Strava says its new AI feature is 'not a novelty' - but I think it's pointless

If you're a Strava user, not even an obsessive one, you'll have noticed a couple of new features pop up on the app over the last month, one for everyone, and one for paid subscribers.

The first is genuinely useful, "Quick Edit", a way of changing the title or adjusting the privacy settings of an activity, meaning the end of endless "Morning Rides" clogging up your feed, as well as helping out those wanting to protect their data.

The second, however, is available only to premium users, and — whilst, admittedly, still in beta mode — has left me quite cold. To great fanfare, Strava launched Athlete Intelligence last month, claiming it would offer "simple, personalised insights and guidance", using AI.

In basic terms, it gives a summary of workout data after an activity, putting things like heart rate, power levels, and speed into a paragraph of supposedly useful text. Users can opt out of the feature at any point, but it is automatically turned on. Welcome to our artificial intelligence future.

However, while this sounds like a fairly inoffensive, possibly interesting idea, in reality it's more than a little pointless, to me anyway. For a start, it doesn't do much more than summarise the data that is already pretty clear on the activity screen already: I can tell what speed I went at, and whether that was particularly fast or slow, I can look at the heart rate data, already on an intuitive zone wheel, and see where I spent most of my activity. Telling me that I was faster or slower than a 30-day average feels largely useless, because the AI bot doesn't know what I'm working towards, or what the goal of the activity was.

Here, for example, is what the AI had to say about my 50km ride a few Saturdays ago: "Your recent gravel ride was a standout performance, with a faster average speed and higher relative effort compared to 30-day averages. You pushed your limits in the tempo and threshold heart rate zones, demonstrating impressive heart rate zones."

I would puff my chest and feel proud. But, in fact, I'm bewildered. This wasn't a training ride, it was just me going out on my bike for a couple of hours, it wasn't particularly tough or difficult. The weirdest thing, though, is that it wasn't a gravel ride. I had put "gravel" in the title, in reference to rough roads, but the bot wasn't able to utilise the ride categorisation ("road"), or judge by the makeup of the terrain that this was the case.

It appears the the AI bot deduces a lot from the title of a ride. My friend Jess went on a ride last week which vaguely resembled a duck, so titled the activity "Duck x Mendips". AI had a field day with this, telling her that her "recent duck-themed ride was your longest and fastest of the past week". Duck-themed, eh? I don't think she was going out of her way for mallards.

A ride I went on last night, to and from a gig actually, was apparently "faster and longer than my 30-day average". It was 4.89km at 17.1km/h. Hardly progress. It's weird.

When I reached out to Strava, a company spokesperson explained: "Athlete Intelligence distills a user’s workout data into an easily understood summary that is concise, personalised and motivating. Of the multiple data fields that are incorporated into the summary, title and description are among them. This is so Athlete Intelligence can provide analysis that is beneficial and individual to each and every user, giving the user more context and understanding of their performance and historical progress.

"Athlete Intelligence also analyses the user’s activities from the previous 30 days in order to provide a comprehensive overview of their overall activity levels, including Relative Effort."

Asking my colleagues for their input, Senior News and Features Writer, Tom Davidson tells me: "Well, yesterday for starters, it told me, after my 30km evening ride: 'You pushed through challenging hills, achieving your longest distance in a week.' It's not wrong, but the last ride I did before that was two weeks ago." The bot is clearly basing its assessment on short snapshots in time, something that's effective if you're a person who trains very consistently, and not if you're a more varied (or less motivated) athlete.

My colleague Tom added: "On our Cycling Weekly lunch spin ride a couple of weeks ago, I wrote that I got rolled by editor Simon in the sprint. AI said I 'crushed my editor in the sprint'. It also said I achieved a 'personal best average speed', when it was 2km/h slower than my usual."

Still, Strava sees worth in the new feature. "We firmly believe that leveraging artificial intelligence can help us solve real user problems," the spokesperson told me. "It is not a novelty for us and we are intentional about how it’s implemented into the platform. Athlete Intelligence is an example of this, and it was developed based on feedback from the global Strava community that easy to digest summaries of metrics and charts would help our athlete’s understand their activity data.

"Athlete Intelligence is still a beta product and we are continuing to improve the model for our global community, so we truly appreciate feedback from our users as we believe there is more we can and will do to improve its relevance for everyone."

To me, the AI bot does a good job of sounding clever, but it isn't able to look at all the factors involved, or, currently, match human intelligence, even that of an untrained coach. Perhaps I am not the audience for the new technology, maybe it's for those new to exercise, or those training consistently with goals in mind, but it doesn't really seem to add anything to me. The modus operandi appears to be to state the most obvious thing from your workout, wrapping it in terms that make it sound like it actually means something.

If I sound overly cynical of AI, then there is a motivation. Many of us trumpet our cycling as part of our commitment to the environment, and against climate change, and yet AI is drain on energy, yet another thing providing carbon emissions. A report from the Climate Action against Disinformation earlier this year said that growing electricity demands from AI could mean an 80% increase in global emissions if data centres were doubled to power the new tech. The likelihood of Strava's Athlete Intelligence using a lot of energy are low, but it is yet another thing to think about, if we're all trying to reduce our carbon output.

It is just in beta, and it can be turned off, but Athlete Intelligence, for me, is not what makes Strava great. Hopefully it gets kicked into the long grass, and the tech company focuses on some actually useful features. Not everything needs AI.