What's that coming over the hill is it a majestic AI Gift-Horse?
Ignoring women until it affects someone who matters
Dear GFPs, hello and welcome to the first edition of the Invisible Women newsletter in 2026! We’re starting the year with the thrilling news that an unlikely ally has emerged in the fight to address GDP’s woman problem. You remember GDP’s woman problem, I wrote about it extensively in Chapter 12 of Invisible Women, and as I’m sure you’ll recall, the problem goes a little something like this:
Back when countries first started getting serious about measuring the economy, they decided to count a lot of stuff. They counted all the things we bought, all the things businesses made, and all the things the government paid for. But there was one thing they decided was just too difficult to count and that thing, inevitably, had a lot to do with women. As Diane Coyle, professor of economics at Manchester University, explained to me at the time, no one disputed that there was an economic value to the unpaid cooking, cleaning, child-rearing and general all-purpose bum-wiping done predominantly by women. It was just too complicated to calculate the value of this unpaid work when it came to summing up a nation’s economic output. And so where this work should be in our calculation of the economy there is instead a great big fat data gap.
As I have said so many times by this point that I’m even starting to bore myself, this matters because GDP is integral to how governments decide to allocate resources and bad data leads to bad outcomes — usually for women. But as I also said at the time, there were growing rumblings about finding a way to start counting unpaid work. Not, you understand, because the lads at the top had finally cottoned on to the economic illiteracy of excluding vast swathes of the economy because “wah it’s too difficult,” but because unpaid work was no longer so heavily weighted towards the unimportant half of humanity.
“With,” I wrote, “the rise of digital public goods like Wikipedia and open-source software (which are displacing paid goods like encyclopaedias and expensive proprietary software), unpaid work is starting to be taken seriously as an economic force — one that should be measured and included in official figures. And what’s the difference between cooking a meal in the home and producing software in the home? The former has largely been done by women, and the latter is largely done by men.”
Still, there has been little serious movement on this subject since then. I did briefly hope that the one silver lining of the pandemic might be that we finally started to take the economic contribution of unpaid work seriously, but alas, that ship has sailed off along with a growing number of women’s reproductive rights. But then, last week, lo! What’s that coming over the hill is it a monster OpenAI??
Aaron Chatterji, OpenAI’s chief economist got feminists all hot and bothered last week, when he started talking about the economic impact of tasks like “unpaid housework” that “mostly fall on women”. Be still our beating hearts! (And thanks to James Ball for sending the heart-fluttering article my way.)
“Whether it’s caring for children or doing household chores, these are things that need to get done,” Chatterji explained in an interview with the Financial Times. “But given that it’s not counted [in GDP], a lot of those benefits will never show up.” And so, Chatterji says, “we have to figure out ways to measure this more carefully.” And as with digital public goods before it, this latest effort to account for unpaid work is not being pushed out of a desire to count the economic contribution of women. Nor is it in fact being pushed in an effort to count the contribution of men, or indeed the contribution of any living breathing human being.
No, it seems as if the thing that might finally drive us to count the contribution of unpaid work to the global economy is…AI. Specifically, how much value AI is contributing to the global economy with all of its selfless unpaid care work. You know, like telling teenagers to kill themselves / helping Sam Altman figure out how to change a nappy.
I mean, look, whatever it takes and I’m not going to look an AI gift-horse in the mouth — can you imagine how terrifying that would be? Sure, it’s a little galling, but women are used to having our contributions and needs and bodies taken seriously only when they start affecting someone, or as in this case, something important.
The very fact that OpenAI has anything to say at all about the impact of AI on unpaid housework is worthy of note, because this is extremely rare. Admittedly, OpenAI may only be talking this up because AI’s impact on paid work is almost universally gloomy (AI coming over here and literally stealing our jobs although probably not our women), but still, I’m now scared of the AI gift-horse and have only nice things to say about its great majesty. But it is also true that machine learning analysts have on the whole tended to stick with the grand tradition of simply ignoring women, with pretty much every paper that considers the impact of AI on work exclusively focused on its impact on paid work. Remind of you anything? Cough GDP cough. In fact, the very first paper on the potential future impact of AI on unpaid work was not published until 2023 — a full ten years after the first major working paper on the impact of AI on paid work was published.
The original 2013 paper, which predicted that 47 per cent of jobs in the US were at “high risk” of automation, kickstarted a “Future of Work” industrial complex, spawning thousands of media citations, research replications, policy responses, funding calls and investment bets — all of them focused on the impact of AI on the paid workplace, even though, as the authors of the 2023 paper (which I will from now on refer to as the DomesticAI group) point out, “people in industrialized countries on average spend comparable amounts of time on unpaid work” and “productive work in the labour market and reproductive work in the domestic sphere are just as vital for the functioning of societies.”
The amount of time people (mostly women) spend on unpaid work also affects their ability to engage in paid work. And this means that if AI-driven automation could reduce the amount of time people (mostly women) spend on unpaid chores, it could mean that more of those people enter paid employment, with a knock-on effect on productivity, tax revenue and government spending. You know, all the important stuff we think is worth measuring as part of The Economy™. Indeed, argue the DomesticAI group in a subsequent paper, if automation does reduce the amount of time people (mostly women) (is this getting old yet?) (probably not as old as women doing the world’s scut-work) spend on unpaid work, “the social and economic implications could be enormous.” Gosh, I wonder why people might overlook an area of research with enormous implications? It’s a mystery!
Quite how enormous the implications are, however, is in the eye of the beholder, and this is where the first DomesticAI paper gets seriously interesting. Both the original 2013 paper and the DomesticAI paper convened a group of experts to estimate how automatable a range of jobs might be. But where the 2013 paper treated the answers given by their group of experts as unproblematic, objective, accurate fact (and indeed, gave little to no information about who the experts were) the DomesticAI group not only made the effort to recruit a balanced group of experts, they also considered the difference in the responses between them based on both sex and nationality. And their findings were fascinating.
The DomesticAI experts predicted overall that “on average 39 percent of the time spent on a given task could be automated within ten years,” with housework being seen as more automatable (44 per cent) than care work (28 percent). But this top-line figure concealed significant differences of opinion. Male experts in the UK were more bullish than female experts in the UK about the potential for automation to reduce the hours spent on housework, which is unsurprising: men tend to be more bullish about technology in general. But in Japan, something surprising did happen: this trend was turned on its head, and here it was women who were more optimistic about the potential for housework to be automated. And the DomesticAI group think this disparity can be explained by looking at differences in how unpaid work is shared between men and women in the UK versus Japan.
In every country in the world, women do more unpaid care work than men. But as Orwell didn’t say, because he too did not spend any time thinking about women’s unpaid care work, some countries are more equal than others. “It would not,” the DomesticAI group writes, “be uncommon for Japanese men in expert positions to have almost no personal experience of major domestic work tasks, which they delegate instead to their wives,” with data indicating that only 52% of working-age Japanese men reporting doing “some” domestic work, in comparison to the 88% of British men who do.
More broadly, working-age women in the UK spend 290 minutes on unpaid work on an average weekday, compared to men who spend 141 minutes; whereas in Japan, working-age women spend 292 minutes on unpaid work on an average weekday, while men spend just 53 minutes. And Japanese men’s relative unfamiliarity with unpaid domestic work, posits the DomesticAI group, may well be affecting their attitude towards its automation. Certainly, the experts themselves explain their pessimism not on the basis that they think the tasks couldn’t be automated, but on the basis of what they saw as “low demand for domestic automation considering its high cost.” And, well, yeah, I’m sure I wouldn’t see the point of a robot cleaner if I already had a full-time human one. By contrast, one of the Japanese female experts pointed out that people are already choosing to pay top dollar for high-end domestic appliances, so why wouldn’t they extend that choice to automated ones?
Well, first of all, because there’s no guarantee they will get a chance to: as women know all too well, just because we would in theory buy a product if it existed, that doesn’t mean that it will, in fact, exist (a car tested to be safe for actual female bodies anyone?). And with AI and automation being such heavily male-dominated fields, it is by no means guaranteed that the voices saying “yes please” to housework automation won’t be drowned out by the voices saying “can’t we just get women to do it for free tho?” This type of thinking may also be behind the finding that “female, but not male, income is positively associated with dishwasher adoption in Japan” while “no such difference is observed in the adoption of air conditioners, PCs, or cellular phones”. It’s almost like women value their time or something?
Still, if washing machines teach us anything, it’s that we can’t rely on technology to fix social problems like women being the sex class voted most likely to enjoy cleaning toilets. Professor Ekaterina Hertog, one of the DomesticAI group’s Principle Investigators reminds us that the washing machine was once heralded as the liberator of women, and, well, look how that turned out:
Washing machines made the task of keeping laundry clean a lot easier and less time-consuming. They were powerless, however, to challenge the social roles which expected women to do the lion share of domestic work. Moreover, as the widespread adoption of washing machines led to an increase of hygiene standards, much of the time freed up by the washing machines was immediately taken by domestic tasks around laundry washing, such as folding and ironing, that became more time consuming with the increased frequency of washing clothes. (Source)
And we shouldn’t assume that automating housework would “narrow the gender disparity in absolute time spent on domestic work,” either: in Invisible Women I reported on a study which found that “even in wealthier couples who pay for domestic help, the remaining unpaid work is still distributed at the same male to female ratio, with women still doing the majority of what’s left.”
Still, whether or not automation actually ends up freeing up women’s time, or if we simply find new and unusual ways to keep women domesticated, I think the techno-optimists have it right and the majestic robots are indeed coming for our unpaid jobs; after all, robot vacuum cleaners and floor mops are already “by far the most popular type of robot today in terms of units sold.” And you know what? I think we’ll probably get around to figuring out the economic value of their unpaid work too.
A paid work AI-themed Coda
Scrolling through Instagram the other day I came across this reel by the FT’s resident data bod, John Burn-Murdoch, all about “How to AI-proof your job.” And it turns out that:
Even within tech jobs themselves, the data says that social skills are more predictive of having a well-paid job than pure technical skills. Sorry, James Damore, guess you were wrong. (Although tbh we already knew that).
Poppy pic of the week
That’s it! Until next time, my dear majestic AI gift horses xoxoxo
They’re only trying to determine the value of the domestic work so that they can charge a higher price for the AI-“replacement”… An ex-colleague is developing nursing AI to “automate a lot of the grunt work that nurses do, and then we’ll charge nursing labor prices”. Meanwhile the nursing staff will be expected to do all the other, more difficult work but at the same labor rates as before. It’s disgraceful. Plus ça change, plus c’est la même chose…