By Thorsten Meyer | November 1, 2026 | ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Here is an offer. You can leave your job tomorrow and keep getting your salary, guaranteed, with your health cover, pension and paid leave intact. Would you take it? Now make it 120% of your salary. Most people say yes long before that. But in a new survey of 8,234 working-age adults in the UK, the US, Denmark and India, about one in five said no, even at 120%.
That number sits at the center of a paper posted to arXiv on 29 October 2026: “What is Work For? Labor, Politics, and Meaning in the Age of AI”, by Julian Jacobs of the University of Oxford and Google DeepMind. It surfaced on Hacker News on 31 October. Its question sounds philosophical, but it is about money and politics: if AI takes over more work, can money replace what a job gives people? The author’s answer is no, not fully. I read the full paper. Here is what it shows, where it is strong, where it is thin, and what it means for companies and for the people who work in them.
Key takeaways
- Pay is only part of it. Of the workers the paper could classify, 33% are mainly driven by the work itself (purpose, learning, contribution), 23% by colleagues and belonging, 17% by status and recognition, and 26% by pay and benefits. In the paper’s own figure, 73.7% put a non-money reason first.
- It is not a luxury of the well-paid. The share driven by the work itself is 32%, 33% and 34% for low, middle and high household incomes. It is about a third among craft and trade workers too.
- Most would still quit for a guaranteed income. By 80% of salary, a slight majority would leave; at 100%, around three in four. But about 20% would stay even at 120%, and that holdout share hardly moves with income.
- The basic income warning. The author argues a basic income “may indeed be necessary” in an AI labor shock, but “could ultimately fail to compensate workers” for what work gives beyond money. That is his interpretation, built on stated answers, not on observed behavior.
- The effects are real but small. After controls, people driven by purpose ask for about 1.9 to 5.8 percentage points more guaranteed income to quit than pay-driven people, depending on the method. At a $60,000 salary, that is roughly $1,100 to $3,500 a year.
- The fear sits on the desk. 40% of white-collar workers think AI is at least somewhat likely to replace their job within ten years, against 29% of manual and service workers.
- The politics are the thinnest part. Purpose- and community-driven workers turn more pessimistic when they expect AI to hurt jobs in their industry. But in the paper’s own tables, the groups do not differ significantly in populist vote intention, and the paper says itself that its analysis is “purely relational, not causal.”
Products Worth Considering
Apple Watch SE 3 GPS 44mm Smartwatch
Anker SOLIX S2000 Portable Power Station, 2,010Wh, 1,500W
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
The paper in brief
- Title and author: “What is Work For? Labor, Politics, and Meaning in the Age of AI”, Julian Jacobs, University of Oxford and Google DeepMind. Single author.
- Status: a preprint, version 1, submitted to arXiv on 29 October 2026 in General Economics (econ.GN), 39 pages and 17 figures. It has not been peer-reviewed. Numbers could change in later versions.
- Data: an original online survey of 8,234 working-age adults (Denmark 2,032, India 2,072, UK 2,052, US 2,078), fielded in June and July 2026 by the polling organization More in Common, with national quotas on age, gender, education and region.
- Funding: the paper says the survey was “supported by Google DeepMind”, the author’s employer, and that the views do not necessarily reflect Oxford’s or DeepMind’s. I found nothing in the analysis that favors any company, but readers should know who paid for the data.
- Method: a nine-item battery on what matters in a job, sorted into four “Work Values Profiles”; an auction-style question on how much guaranteed income would make people quit; a question on how they would spend their time; and questions on AI threat and politics. The models are cross-sectional regressions with controls for income, occupation, age, gender, education, part-time work and country.
The thought experiment: would you quit for a guaranteed salary?
The core of the paper is a simple auction. Respondents were offered a guaranteed basic income at 60% of their current salary, then 70%, and so on in steps of ten up to 120%, with healthcare, pension and paid leave kept. At each step: would you voluntarily quit?
The answers, pooled across the four countries:
- By 80% of salary, a slight majority would leave.
- By 90%, around 65%.
- At 100%, around 75%.
- At 120%, about 20% still refuse. The paper calls them holdouts.
The holdout share is strikingly flat across incomes: 21% for low-income, 18% for middle-income and 22% for high-income workers. It is 21% for white-collar and 17% for blue-collar workers. By country, it is 16.0% in the UK, 17.9% in both the US and Denmark, and 28.8% in India, where the online sample is younger, more urban and far more educated than the country as a whole.
Among the people who did accept an offer, the median quit point was 70% of salary, the average 76%. In other words, most respondents would accept quite a lot less than their pay to stop working, if the money were guaranteed. That is the first surprise, and it cuts against a romantic reading of the paper: for most people, work is something they would leave.
The second surprise is the size and spread of the minority who would not.
WHAT WORK IS FOR, ACCORDING TO WORKERS
Three in four put something other than pay first
The paper sorts workers by what they rate most important in their job. In its own figure, 73.7% of classified workers put a non-money reason first.
The work itself: purpose, learning, contributing to society.
Pay and benefits come first.
Belonging, colleagues, meeting people.
Recognition and prestige.
Pay-first share, by country
Share of classified workers whose top value is pay and benefits
Meaning is not a luxury
Share driven by the work itself (intrinsic), by household income third
Also about a third in craft and trade jobs (33%) and in services and sales (31%).
Shares of the 6,224 working respondents the paper could classify; 2,010 respondents (non-working, or no clear top value) were left out. 2,000 tied answers were assigned by a tie-break rule. Satisfied with their job: 81%.
Source: Jacobs, What is Work For?, arXiv:2610.39375v1, 29 Oct 2026 · preprint, not peer-reviewed · survey of 8,234 working-age adults in the UK, US, Denmark and India by More in Common, June–July 2026, supported by Google DeepMind.
Powered by Thorsten Meyer AI
Four reasons to work
To sort people, the survey asked how important eight things are in their job: pay, non-wage benefits, a sense of purpose, learning new skills, contributing to society, belonging, meeting people, and social status. (A ninth item, “having time and space to be myself”, was left out of the index.) Each worker was assigned to the domain they rated highest:
- Intrinsic, 33%: the work itself matters, for purpose, learning or contribution.
- Transactional, 26%: pay and benefits come first.
- Community, 23%: belonging and the people.
- Status-driven, 17%: recognition and prestige.
The mix differs by country in a way that fits comparative politics. The US sample has the most pay-first workers (38% of those classified) and the UK 32%. In Denmark, with its generous safety net, the share drops to 19%, and in the Indian sample to 15%. The paper reads this as national context shaping how people relate to work. That is plausible, but the survey cannot prove it.
The finding I consider most important is about who holds these values. Being driven by the work itself is almost evenly spread: 32% of low-income, 33% of middle-income and 34% of high-income workers. It shows up in 38% of professionals, 34% of managers, 31% of service and sales workers and 33% of craft and trade workers. In a model with controls, high household income does not significantly predict any profile, and occupation and party preference show virtually no association either. Meaning, in this data, is not a perk of the corner office.
Satisfaction is high, too: 81% of respondents say they are satisfied with their job, and Danish workers top the list at 82%.
What people would do with the money
The survey also asked: if you were made redundant and received a basic income, how would you spend your time? Respondents could pick several of nine activities. The profiles split sharply:
- Pay-first workers mostly retire into leisure: 79% picked at least one leisure option, such as rest, hobbies or travel. Only 11% would look for another full-time job.
- Status-driven workers keep climbing: 55% would continue in paid work or start a business. The paper reads this as using a basic income as a floor to reach higher-status work. It also explains why this group has the lowest quit point (a median of 60% of salary among those who accepted an offer).
- Purpose-driven workers turn to others: 57% would volunteer or care for family.
- Community-driven workers blend both: 51% would volunteer or care, and many would also look for work.
All of these differences are statistically significant after controls. Again, these are stated plans, not observed behavior.
How big is the “meaning premium”?
Purpose-driven workers need more to quit. Among those who accepted an offer, their median quit point is 80% of salary, against 70% for pay-first and community workers and 60% for status-driven ones. And 23% of them refuse even at 120%, against 14% of pay-first workers.
After controls, the most conservative estimate puts the purpose premium at 1.85 percentage points of salary. The paper stresses that this is a lower bound, because the strictest model drops the holdouts, and those are disproportionately purpose-driven. Keeping them in, under different assumptions about what they would accept, the premium rises to 4.58 points (holdouts counted at 120%) and 5.77 points in a statistical model built for capped answers. The paper translates this, at an approximate US median full-time salary of $60,000, into about $1,110 a year at the low end and $3,462 a year for the capped-answer model.
Is that a lot? The author points out that even the smallest version is larger than the difference between the bottom and top third of household incomes in the same model. Fair. But the models explain very little of the overall variation (an R² of about 0.05), so most of what drives a person’s quit point lies outside these profiles.
Two more patterns deserve a mention. The premium grows with age: it is small and not significant among 18 to 29-year-olds and largest among workers aged 55 and over. And it does not seem to be just about future pay: young workers, who have the most pay growth ahead, have the lowest quit points, and people confident they could find an equally good job ask for less, not more.
THE GUARANTEED-INCOME TEST
Most would quit for a guaranteed salary. One in five would not
Respondents were offered a guaranteed income from 60% to 120% of their salary, with health cover, pension and paid leave kept, and asked at which step they would quit.
Who holds out, by profile
Refuse every offer up to 120% of salary
Median quit point among those who accepted an offer: intrinsic 80%, transactional and community 70%, status-driven 60% of salary.
What they say they would do instead
If made redundant with a basic income (several answers allowed)
How big is the meaning premium?
Extra guaranteed income intrinsic workers ask for to quit, vs. transactional workers, after controls
At an approximate US median salary of $60,000, the paper puts this at about $1,110 to $3,462 a year. Stated answers in a survey, not observed behavior; the models explain about 5% of the variation (R² ≈ 0.05).
Source: Jacobs, What is Work For?, arXiv:2610.39375v1, 29 Oct 2026 · preprint, not peer-reviewed · survey of 8,234 working-age adults in the UK, US, Denmark and India by More in Common, June–July 2026, supported by Google DeepMind.
Powered by Thorsten Meyer AI
AI fear sits on the desk, not the factory floor
The survey asked how likely it is that AI or automation will replace your job within ten years. In white-collar occupations, 40% said at least somewhat likely; in manual and service jobs, 29%. The paper notes that this lines up with recent studies of where AI use and task exposure concentrate. Earlier waves of automation hit routine manual work first. This one, people feel, is aimed at the office.
Younger workers (41 and under) perceive a much higher risk than older ones, across every profile. The paper adds a caution: that may reflect greater AI awareness or general anxiety, not a correct forecast.
One finding is easy to misread. The more people believe AI will take their job, the less guaranteed income they demand to quit: about two to four percentage points less per standard deviation of perceived threat, and the drop is steeper for purpose- and community-driven workers. The author is clear that the direction is unknown. Pessimists might both fear AI and accept less, or fear might make people take a safe offer. Either way, it is a correlation.
The politics: where AI anger could come from
Political scientists have long puzzled over why automation shocks feed support for anti-establishment parties even where welfare is generous. One answer in the literature is lost status, which money does not restore. This paper tests a broader version: if work gives meaning, community and status, then a cheque alone cannot buy off the anger.
What it finds is mixed:
- At baseline, the pay-first group is the most negative. They expect AI to harm their country’s political system the most (+0.19 standard deviations, against roughly zero for purpose- and community-driven workers and −0.16 for status-driven ones). In the UK module, they also sit furthest to the economic left and are more inclined than purpose-driven workers to want broken institutions torn down.
- But when the threat gets concrete, the others react harder. Among workers who expect AI to harm job quality and job numbers in their own industry, purpose-driven (+0.98 SD) and community-driven (+0.96 SD) workers are more pessimistic about AI’s political impact than pay-first workers (+0.84 SD). In the UK-only module of about 1,500 classified workers, expecting industry harm goes with a stronger belief that the system is “rigged”, most clearly for status- and purpose-driven workers.
- Vote intention does not follow. My reading of the paper’s tables: in the main four-country model of right-wing or populist vote intention, none of the differences between the profiles, in levels or in how they react to AI threat, is statistically significant. In the appendix version that uses expected harm to one’s industry, purpose-driven workers who expect harm are, if anything, less likely to say they would vote for such a party. The pessimism shows up in attitudes, not in stated votes.
The author calls this “potential evidence” and writes that “more research is needed”. I agree, and would put it more bluntly: the political chapter is a promising hypothesis, not a finding.
Where the paper is strong
- The question is the right one, and it is rarely measured. Much of the AI-and-jobs debate is about income: wages, transfers, taxes. This paper puts numbers on what money may not replace, across four very different economies.
- The sample is large and cross-national, with separate checks for the three Western countries alone, which matters because the Indian sample is skewed.
- It tests the obvious objections. The purpose premium survives controls for job satisfaction and for confidence in finding another job, alternative cut-offs for the profiles, and several ways of handling the people who refused every offer.
- It is honest about its limits. The paper has its own limitations section, flags the Indian sample repeatedly, and warns that survey answers can diverge from real decisions.
- The most policy-relevant result is robust in its simplicity: non-money reasons to work are found in about equal measure across income groups and occupations.
Where it is thin
These are my own reading of the paper, marked as such:
- Stated, not revealed. Nobody was actually offered the money. As the paper puts it: “No respondent actually left their job as a consequence of participating in the survey.” The paper cites the economist Jerry Hausman’s critique of such hypothetical valuations, in an article titled “Contingent Valuation: From Dubious to Hopeless”. Real cash experiments, which I covered in Will Robots Make Us All Rich?, measure what people do with money, not what they say they would do.
- The profiles depend on a tie-break. 2,000 workers, 32% of those classified, rated two or three domains equally high. They were assigned to the domain that is rarest at that level, which often puts people who rated pay at the top into a non-money group. The author argues this makes the comparison conservative and shows robustness checks without the rule. Still, the headline split of 33, 26, 23 and 17% is partly a product of that rule, and 882 working respondents who rated everything the same, or nothing as important, were left out.
- Small effects, low explanatory power. Premiums of two to six points of salary are meaningful, but the models explain about 5% of the variation in quit points.
- One thing work gives is not measured. The paper’s model folds “daily temporal structure, habit, or workplace amenities” into the error term. Routine, a reason to get up, a place to be: these are exactly what many people miss after losing a job, and the survey does not capture them as a separate value.
- Political participation, as such, is missing. The paper measures attitudes and vote intention. It does not measure whether people vote, organize or join unions, which is what “participation” usually means.
- Correlation throughout. The author writes: “I insist again that this analysis is purely relational, not causal.” Values may follow from good jobs as much as the other way round; the paper itself suggests the profiles describe a worker’s “realized relationship” with the current job.
- Single author, preprint, one funder. None of that makes it wrong. It means the work has not been through peer review, and the data was funded by an AI lab with an interest in how the AI transition is perceived.
THE PAPER, WEIGHED
Strong on the question, thin on the politics
What the evidence in the paper supports, where it is weaker than the headline, and what it does not measure at all.
Where the paper is strong
- strongLarge original survey: 8,234 adults in four very different economies.
- strongNon-money reasons to work found at every income level and in blue- and white-collar jobs.
- strongThe purpose premium survives controls for job satisfaction, re-employment confidence and alternative cut-offs.
- strongOwn limitations section; separate checks without the skewed Indian sample.
Where it is thin
- thinStated, not revealed: nobody was actually offered the money.
- thinProfiles rely on a tie-break for 2,000 workers (32% of those classified).
- thinSmall effects (about 2 to 6 points of salary); models explain about 5% of the variation.
- thinPolitics: attitudes shift, but populist vote intention shows no significant gap between the groups.
- thinCross-sectional: the author calls it “purely relational, not causal.”
- missingDaily structure and routine are not measured as a value.
- missingPolitical participation (voting, organizing) is not measured.
Status: arXiv preprint, version 1, single author (University of Oxford and Google DeepMind), not peer-reviewed. “Strong” and “thin” are Thorsten Meyer AI’s reading.
Source: Jacobs, What is Work For?, arXiv:2610.39375v1, 29 Oct 2026 · preprint, not peer-reviewed · survey of 8,234 working-age adults in the UK, US, Denmark and India by More in Common, June–July 2026, supported by Google DeepMind.
Powered by Thorsten Meyer AI
So, what is work for?
The brief question in the title deserves a direct answer. Here is mine, as opinion, informed by the paper.
Work is for income, first. Most respondents would take less than their pay to leave if it were guaranteed, which tells you how many people work mainly because they must. Any AI transition that ignores income will fail on its own terms.
Work is also for meaning, company and standing, for most people and at every income level, if this survey is right. Those three are not interchangeable. Meaning seems to stick to the role itself. Community can partly be rebuilt outside work, in families, clubs and neighborhoods, which may be why community-driven workers ask for a smaller premium. Status can be chased elsewhere, which is why status-driven workers would happily take a basic income and start something new.
And work is for structure and, for some, a public role: a reason to leave the house, a place in the social order, a stake in how the economy is run. The paper hints at the second and does not measure the first.
If that is right, then the policy debate about AI and jobs has been asking half the question. “How much money do people need?” matters. “How do people keep contributing, belonging and being recognized?” matters too. The author’s own conclusion points there: an effective response “may need to reckon with the simultaneous shocks to both income and the deeply-embedded social, cultural, political, and normative value that work provides.”
How this fits with this week’s debates
On the same day the paper surfaced, I asked whether automation has become an imperative for governments and companies, in 25,000 civil servants and 60,000 exit offers on the same day. That article is about the push to automate. This paper is about the other side of the ledger: what the people being automated lose.
The basic income debate, from Larry Ellison’s “universal high income” to South Korea’s rural pilot and the US, Finnish and Kenyan trials, is in Will Robots Make Us All Rich?. This paper does not test a basic income. It tests how people say they value work against one.
And when companies blame layoffs on AI, as covered in Is AI really behind the layoffs?, the framing matters more than it seems. This paper suggests that the belief that AI will hurt one’s industry goes with sharper political pessimism. A company that says “AI did it” when budgets did may be feeding that belief.
What this means for companies
- Don’t price automation at the wage line alone. In this sample, roughly three in four workers put something other than pay first. If you automate away the parts of a job that give it purpose, you may keep the payroll cost and lose the reason people stay. That is my inference, not a tested result.
- Ask which tasks people value before you automate them. Let AI take the drudgery, not the craft. A short survey of your own teams, using the paper’s eight items, is cheap.
- Plan for lost company, not just lost tasks. If AI tools reduce daily contact, community-driven employees lose something real. Shared work, mentoring and in-person time are design choices.
- Recognition is a lever. Status-driven workers are the most mobile in this data. Visible expertise, titles that mean something and credit for AI-assisted work cost little.
- Severance replaces income, not belonging. When roles go, transition support that offers new work, retraining, alumni networks or paid time for volunteering speaks to what the paper says people would actually miss.
- Say what is true about AI and jobs. The political chapter is thin, but the direction is plausible: people who believe AI will damage their industry turn more pessimistic about the system. Overselling AI as the reason for cuts carries a cost beyond the press cycle.
- Older staff may be the most attached. The purpose premium is largest among workers aged 55 and over. Treat their knowledge, and their exit, accordingly.
What this means for workers and users
- Ask yourself the paper’s question. At what share of your salary would you quit, and what would you miss? The answer tells you whether your job is mainly income, meaning, people or standing, and what to protect if it changes.
- Diversify where meaning comes from. If work is your main source of purpose, community or status, build at least one more outside it. That is good advice with or without AI.
- A floor can be a springboard. In this survey, status-driven workers saw guaranteed income as a way to start something new. Savings, training and side projects play the same role today.
- Don’t let fear set your price. People who expected AI to take their job were willing to accept less to leave. The paper cannot say why. Before you accept an offer, check the facts about your role, not the headlines.
- Read survey numbers for what they are. “One in five would stay at 120%” is a stated answer in an online poll, not a forecast of what will happen.
What to watch next
- Revisions and review. Whether a later arXiv version or a journal publication changes the numbers or the profile method.
- Behavioral evidence. Cash-transfer studies that track what people actually do, including volunteering and care, not only paid hours.
- Causal work. The author calls for research that can identify cause and effect between AI threat, work values and politics.
- Company practice. Whether employers start measuring the non-money value of roles before automating them.
Honest limits
- This article is based on version 1 of a preprint, read in full on 31 October and 1 November 2026. It has not been peer-reviewed.
- I report the paper’s numbers as published and did not re-run its analysis. Where my reading differs from the author’s (the vote-intention result, the measurement gaps), it is labelled.
- Some figures in the paper are given as approximations (“around 65%”); I kept them approximate.
- The dollar values are the author’s translation at an approximate US median salary of $60,000; they do not apply to other countries or salaries.
- Hacker News was the tip, not a source. No outlet coverage of the paper was used.
- Nothing here is career, legal or financial advice.
Sources
- Julian Jacobs, What is Work For? Labor, Politics, and Meaning in the Age of AI, arXiv:2610.39375v1 [econ.GN], submitted 29 October 2026 (PDF). Survey data collected with More in Common, supported by Google DeepMind. All figures in this article come from this paper.
- Cited within the paper and quoted here only as the paper cites it: Jerry Hausman, “Contingent Valuation: From Dubious to Hopeless”, Journal of Economic Perspectives 26(4), 2012.
- Related on Thorsten Meyer AI: Is automation now an imperative? (31 October 2026), Will Robots Make Us All Rich? (31 October 2026), Is AI really behind the layoffs? (31 October 2026).
