I thought that was interesting too. Looks like the research on it came out 10 years ago. A couple of links from some quick searching. The second link has some charts and graphs, though it limits its estimates to 65+ age groups.

https://www.scientificamerican.com/article/walking-speed-sur...

https://jamanetwork.com/journals/jama/fullarticle/644554

Edit: I'd add that you likely don't need any tracker to get a rough estimate.

If you walk faster than people around you who are roughly the same age, then you'll likely outlive them.

That may be a reason to take it slow and smell the roses, since you have more time :)

Just to respond seriously for a second: it's good to remind ourselves that ML isn't magic. If it knew your weight and age it would use those for predicting life expectancy instead. Same for smoking history, sleep habits, etc. But all it has access to is the accelerometer, and a weak correlation is better than nothing, so that's what it uses.

Someone previously illustrated sceneries where the algorithm hiccupped on your data and erroneously labeled you as a pedophile and/or terrorist (at the same time!).

If you think about it, bad ML (or your words: "...ML isn't magic.") is just as bad, if not worse, than infallible ML.

Apple Health and many fitness tracker apps can estimate your Vo2Max score based on your height, weight and how your heart rate varies during brisk walks and runs.

It's not incredibly accurate but Vo2Max is regarded as an important indicator of your cardiovascular health.

They only know in a statistical sense. Useful for insurance companies, but not on a personal level.