Can Garmin Learn from Whoop and Introduce an AI Coach?

More: Whoop In-Depth Review

More: Whoop AI Coach

Men’s Health magazine recently interviewed Whoop’s CTO, Jaime Waydo who outlined the company’s rapid response to AI coaching by introducing a service based on ChatGPT-3. It was initially called Whoop Bot by the early developers and was working internally within 2 months. Here are some of the highlights from that and further thoughts.

Q: …Whoop Coach. How did it come about?

A: In 2022, when ChatGPT-3 came out, I started playing with it, and it actually sounded like you were having a conversation. I thought, ‘What if you could do that for all of the metrics that Whoop [provides] to help you understand your own body?’ I just sent an email to the software engineering team: ‘There’s this new tech. If you haven’t played with it yet, I encourage you to.’ About a month later, I was walking by a conference room late at night, and there were a bunch of people in there eating pizza. I walked in and I was like, ‘Hey, what are you all working on?’ And they were building the first version of, what at the time we were calling, ‘Whoop Bot’. They showed me this prototype where they were piping their Whoop data into ChatGPT and asking it questions. It was super rudimentary at the time: ‘What is your HRV?’, ‘How do you improve your HRV?’ By the end of the week, I had a demo that I could run with my own data. By the end of that month, we had something that everyone in the company could use.

Q: What kinds of questions is Whoop Coach good at answering?

A: The thing Coach does that’s really powerful is it helps you understand your data and slice it and dice it in ways that we would never build a feature to do. As an example, do you sleep better in the summer or the winter? Whoop Coach can tell you that. And is it because of your allergies? Is it because you drink less in the summer? Is it because you’re getting more sunlight? If you’re [using the Whoop Journal to track habits] and you’re asking questions like that, you can get rich answers: ‘The best month you sleep in is July.’ ‘The best night of sleep you had was on July 14.’ ‘Here are some conditions for that.’ And then you can go right into that data and look at it and learn even more.

Q: The technology behind Coach is still pretty new. What will it look like when you get to version eight, nine or 10?

A: I ask the team right now questions such as, ‘How do we build a Coach that coaches you to have the best sleep of your life?’ ‘What does the Coach need to know?’ ‘What is the research we need to be doing to inform that and what is the best way to communicate that?’ I think you’ll start to see more science that’s going in to fuel that coaching. And then I think we’ll see different media coming in: you could record sounds, you could take pictures, you could talk to the Coach. And then I think you’ll be able to say, ‘Give me the TL;DR.’ Then, if I want to go deeper, I can go deeper.

Q: A lot of the conversations around the use of AI in the fitness industry have been about AI replacing human coaches. How do you see that relationship evolving?

A:…Where AI is really useful is… imagine, your client comes in and you say, ‘Hey, Whoop Coach, how is Jamal doing this morning? What’s Jamal ready for?’ And then the Whoop Coach can give you insights into how Jamal’s been sleeping and how Jamal’s stress has been. That becomes a really powerful partnership. The Whoop Coach is with you 24/7, but your personal coach is the one that really knows what your goals are, exactly where you’re struggling and where you’re ready to take on more.

How Whoop Coach Works

Whoop Coach doesn’t do anything you couldn’t do yourself but it does in seconds what may take you days or weeks even if you had access to all the relevant data and scientific knowledge.

Whoop considers several sources of information: your performance, activity and sleep data; your subjective feedback from Whoop Journal; Whoop peer group data; scientific and other data. On top of that is Chat-GPT which interprets your ad-hoc questions to give language-based responses.

Here are some real examples

What Garmin Needs To Do

You and Garmin already have sufficient activity and sleep data for a language model like Chat-GPT to interrogate and answer certain types of questions. For example, you might be able to determine which workouts elicit the biggest increases in FTP or what kinds of taper appear to work best for you. You might be able to see if the volume of sleep of HRV level affected your performance at different times of the year. Things like that.

However, the power and scope of that is limited. You might instead be interested in how you recover from illness, how alcohol or sex affects your performance, or how the time of your last meal impacts sleep quality. There are a myriad more things that might only interest a small number of people and Garmin will never directly record those. If Garmin followed Whoop’s approach it would create a Journal to let you add personal tags such as ‘alcohol consumption’ which then become new data points and can be assessed alongside everything else in Garmin Connect.

Will Garmin do Something Similar?

Yes, I would have thought so. Although there is no inkling that this is happening.

I suspect the upfront costs linked to changing the Garmin Connect app and data storage beneath it would be reasonable. But the ongoing cost of using the AI model could be a stumbling block, AI queries are expensive and energy-consuming as we will become generally more aware of in the months and years ahead. AI is yet another climate issue.