OUTSIDE MAGAZINE 61
MA
RK
M
AN
N/
AU
GU
ST
Coach in
the Machine
IS ARTIFICIAL INTELLIGENCE THE
NEXT FRONTIER OF CUSTOMIZED
WORKOUTS?
BY PETER ANDREY SMITHFITNESS TRACKING took a leap forward in
the early 1960s, when the Japanese company
Yamasa released a pedometer called Manpo-
kei, which roughly translates as “the 10,000-
steps system.” This early wearable helped
establish the myth that there’s an optimal
number of steps you should take every day.
Decades later, even as GPS and app-based
tracking have proliferated, our increasingly
sophisticated devices are often little more
than counting tools that log activity. The
newest innovations, by contrast, promise to
adapt with you, using past performance data
to craft customized workouts using machine-
learning algorithms. It’s almost like having an
automated personal trainer.
For the uninitiated, artificial intelligence
is the much hyped term for computational
systems designed to mimic human reason-
ing. Wearables that incorporate AI, in addi-
tion to logging activity, can replace human
decision-making, using algorithms to find
patterns and make predictions. In theory,
the more data collected, the better the algo-
rithm becomes at customizing your training.
(These so-called recommender systems are
not unique to the fitness world, of course.
They’re also how Netflix turns your viewing
habits into a list of suggestions.) Personal-
ized workout algorithms make up a small but
growing portion of the fitness-app space,
which is expected to eclipse $15 billion by- And there are signs that the trend is
going mainstream: even Planet Fitness, the
budget-friendly gym chain, has expressed
interest in AI to adjust members’ routines ac-
cording to performance gains and goals.
This spring, feeling bored with my own
routine, I decided to give digital coaching
a try. I installed Freeletics, which with 30 mil-
lion users is one of the more popular apps
claiming machine-learning capabilities. The
Munich startup has a large client base in Eu-
rope and last year raised several million dol-
lars to expand its reach in North America.
The app offers a menu of individual exercises
as well as a handful of signature workouts;
Prometheus, for example, is a core and lower-
body circuit featuring push-ups, mountain
climbers, sit-ups, squats, and jumping jacks.
There’s also the option to join a premium
subscription service that gives the user ac-
cess to a feature called Coach, which uses an
algorithm to devise an individualized training
schedule. (Three months costs $35.)
I signed up for the premium version. An
initial assessment consisted of timed high-
intensity interval exercises—no equipment
required. The app cycled through a series of
moves (each with a video tutorial) and tracked
how long I took to complete each. Afterward
it asked me to rate my level of exertion: “I can
do even more,”“It was OK,” or “It was too
much.” I was prompted to do the same for
my form, rating myself on a scale from poor
to excellent. Factoring in the time I took to
complete these benchmark activities, as well
as my subjective self-assessment, Coach
built a roughly 30-minute custom workout
plan. (The app can put together a series of
plans, based on how many days a week you
exercise, and allows you to exclude muscle
groups if you have an injury or need a rest.)
As I sped through squats and took breaks
between sit-ups, I couldn’t tell how closely
the recommended workouts had been geared
toward my initial benchmarks. But right off
the bat, Coach encouraged me to do exer-
cises, like burpees, that I have trouble moti-
vating myself to do.
For comparison I downloaded a few simi-
lar apps, though I didn’t try anything that
required stand-alone hardware such as a Fit-
bit or smartwatch. Most begin with a base-
line assessment or calibration workout and
request some general information, such as
weight, gender, and fitness goals. Keelo ($31
for three months) offers a CrossFit-like chal-
lenge that focuses mostly on high-intensity
interval training with an optional weight-
lifting component. Fitbod ($60 for a year) is
tailored almost exclusively to lifting weights
and accounts for available equipment whiletargeting specific muscle groups. Some apps,
such as Kudos ($95 per month), generate per-
sonalized regimens but also put you in touch
with a human trainer as part of the subscrip-
tion fee. Still others are geared toward spe-
cific activities: Podium ($55 for 14 weeks), for
example, helps runners hit a target pace and
distance.
According to Julian McAuley, a professor
of computer science at the University of Cali-
fornia at San Diego, many of the available op-
tions are pretty rudimentary. The algorithms
tend to be created from global models; in
other words, the recommendations are based
on how you compare with people of similar
age, weight, and gender, rather than being
tailored to fit your specific biometric data.
But McAuley says that machine learning has
the potential to become far more sophisti-
cated. (And because successful approaches
to achieving health and fitness vary widely,
personalization could be especially useful in
a training context.) McAuley’s team has used
tracking data to generate customized running
routes, which could tell you what distance
and elevation gain might bump up your heart
rate to a given level. But, he says, “This is
really cutting-edge stuff that’s a couple years
away from being totally practical.”
I’m still using Freeletics, but only once a
week at most. Coach keeps pushing me to do
a variety of quick, difficult workouts, and I
find that the intensity rarely wavers. This is,
I suppose, the whole point of setting goals
and having someone hold you accountable.
For now, though, Coach doesn’t motivate me
quite like a personal trainer might.
Since these apps generally don’t offer much
feedback on form or technique, they probably
work best as a supplement to advice from a
human expert or a group class. And I don’t
exactly relish my time with Coach. Freelet-
ics and other apps require you to be plugged
in, an obvious downside if, like me, you look
forward to your workouts as a respite from
screen time. And while these products offer
ways to share results on social media (Free-
letics also has an in-app leaderboard), in my
experience these features pale in comparison
with the flesh-and-blood camaraderie I’ve
found among fellow gym-goers.