I’m not a fan of AI well being and health options. Not solely do they regurgitate Captain Obvious-level summaries of what you simply did, however the “insights” are so generalized {that a} Google search is usually extra useful. So it was with nice skepticism that I walked right into a demo to find out about Fitbit’s forthcoming AI-powered private well being coach. To my shock, I left cautiously interested by the place Fitbit goes.
“We actually wish to transfer in the direction of this world of teaching,” Andy Abramson, Google’s director of product administration for Fitbit and Well being, says throughout a demo of the characteristic. Skilled athletes have a complete staff of individuals serving to to craft their health regimens. “We requested ourselves, what if everybody may have one thing like this?”
On paper, Fitbit’s well being coach isn’t providing something a dozen different well being and health tech corporations haven’t already promised. It’s a chatbot constructed off Gemini that lives in a spiffy, redesigned Fitbit app (now with darkish mode!). Every week, it builds customized routines with detailed exercises and exercise targets based mostly in your private well being targets. These exercises will regulate based mostly in your real-time information. So if in case you have a crap night time of sleep, the subsequent day it’ll tweak your instructed exercise to compensate. It’s also possible to proactively inform the bot anytime you’re sick, injured, or have a brand new aim, and it’ll take these issues into consideration. If it notices traits in your information, like improved sleep high quality, it’ll flag them to you.
Fitbit’s coach is an try to deal with the age-old downside of wearables offering customers means an excessive amount of information with out the suitable context. What’s been sorely disappointing from opponents to this point has been the execution. Usually, it appears like AI has been tacked on as a gimmick to please buyers reasonably than developed as a device that may present genuinely useful insights.
The place Google and Fitbit’s take feels totally different is that the product has been utterly but cautiously overhauled across the idea.
“It’s not similar to a brand new coat of paint. It’s not simply AI bolted on. We’ve actually requested the query of: How will we put the AI coach in each a part of the app?” Abramson says.
Fitbit’s coach actually is prominently baked into each nook of the brand new app. The At the moment tab, which shows your each day metrics, has been reorganized right into a smaller information dashboard with an AI chatbox proper beneath. While you scroll down, there are blocks calling out insights based mostly on metrics like sleep. Beneath every are prompts to interact with the coach additional on every matter.
In our demo, Abramson reveals me how the AI coach interprets his personal private information. Some elements seem to be the identical outdated regurgitated ebook experiences. In others, nevertheless, I can see glimpses of the promise. For instance, Abramson is ready to inform the AI that his overarching aim is to get higher at path working, however that he’s touring and has restricted entry to tools. In response, the app instructed a exercise incorporating the Peloton bike he has entry to on the lodge. The coach additionally notes that due to jetlag, Abramson’s had much less sleep the night time earlier than however with fewer interruptions in comparison with his ordinary. It then asks to verify in on his power ranges for the day.
The important thing right here is real-time adjustment based mostly on dialog. Abramson relays one other story of a staffer who damage their finger and requested the coach to take away power exercises in the intervening time. Per week later, the coach checked in asking whether or not the finger had healed and if it was okay so as to add these exercises again in.
1/2
In comparison with different health AI I’ve examined, this demo coach can also be pretty chatty. In Abramson’s logs, there are prolonged blocks of textual content peppered with metrics and information breakdowns. Google VP of Fitbit and Well being Rishi Chandra says that is intentional.
“The LLM can summarize it if you need three strains, however it is going to be so generic that it doesn’t really feel prefer it’s telling you something,” Chandra says. The staff explored shorter summaries, however early testers instructed them that they weren’t in any respect useful. “It is a balancing act now we have proper now. We’ve proper now listed on getting extra depth for customers after which determining how you can trim that.”
Fitbit can also be shifting away from each day targets towards weekly ones like an precise private coach would do. “A coach wouldn’t say day by day it’s important to get this precise 10,000 steps or no matter it may be,” Chandra says.
Cardio Load, launched final 12 months, was initially designed as a each day aim for individuals to know what they wanted to do to enhance their cardiovascular well being. Going ahead, this characteristic will likely be a weekly goal. Sleep insights may also be based mostly in your weekly and long-term patterns, and the coach may also recommend adjusted schedules if it finds your sleep debt is extreme or if it determines you want additional relaxation from a tough exercise.
That is only a sensible change that enables for better flexibility. A each day cardio goal doesn’t work in case you’re caught on a 14-hour aircraft trip and all you possibly can handle is a chill yoga session if you arrive at your lodge. The change lets customers and the app account for all times getting in the way in which.
A part of making this all work is ensuring Fitbit’s app really has the information it wants. The sleep monitoring algorithm, which processes information from Fitbits and Pixel Watches, is meant to be extra correct. The Pixel Watch 4 can also be including retroactive exercise logging so that you just by no means lose credit score in case you overlook to log a stroll or a exercise. Customers will be capable of prepare the AI to extra precisely acknowledge sure actions over time by means of tags, too. The AI coach may also be capable of soak up information logged from third-party apps by means of Well being Join and HealthKit — so that you don’t have to do all of the native exercises within the Fitbit app. And, as a result of Fitbit units are platform-agnostic, technically the coach can work with iOS too. The aim is to turn into extra of a linked AI well being hub, finally branching from sleep and health towards different points like diet and psychological well being.
After all, something is feasible when a characteristic is in growth. When the characteristic really launches in October, it’ll be an opt-in preview for Fitbit Premium customers solely. (One perk is that the preview isn’t restricted to the Pixel Watch 4; it’ll work with any Pixel Watch or Fitbit {hardware}.) There are additionally nonetheless a number of considerations I’ve. LLMs are restricted, are liable to hallucination, and will teeter dangerously on the road between medical tech and wellness. Information privateness is one more can of worms.
However having examined a dozen lackluster AI well being coaches, this feels closest to reaching the platonic supreme that I’ve seen but.
