Discussions about how AI might disrupt the economy have a blind spot: they largely fail to account for the valuable activities people do outside of paid work.
Statisticians call such activities “household production,” but they collect very little data about them. The American Time Use Survey (ATUS) is conducted annually in the United States, but its scale is small, and most OECD countries collect such information less frequently.
As a result, there is very little economic research, and next to no policy debate, on how most of us spend most of our time.
But investors and policymakers should care about unpaid activities. When social or technological shifts move activity across the production boundary—from paid to unpaid or vice versa—our interpretation of economic signals must also change.
In the 1970s and 1980s, official GDP- and productivity-growth figures rose in countries like the United Kingdom and the US partly because changing social conventions were causing more married women to spend more of their time doing paid work.
Digital technology has already led to major shifts across this production boundary, mainly from paid to unpaid.
People now use digital tools to handle a wide range of tasks, from banking to travel planning, that once required consultation with a professional, not to mention ringing up their own groceries at the supermarket checkout.
Meanwhile, many people share their output—open-source software, entertaining videos, music lessons, or DIY tutorials—online for free.
Where people find AI most useful outside paid work
All of these examples represent the substitution of unpaid for paid work. And, for the most part, they have created a good deal of consumer value.
With AI promising to move yet more activity across the production boundary, the question is how to ensure that this, too, will prove valuable for households.
Answering this question will require us, first, to get a sense of where people find AI most useful outside paid work. Google’s ATLAS project (on which I was an adviser) provides some early insights.
It mapped millions of anonymized Gemini interactions to the task categories used by the ATUS and contrasted the volume of US users’ AI conversations related to each category against the actual hours people dedicate to it.
Topics like “Eating and Drinking” and “Traveling” were significantly underrepresented in AI interactions
By this metric, topics like “Eating and Drinking” and “Traveling” were significantly underrepresented in AI interactions, compared to actual time spent. Interactions relating to “Socializing, Relaxing, & Leisure”—the category where overall AI usage is the highest—were relatively close to parity with actual time spent.
“Education,” however, was overrepresented, with its share of AI interactions about 5.8 times larger than its share of people’s time. “Professional & Personal Care Services” and “Consumer Purchases” were also over-indexed.
But nothing compared to the overrepresentation of “Government Services & Civic Obligations,” which appeared nearly 20 times more often in the AI interactions than in official time-use data.
Dealing with government bureaucracy
In her new book The Time Tax, Annie Lowrey points out that, in the US, people spend six working days per year, on average, dealing with government bureaucracy, completing tasks like renewing licenses, filing taxes, and applying for assistance.
Private-sector administration, from negotiating with insurers to canceling subscriptions, compounds this time tax.
There is good reason to think that AI can help to reduce this burden. Good government digital services already offer efficiency and convenience:
I recently watched a friend apply for a passport online in just a couple of minutes. New York City Mayor Zohran Mamdani’s “click-to-cancel” rule shows how policy can also ease the burden of private-sector administration.
New York City Mayor Zohran Mamdani’s “click-to-cancel” rule shows how policy can ease the burden of private-sector administration
Governments should seek to leverage AI to advance progress on both fronts. Unfortunately, while many governments are keen to use AI to make public bureaucracies more efficient, they have so far tended to focus on cost savings for the state, not time savings for citizens.
Even without government action, the use of AI tools can save people both time and money.
A high-street lawyer in the UK might charge a client £12,000 ($16,220) to apply for probate and administer their deceased parent’s estate, but LLMs are increasingly capable of completing much of the paperwork involved for free.
Other examples of household administrative tasks that might be tackled using a chatbot, or even delegated to an AI agent, include designing interiors or gardens, researching consumer purchases, and queuing up for items like concert tickets. The list is long.
Productivity gains in unpaid work
Just as digital technologies led to the disintermediation and re-intermediation of many services, generative AI models will disrupt a wide range of services.
New law-tech startups might find themselves competing with free chatbots or agents, rather than with expensive traditional law firms. Similar dynamics will emerge in industries like education, health, finance, and entertainment.
This need not be entirely bad news for human service providers. Free AI access will expand the market for these types of services, in addition to substituting for expensive versions them.
The productivity gains in unpaid work—which is economically valuable, despite being largely left out of official statistics—could be tremendous
There will be growth in complementary products, too: If I can redesign my own dining room for free, maybe I will be more inclined to buy wallpaper, drapes, and new furniture.
Meanwhile, the productivity gains in unpaid work—which is economically valuable, despite being largely left out of official statistics—could be tremendous.
Any economist attempting to predict how AI will reshape the economy, and any policymaker seeking to update consumer protections for the AI age, must account for the household side of the production boundary.
Diane Coyle is a Professor of Public Policy at the University of Cambridge.
