Defaulting to male seems particularly endemic in sports tech. Starting with the most basic, the calorie count on treadmills is perfect for practically no one, but it will be more accurate for your average man because its calculations are based on the average male weight (the default setting for calorie count on most exercise machines is for a person who weighs eleven stone). And although you can change the weight setting, that still leaves a calculation based on an average male calorie burn. Women generally have a higher fat and lower muscle distribution than men as well as different ratios of various muscle fibres. What this means at a basic level is that even after accounting for weight difference, men on average will burn 8% more calories than a woman of the same weight. The treadmill does not account for this.
There’s no reason to think that things improved much with the advent of wearables, either. One study of twelve of the most common fitness monitors found that these underestimated steps during housework by up to 74% (that was the Omron, which was within 1% for normal walking or running) and underestimated calories burned during housework by as much as 34%.22 Anecdotally, Fitbits apparently fail to account for movement while doing the extremely common female activity of pushing a pram (yes of course men push prams too, but not as often as the women who do 75% of the world’s unpaid care). Another study, which unusually did manage to include almost 50% female participants, found that fitness devices were overestimating calorie burn by significant amounts.23 Unfortunately, they failed to disaggregate their data so it is impossible to know if there were any sex differences.
Tech developers even forget women when they form the potential majority of customers. In the US, women make up 59% of people over the age of sixty-five and 76% of those living alone, suggesting a potential greater need for assistive technology like fall-detection devices.24 The data we have suggests that not only do older women fall more often than men, they also injure themselves more when they do.25 Data analysis of a month’s worth of emergency department visits in the US found that of the 22,560 patients seen for fall injuries, 71%, were women. The rate of fracture was 2.2 times higher in women, and women had a hospitalisation rate 1.8 times that of men.26
And yet despite women’s arguably greater need (as well as research indicating that women tend to fall differently, for different reasons, and in different places), gender analysis is missing from the development of this technology. In one meta-analysis of fifty-three fall detection device studies, only half of them even described the sex of participants, let alone delivered sex-disaggregated data;27 another study noted that ‘Despite extensive literature on falls among seniors, little is known about gender-specific risk factors.’28
The Proceedings of the 2016 International Conference on Intelligent Data Engineering and Automated Learning points out that ‘a notable motivation for elders to reject fall-detection devices is their size’, suggesting mobile phones as a solution.29 Except this isn’t really a solution for women because as the authors themselves note, women tend to keep their phones in their handbags, ‘where fall-detection algorithms will likely fail because they are trained to detect falls through acceleration sensors close to the body trunk’.
In acknowledging this, the authors are unusual. Whitney Erin Boesel, a researcher at the Berkman Center for Internet and Society at Harvard, is a member of the ‘quantified self’ community, which promises ‘self-knowledge through numbers’. These numbers are often collected via passive tracking apps on your phone, the classic being how many steps you’ve taken that day. But there’s a pocket-sized problem with this promise: ‘Inevitably some dude gets up at a conference and [says] something about how your phone is always on you,’ Boesel told the Atlantic.30 ‘And every time I’ll stand up, and I’ll be like, “Hi, about this phone that is always on you. This is my phone. And these are my pants.”’
Designing passive tracking apps as if women have pockets big enough to hold their phones is a perennial problem with an easy solution: include proper pockets in women’s clothing (she types, furiously, having just had her phone fall out of her pocket and smash on the floor for the hundredth time). In the meantime, however, women use other solutions, and if tech developers don’t realise women are being forced into workarounds, they may fail in their development.
A Cape Town-based tech company fell into this trap when they developed an app to help community health workers monitor HIV-positive patients. The app ‘fulfilled all the usability requirements; it was easy to use, adaptable to local language’ and solved a very specific issue. More than this, the community health workers were ‘excited at the prospect of using it’.31 But when the service was launched, it proved to be a flop. Despite several attempts to solve it, the problem remained a mystery until a new design team took over the project. A team that happened to have a woman in it. And this woman ‘took only a day to discover the problem’. It turned out that in order to more safely complete their daily commute into the townships where their patients lived, female health workers were concealing their valuables in their underwear. And the phone was too big to fit in their bras.
Gender affects the kinds of questions we ask, says Margaret Mitchell, senior research scientist at Google. Limiting AI
