Government Doesn't Just Waste Your Money, It Wastes an Hour of Your Week
Plenty of countries file taxes for their citizens; the IRS is legally barred from doing it for you. The time tax hits the poor, immigrants and the formerly incarcerated hardest — and AI may not save them.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The IRS is legally barred from filing your taxes for you
Lowrey's core example: in Japan, salaried workers never touch the tax process at all; many countries send you a pre-filled return and you just correct the errors. The US could easily do this — nine in ten households take the standard deduction, so their tax situations are fairly simple — but the IRS is not legally allowed to. This is not a technical problem, it's an institutional choice. She defines the "time tax" as the portion of cost the government shifts onto individuals in the form of administrative labour, and unlike a monetary tax it's different in kind: you can earn more money, you cannot earn more time.
— Annie LowreyThe worst program is TANF, not the IRS
Lowrey names TANF (cash welfare) as the single worst program in government: it nominally exists to address child poverty, in practice it doesn't move the child poverty rate, its compliance burden is extremely heavy and punitive, and its $16.5 billion a year is essentially wasted. She offers a rule — programs aimed at high-income Americans are far better to deal with, safety-net programs are generally bad, and welfare is the worst. This yields a counterintuitive judgment: the quality of government service is inversely proportional to the political influence of its beneficiaries.
— Annie LowreyPandemic fraud is not something the time tax can prevent
Cowen raises a common objection: after deregulation in 2020 fraud exploded, so maybe the time tax is the necessary cost of fraud prevention. Lowrey rejects this outright: the fraud came mainly from criminal gangs and fake small businesses, and from the fact that the UI system itself runs on ancient COBOL infrastructure. Her mechanism is misaligned incentives — governors have no incentive to upgrade UI systems when unemployment is low, so the same conversation repeats every recession. The conclusion: building state capacity and agile digital systems prevents fraud; pushing the work onto individuals does not.
— Annie LowreyLarge-scale fraud is almost entirely on the supply side
Lowrey points out that Medicare and Medicaid fraud is almost entirely on the provider side: nursing homes, clinics and hospital systems defrauding the government. The ceiling on individual fraud is low, because the amount you can extract from the government is limited; to commit large-scale fraud you have to process a lot of paperwork. So her policy ordering is: reduce administrative overhead in other programs, concentrate enforcement resources on criminal organisations and legitimate providers, rather than loading burdens onto the 99.9% who don't commit fraud. She adds a piece of political economy: fraud narratives are often used to make programs look illegitimate, so voters tolerate cuts.
— Annie LowreyFifty hours per person per year, and highly unequal
Lowrey's quantitative estimate comes from the American Time Use Survey: total bureaucratic time is about 50 hours per person per year, close to an hour a week. She also concedes the number is understated — ATUS can't capture multitasking (waiting on hold while cooking counts as one thing), and time taxes are scattered across categories without being aggregated. Who pays the most? People entangled with the criminal justice system, immigrants, the poor. The rich can "skip the line" with staff and accountants, but there's one exception: renovating a house in San Francisco requires waiting 500 days and going through multiple rounds of public review no matter how rich you are.
— Annie LowreyAI can fill out forms, but it can't fill a urine test
Cowen asks whether AI agents could handle all of this within five years. Lowrey says probably, but raises two caveats: first, the users of the most burdensome programs are low-income, and whether they know about or trust these agents is unknown; second, much of the burden isn't paperwork at all — it's mandatory interviews, urine screening, data that's hard to collect. She gives SNAP as an example: benefits are calculated by household, but the government has no stable definition of "household," and you have to explain how you split staple food costs with the people you live with — a question AI would struggle to answer for you. She notes she was originally sceptical, but a civic technologist who has worked in this field for years firmly believes AI will help a great deal.
— Annie LowreyDisability insurance is already hard enough to apply for
Cowen presses with data: the number of people on disability insurance roughly quadrupled between 1990 and 2014, while workplace deaths kept falling — shouldn't the threshold be raised? Lowrey doesn't accept the inference. She offers two alternative explanations: population ageing and rising obesity-related chronic illness. Then she gives the counter-evidence: 5% of applicants have to submit more than 1,000 pages of medical material; 11,000 Americans die each year while waiting for a disability determination. She concedes she worries about both sides — about paying people who don't need it, and about denying people who genuinely do — so she doesn't think this is the moment to add more paperwork.
— Annie LowreyThe research moved her off her UBI position
Lowrey's previous book, Give People Money, was pro-UBI; here she says explicitly that the research findings are "sobering," and that she has updated her priors: the side benefits she expected cash to deliver didn't materialise. But her defence is that the point of giving people money is to give people money to buy things, and that rationale still holds; improvements in life falling short of expectations is not an argument against the policy. On AI and UBI, she admits she was extremely dismissive back then of the claim that "AI will force UBI," and she wouldn't go that far now, but if GDP and productivity surge while unemployment also rises, that would be very strange, and it would be time to start considering UBI-type policies.
— Annie LowreyIn their own words · checked verbatim
We could do this. Nine in 10 American households take the standard deduction and have fairly simple tax situations. We do not do this. The IRS is legally not allowed to do it.
Annie Lowrey0:58
If you’re a governor and you have a low unemployment rate, what governor is going to devote a huge amount of money to upgrading a state UI system?
Annie Lowrey7:11
To do big fraud, you have to be doing a lot of processing.
Annie Lowrey11:19
If you’re looking at what I think of as roughly as bureaucratic time total spent, it’s 50 hours per person per year.
Annie Lowrey13:25
I’m convinced that white-collar technology, labor-saving stuff, I’m convinced it’s all Jevons paradoxical. I think that basically anything that is designed like email to make your life easier will at some point make it worse.
Annie Lowrey21:49
In 5 percent of applications, the applicant submits more than 1,000 pages of medical paperwork. It’s a lot. 11,000 Americans die waiting on a disability determination each year.
Annie Lowrey32:24
I think it is noncorrupted, that their lives don’t improve as much as I would have thought, I think, isn’t necessarily a case against the policy.
Annie Lowrey39:41
I note that I think the best way to make friends is not to make yourself available for what other people need but to ask for help.
Annie Lowrey53:17
Figures
| Annual TANF spending | $16.5 billion | 1:02 |
| Share of US households taking the standard deduction | nine in ten | 0:58 |
| Estimated total bureaucratic time | 50 hours per person per year | 13:25 |
| Wait for a San Francisco home renovation permit | 500 days | 15:26 |
| Share of disability insurance applicants submitting more than 1,000 pages of material | 5% | 32:24 |
| Americans who die each year while waiting for a disability determination | 11,000 | 32:24 |
| Number of TANF programs | more than 100 | 35:34 |
| Number of Medicaid programs | 56 | 35:34 |
| Change in the number of people on disability insurance, 1990 to 2014 | roughly fourfold | 31:18 |
| Staten Island Ferry fare | $0 | 36:37 |
Glossary
- time tax
- The time cost imposed on individuals when government shifts administrative labour onto them; unlike a monetary tax, it cannot be offset by earning more.
- TANF
- Temporary Assistance for Needy Families, the US cash welfare program; Lowrey considers it the most administratively burdensome and least effective program in government.
- submerged benefits
- Benefits delivered through the tax code, such as the child tax credit, that recipients often don't realise they are receiving.
- Jevons paradox
- Efficiency gains increase total consumption rather than reducing it; Lowrey uses it to explain how labour-saving tools like email ended up adding to the task load.
- improper payment
- An erroneous payment caused by complex rules, distinct from deliberate fraud, and reducible by simplifying those rules.
How to listen
People working in public policy, government digital services or welfare products; engineers wondering whether AI can replace administrative processes; and readers who want to understand why the American welfare state is so inefficient.
The argument about voting difficulty from 36:25 to 38:28 — both sides talk past each other, low information gain.