeping on top of it all feels impossible. An AI agent to help me sift through this would be far more useful than in my professional life. But could it cut through the weird mess of strange and clunky apps that schools and kid activities all use?
The task: Plan a kid’s birthday at a trampoline park. This involved getting pricing and availability from the venue, and sending invites.
The result: Failure, but mostly human failure. Muse tried, but ultimately, there are too many decisions about this to be made by the birthday boy (and needed real phone calls to the venue), so this wasn’t a good job to outsource to AI. (Since then, Meta has updated Muse to allow it to make calls, but I don’t love the idea of subjecting a human to an AI call.)
The task: Keep on top of my kids’ school and activity notifications. Staying on top of things across a variety of apps and emails for school and activities is my personal nightmare — so send me a digest.
The result: Mixed results. While it flagged items that came in via email, the school’s Russian nesting dolls of information sometimes stymied Muse’s ability to actually help me.
For example, I needed to fill out a Google Form for band signup, and that Form lived inside a link from ParentSquare. But my ParentSquare login is via my Google account — and Muse’s browser version of the site didn’t include the Google login option. (I told a Meta PR person a few days later about my Google login issue. They conferred with the engineering team, and gave me steps to troubleshoot a way to log in. Success, although that kind of concierge service isn’t available to most users.)
It also couldn’t track any of the Girl Scouts planning information, because that app happens in a busy WhatsApp group — and it couldn’t view WhatsApp messages.
The task: Renew Girl Scouts annual membership.
The result: The Girl Scouts of America defeat AI! The Girl Scouts’ website has always been really frustrating to use, and although Muse almost could do it, it couldn’t complete the final payment section without the site timing out. I ended up having to do it on my computer.
2. Personal health data
I needed to go full Muse guinea pig and give it all the sensitive data I could think of. I connected it to my Apple health and fitness tracker data. Could Muse make me healthier?
Here you go Muse, have my Apple Health data!
The task: Help me lower my resting heart rate. I’ll be honest, I wasn’t that interested in doing this but I wanted to give it something to do.
The result: Sure, fine. It looked at my data and told me my resting heart rate was fine, but suggested I could improve it by sleeping better (no thanks).
The task: Book my annual physical. I gave it my doctor’s name and told it to book for November.
The result: Failure. After giving it a bunch of info like my date of birth and last four digits of SSN (yikes), it couldn’t get past updating the insurance info and told me to call the office.
The task: Order cookies from Walmart+ to arrive 2 days before my period each month.
The result: Failure. The menstrual cycle data from Apple doesn’t sync with Muse, and it won’t connect to the other period tracker app on my phone.
It was able to start a Walmart order, and I told Muse to only order with free shipping, but it took that too literally and didn’t notice it was tacking on a bunch of extra same-day delivery fees. I told it to cancel the task and not place the order.
My takeaway: Muse was helpful, but not ready to fix my life yet
I can see the promise here — it was able to accomplish (or at least half-accomplish) some stuff that could be genuinely helpful. A colleague of mine used Muse to successfully dispute a health bill with his insurance. But it can’t yet do the really annoying stuff I don’t want to do myself.
Meta acknowledged that Muse can’t do everything (yet). “Some of that friction is intentional,” said Jimmy Raimo, spokesperson for Meta told me. “If fees look high or a login looks off, Muse is designed to stop and check rather than push through. We’d rather it ask than get it wrong.”
Although the failures were disappointing, nothing was life-ruining. I just simply had to do the tasks myself on a computer. I’d rather it fail than accidentally spend $5,000, email my whole contacts list, or sign me up for a marathon in an effort to increase my daily steps.
One thing I’ve realized is that 80% of the battle here isn’t how well Muse can execute a task; it’s how well you can come up with a task for Muse to do. It requires a huge amount of knowledge about what kinds of tasks AI agents are suited for, and some creativity to brainstorm for ideas it could do. I suspect the biggest hurdle here was me — I wasn’t good at coming up with good tasks for it.
Maybe I need a muse to help inspire me.
Read the original article on Business Insider
The post Meta’s Muse AI agent is very powerful — but not enough to overcome my kids’ annoying school apps appeared first on Business Insider.
