The marathon, the algorithm and me
Twenty-five centuries ago, after the Greeks shattered the Persian army at Marathon, brave Pheidippides ran 26 miles to Athens with the news. Robert Browning’s poem tells the tale:
“Rejoice, we conquer!” Like wine thro’ clay
Joy in his blood bursting his heart, he died — the bliss!
With the death of Pheidippides began the legend of the marathon, a feat of running so arduous that the very attempt could kill you. I plan to run my first marathon in April in London, hoping to avoid his blissful fate. After all, I have an ally that he did not. Pheidippides, for all his valour, lacked a sports watch.
I was never a runner; my knees weren’t up to it, I’d tell myself. But one thing led to another and, after a couple of years at my local Parkrun, I bought an entry-level running watch, with no aim beyond pacing myself evenly. I didn’t realise that I was plugging my body into the exercise yard of the digital panopticon, with the watch’s app estimating everything from my heart-rate to my step count, and hazarding a guess at my body’s capacity to use oxygen, not to mention my “fitness age”. I had never dreamt such a small box of tricks could provide so many numbers, all claiming in some way to — and here I quote the watch manufacturer, Garmin — “support your efforts to improve and maintain your health”.
There is no denying the technological cleverness here. My watch uses a network of 24 satellites, time signals to within three billionths of a second and calculations adjusting for the irregular shape of the planet in order to pinpoint my location to within 5m. It adds an accelerometer, a device that detects changes in speed or direction using interleaved combs of conductive material etched on silicon that flex and touch as my wrist moves. A strip of flashing green lights on the underbelly of my watch monitors my heart rate by detecting how much light bounces off my wrist rather than being absorbed by the red blood swelling and shrinking my tiny capillaries.
It is all something of a miracle, but more interesting still is the panoply of behavioural nudges, everything from inviting me to share my runs on social networks to tracking my “streaks” of exercise. Last year, I began training for a 10k race, then a half-marathon (more than 21km), using the free coaching software bundled with my watch.
Over 12 weeks of training, my virtual coach would send me off on several runs a week, gradually sharpening the pace and increasing the distance, mixing things up with easier runs or fierce sprint intervals. From time to time, I’d get a short article or a canned video message and, after every run, an upbeat verdict: “Great job!” or “Room to grow.” A coloured dial, purportedly indicating my coach’s confidence, but actually the output of some unknown algorithm, told me how likely I was to achieve my goal on race day.
Without a doubt this coaching programme worked; it prompted me to exercise regularly, and I became faster and fitter. But the longer I used it, the more questions arose in my mind.
There is something about the fitness watch that feels unnervingly familiar after two decades of smartphones and social media. An amazing technology flipping from unimaginable to indispensable almost overnight; the endless tracking, nudging, sharing; the datafication of something that previously had eluded measurement; and a sense of mystery about where all this data is going and how it is being used. On top of all that is something new and visceral: a device worn on my skin, measuring blood, breath, speed and sleep.
Is the fitness watch really to be trusted with my fitness? And can it teach me a lesson about the way so many parts of my life have been transformed into numbers, rewards and targets?
*
Automated fitness tracking began before I was born. In the 1960s, worried that their Japanese compatriots were becoming sedentary, a doctor named Iwao Ohya and an engineer named Jiro Kato developed a simple step-counter. They called it the manpokei or “10,000 stepmeter”. There are various origin stories for the figure of 10,000 and all of them acknowledge that there is no scientific logic for the threshold. That didn’t stop the idea catching on in a big way in the 21st century, when smartphones and fitness trackers began to number our steps and tut disappointedly whenever we missed their arbitrary target.
These tuts make a difference: Katy Milkman, a professor at the Wharton School and author of How To Change, showed me step-tracking data from an unpublished study. Her study subjects walked a variety of distances, but the data displayed a huge spike just beyond 10,000 daily steps, evidence of the powerful urge to satisfy the fitness tracker’s meaningless target.
Still, motivation is motivation. “There is a widespread perception that fitness trackers don’t work, which is incorrect,” says Carol Maher, a professor of population and digital health at the University of South Australia, who has conducted many studies into the effects of fitness tracking. “When you put all the evidence together, it’s very clear that they do help people walk more and take more steps. It’s a modest change but even modest changes are very beneficial.”
Maher and a team of researchers conducted a wide-ranging review of different studies of fitness trackers, covering 164,000 participants. They found all the effects that one might hope for: people tend to be more active, walk more, lose fat, lose weight and gain fitness.
This should not be a surprise. Fitness trackers set us simple goals, record our progress and share our achievements with our friends. All of these behaviour nudges are calculated to prod us into action.
Milkman sent me a short reading list of relevant studies, along with a rapid-fire summary. “Reminders change behaviour,” she told me. “Bite-sized, short-term goals change behaviour and round-number goals are particularly helpful. Self-monitoring changes behaviour. Symbolic rewards like badges change behaviour. Social accountability, such as sharing your exercise, changes behaviour.”
Both Milkman and Maher are convinced that fitness trackers help, and so am I. But help who? And to do what? It’s one thing to coax a couch potato to get up and go for a walk; it’s another to guide an ageing writer to his first marathon. Yet I had put my watch in charge of reaching this all encompassing goal.
*
At the heart of the matter is a piece of human behaviour identified by Milkman in a study conducted with behavioural scientists Linda Chang, Erika Kirgios and Sendhil Mullainathan. The researchers asked a simple question: “Do we decide differently when some dimensions of a choice are quantified and others are not?”
The answer emerged loud and clear from a series of experiments: yes, we do. Whenever experimental subjects were offered a choice between two options, they would tend to favour whichever option looked better on numerical measures and overlook qualities that were expressed as graphical elements, letter grades, star symbols or in words (“moderate”, “excellent”, “highly likely”). This was true whether the choice was between hotels, job applicants, conference locations, public works projects, restaurants or charitable causes. Numbers loomed large. What was quantified, got attention.
This matters because fitness trackers purport to excel at quantifying some things and do not pretend even to quantify others. If quantification fixation applies, we would expect to see such trackers systematically pushing people towards the quantified behaviour at the expense of other things.
An early hint of this came in 2016, when the results of a study of weight loss in 470 people were published. All these people were trying to lose weight, all of them were prescribed a low-calorie diet and all of them were encouraged to exercise. Only half of them, however, were given fitness trackers. To the barely concealed glee of journalists, who love a counter-intuitive finding, the results of the study showed that, after two years, the people who had lost more weight were the ones without the fitness trackers.
Subsequent, larger studies strongly suggest that fitness trackers do not usually hinder weight loss, but the surprising and disheartening finding is an example in miniature of the quantification-fixation problem.
In this case, both groups were equally active, but those using a fitness tracker were getting automatic, effortless validation of their effort, which they could then use to justify more indulgent eating. The lead researcher, John Jakicic, speculated at the time: “People would say, ‘Oh, I exercised a lot today, now I can eat more.’ And they might eat more than they otherwise would have.” Calorie counting is joyless, easily fudged — and not automated by the watch.
We’re all familiar with the tendency to be virtuous in one aspect of our behaviour, then let ourselves off the hook somewhere else — choosing a healthy salad, then using it as permission to order dessert. Psychologists call this behaviour “self licensing” and fitness trackers encourage it by supplying us with asymmetric data. We are told how much we moved, but not what we ate. We get stark feedback on heart rate and step count, but the tracker looks the other way if we order french fries and a glass of beer.
Here’s another instructive example of the way quantification can lead us astray. In a small experiment conducted by Rob Copeland of Sheffield Hallam University, some volunteers were asked to hit the timeworn target of 10,000 steps a day, while others were told instead to take three brisk walks a day, each of about 10 minutes. One of these exercise regimes requires a wearable computer; the other, nothing more than a pair of shoes. Three brisk walks aren’t close to 10,000 steps; in total they are more like 3,000 — not that anyone is counting.
When Copeland studied fitness-tracking data from all the volunteers, he found that those who had done the human-centred exercise of a few short walks had actually done almost a third more “moderate to vigorous” physical activity than the ones grinding out a step count for the algorithm, and found the task less of a chore.
Even on the narrow grounds of cardiovascular activity, the unquantified walk beats the quantified one — and that is before we take into account the benefits of a chat with a friend or the feeling of the wind in your hair. The fitness tracker will handle quantity all day long. But the quality of a walk? That’s up to us to defend.
Our digital devices are quantification machines. Try to count 10,000 steps as you go about your day and you’ll drive yourself mad, but your watch will do it for you without you even noticing. But what gets counted isn’t always what counts.
A brutal callisthenics session in the gym may leave me feeling that I’ve given everything, but the watch sees only my heart rate and is unimpressed. My Taiji practice is a form of gentle exercise that I greatly value, but as far as my watch is concerned I’m not really exercising at all. None of this would matter much if quantification fixation didn’t exist, but it does. It is human nature to take the watch and the activities it quantifies more seriously than they deserve.
*
Over the past 18 months, my virtual coach has paced me to my longest runs and my fastest runs, prodded me to pull on my running shoes and head out the door even when I didn’t feel like it, and broadened the (admittedly narrow) horizons of my training routines. But it has also nudged me into some decisions I regret.
Last winter, I went out running when some of the roads were covered in sheet ice. I avoided mishaps by gingerly picking my way over the obstacles, only to find the algorithm grumbling that I had not run fast enough.
A month ahead of my first flat 10k race, I picked up a minor injury. A coach would have told me to rest and heal, but I worried that the algorithm’s “adaptive” training plan would be derailed if I didn’t keep going. (Many of these training plans call themselves “adaptive” but I have yet to find one that explains how this adaptation works.)
In the end I resolved the tension between my need for recuperation and my desire for a personal best by going to my local park three weeks before race day, gritting my teeth and running the PB I’d been aiming for. Then I switched off the training plan so that I could heal. I doubt I would have achieved the PB without the watch — but I also would never have behaved so oddly.
There’s a word for losing sleep because you’re worried about being judged by your sleep tracker: “orthosomnia”. I’m lucky enough not to worry much about my sleep, but I do worry about my running. It’s easy to see how the powerful lure of a training plan that understands neither ice nor injury could prompt me and others like me into counter-productive overtraining — even permanent damage.
Some of these risks come from poor product design. Garmin’s Connect app, for example, prominently celebrates “streaks” of exercise, meaning the number of consecutive days in which I’ve recorded some kind of activity. Yet any coach will tell you that rest days are vital, so it is strange that my main fitness app applauds me for the number of consecutive days in which I have failed to take a rest.
Other risks are more subtle. When I signed up for the Runna app, for example, it suggested what seemed an absurdly aggressive target for my first marathon time — almost an hour faster than Garmin’s race prediction. The first training run the app proposed was at a blistering pace.
I spoke to Walter Holohan, the chief technology officer of Runna, who was keen to emphasise that the Runna training plan was personalised and it would use a proprietary algorithm to adapt the training schedule to my performance. Could he share any details?
“Obviously, we wouldn’t want to share our proprietary algorithms,” he explained. Obviously. I’ve not yet found a company that will. But that leaves users taking things on trust.
“It’s understandable, of course, because they’ve got competition between one another,” says Joe Warne of the Sports Science Replication Centre at Technological University Dublin. “They don’t want to share their secrets of how they’ve arrived at these values. But the more that we continue to do that, the less that we’re going to have any real insight.”
Given that the history of fitness trackers begins with someone picking 10,000 steps because it’s a nice-sounding round number, the lack of transparency and independent verification of these apps and devices is not wholly reassuring. They are not being sold as medical devices, so regulators do not get involved. I am often told that older runners need more time to recover between each run, so I asked Runna’s Holohan to reassure me that Runna would take into account the fact that I was 52 years old. Alas, he could not. Age-adaptive plans were still on the drawing board, he told me. So were training plans that reflected the menstrual cycle of female athletes.
Reassurance was no more forthcoming from Garmin. The company wouldn’t make anyone available for an interview, and ducked every question about whether the Garmin training recommendations took into account my age.
Facing a marathon, then, which app should I choose? I respect their behavioural savvy and would expect any of them to tug my strings like an expert puppet master, but I am less confident of the physiological science behind their recommendations, as their methods are secret and their pretensions to rigour largely untested.
I don’t mean to be ungrateful: my inexpensive Garmin watch and the free coaching app that was bundled with it took me from weekly wayward 5k runs to a well-paced half-marathon. But perhaps I have come to expect a little too much from my silicon coach.
Iefore my half-marathon, my Garmin app told me my predicted time was 1hr 54 minutes and 56 seconds. Strava, looking at exactly the same data, told me I could go a full 11 minutes faster. Even over a distance of more than 21km, 11 minutes is a huge difference. This put me in a quandary before the race. Everyone warned me not to go off too fast — but given the yawning gap between the algorithmic forecasts, what did “too fast” even mean?
“If you spoke to two different humans they might do the same thing,” says the digital health expert Maher. “It’s easy to believe that technology just has the answer.”
A fair point. I’d never tried to set a half-marathon time before, so any forecast would be little better than a guess. Yet that did not stop both Strava and Garmin making their race predictions to within the nearest second. And it did not stop me taking both of them seriously, and hesitating when they contradicted each other.
*
It is a sobering experience to stare at a marathon training plan.
Monday — Strength Training — 30 minutes
Tuesday — Fartlek (“speed play”) run — 10 minutes @6:05/km. 8 mins @5:35/km, 2 mins easy. 5 mins @ 5:15/km, 90 sec easy. 4 mins @5:10/km, 90 sec easy. 3 mins @5:00/km, 90 sec easy. 2 mins @4:50/km, 90 sec easy. 1 min @4:35/km, 90 sec easy. 10 mins @6:05/km
Wednesday — Easy Run — 45 mins @6.05/km, 15 mins @5.45/km
Thursday — Cross Training — 45-60 mins
Friday — Strength Training — 30 mins
Saturday — Threshold Run — 15 mins @6:05/km, 5 x 5 mins @5:05/km with 1 min rest after each, 15 mins @ 6:05/km
Sunday — Long Run — 120 minutes @6.05/km
That’s week one. It would be an oversimplification to suggest that the following 15 weeks are the same, but further and faster — but not a grotesque one. Although such a training block isn’t easy, it isn’t complicated either. With your fitness watch on and the training schedules programmed in, just pull on your shoes, head out the door and follow the watch’s orders.
But the longer I have followed this sort of plan, and the more I spoke to people in the world of fitness trackers, the more I feel that there is something missing — something unquantifiable. Serendipity, perhaps? Variety? Playfulness? Look again at that Tuesday “speed play” session. Speed, yes. But there is nothing playful about it.
These training plans are relentless and not just in the obvious fashion, where a 52-year-old body with niggles and twinges and the occasional 14-hour work day faces an implacable silicon coach which refuses to negotiate. My physiotherapist shook her head in exasperation when I told her I was planning to use the Runna app for my marathon preparation. Having seen too many people allow an app to overtrain them into injury, she urged me to think again.
But the relentlessness comes in another guise, too. It isn’t just the grind and the risk of injury, but all the times I passed up opportunities that the watch and the training plan could not quantify — opportunities to run with a friend or my wife or my informal local running club. The watch tends to have other plans, and I do not want to disappoint the watch. That is the nature of quantification fixation.
As I reflected on these missed opportunities, I realised that running apps could, in principle, set us a very different kind of training programme.
*
In 1976, David Bowie fled to West Berlin. Beset by legal troubles, drug abuse and a disintegrating marriage, he later recalled, “It was a dangerous period for me.” In the shadow of the machine gun nests along the Berlin Wall, it seemed an unpromising place to make a record. But Bowie had a way of finding new challenges and constraints, which may be why he asked Brian Eno to join him.
Eno began showing up at the Hansa Studios with a selection of cards he called Oblique Strategies. Each card had a different, often baffling instruction:
Emphasise the flaws
Only a part, not the whole
Change instrument roles
Eno would draw a single card at random, and push the musicians to respond. They did not necessarily approve of his randomised provocations — “This experiment is stupid”, complained guitarist Carlos Alomar — but it is hard to argue with the results: two of the decade’s most critically acclaimed albums, Low and “Heroes”.
Years later I asked Eno what the idea of these cards was supposed to be. “The enemy of creative work is boredom,” he told me. “And the friend is alertness.” The random inscrutability of the cards kept generating new situations and new problems. And, as a result, pushed the musicians into situations that could be frustrating but could also be exciting.
So what about injecting a little excitement into marathon training with the occasional Oblique Training Run?
Monday, gym. Tuesday, easy run. Wednesday, go for a run dressed as superman.
Monday, gym. Tuesday, easy run. Wednesday, pack a picnic, run somewhere nice, get the bus home.
Monday, gym. Tuesday, easy run. Wednesday, get a head torch and run in the dark.
Run with a fast friend.
Run with a slow friend.
Make three people smile.
Run a route that draws a picture on the Strava map.
Run with a different soundtrack.
Run in silence.
I’m in training now; wish me luck. My fitness watch will be a vital part of my training practice, but it won’t be the only part. If you see an economist running up the river Thames dressed as Superman or carrying a picnic, that is because in running, as in life, much of what matters cannot be measured.
In their ability to track our running metrics, plot out complex progressions, and push us hard, fitness watches are a wrist-borne marvel. If I make it to the start line of the London marathon in April, I will have my watch on my wrist, pacing every step.
But like Pheidippides, I’ll also hope to have joy in my blood.
I’m running the London marathon run is in aid of the Teenage Cancer Trust. tinyurl.com/HarfordMarathon
First published in FT Magazine on 17 January 2026
