2041 (ASI) ── 2041: Work, Ownership and Citizenship Under Advanced AI

*Intelligence Synthesis / Written 2026-10-10 / Horizon year 2041*

1. Thesis

In 2041 most American adults will still live mainly on wages, and US payroll employment will be higher than the 159.0 million of September 2026. But the college-educated office middle class built by the Third Industrial Revolution will be clearly smaller, most net new jobs will be in health care, care work, skilled trades and in-person services, and the central political question will be who owns and governs the most capable AI and how its output is shared. Our probabilities for 2041: (a) artificial superintelligence (ASI) exists and is widely deployed: 20%; (b) AGI-level systems match skilled professionals in most office work, but not ASI: 45%; (c) progress is slower, and AI remains a strong tool that needs close human direction: 35%.

2. Society in 2041

This section describes the base case (b). Every point is Inference. The evidence behind each point is named in brackets, and Section 3 gives the full figures. Section 6 explains how (a) and (c) differ.

Social structure

• A smaller office middle class. Jobs in information and finance are well below 2026 levels. (Inference. Evidence: in the year to September 2026 information lost 120,000 jobs and financial activities lost 107,000, while total payrolls grew by 496,000; BLS)
• Who is hit hardest. People who tried to start office careers between about 2024 and 2040. Some have moved into health care, teaching, trades and public service. The rest work in offices with fewer steps of promotion. (Inference. Evidence: computer systems design employment fell by 32,200 in a year; BLS)
• Concentrated ownership. A few firms that own models, chips, data centres and power contracts, and their shareholders, receive a large share of AI profits. The labour share of income is lower than in 2026. (Inference. Evidence: Investment and capital rose from 6.28% to 8.32% of AI news in 12 months)
• Regional divides. Counties that host data centres gain tax revenue but argue about electricity bills, water and cooling. Cities that depended on back-office work lose jobs. Places with large hospitals gain jobs. (Inference. Evidence: [C05])
• Families. Fewer households have two office earners. More combine one care or trade job with one office job. Robots and monitoring AI take over part of the care of elderly parents, but the burden on families does not disappear. (Inference. Evidence: the sector figures above; [C05])

Work

• Office work means directing and checking. Most office staff give instructions to AI systems, check the output and take responsibility for it. In fields where the law requires a licensed person's signature, such as law, accounting, medicine and architecture, people remain as the accountable role. (Inference. Evidence: Enterprise adoption is the largest theme in AI news, at 15.28% to 19.88% of coverage each month)
• Entry through apprenticeship. Firms hire fewer graduates than in 2026. New staff learn through rotations and supervised work, not routine junior tasks. (Inference. Evidence: falling employment in information and computer systems design)
• Where jobs grow. Health care, elder care, child care, electrical and construction trades (including grid and data centre work), restaurants and personal care. (Inference. Evidence: professional and business services as a whole still grew by 123,000 in a year, so the shift is gradual; [C05] on a grid that is slow to catch up)
• Robots. Humanoid and other general-purpose robots are routine in warehouses, factories and hospital logistics. In care homes they lift, carry and check rooms at night. Bathing, feeding and conversation are still mainly done by people. Few homes have one, because of price, safety and liability for accidents. (Inference. Evidence: Robotics stayed between 3.72% and 6.0% of AI news over 12 months, with no upward trend)
• Pay and hours. Output per hour is higher than in 2026, but median wages grow more slowly than output. Weekly hours are slightly shorter. More firms offer a four-day week, but it is not the general standard. (Inference. Evidence: Robert Allen's work in Section 4)

Attitudes and values

• Conditional trust. People use AI every day but do not trust the firms that run it. Who is responsible for accidents and errors is the strongest public concern. (Inference. Evidence: Safety and incidents rose from 6.8% to 11.32% of AI news in 12 months)
• The meaning of work. Jobs that involve direct contact with people, or making things by hand, gain social standing relative to pay. The belief that a university degree alone secures a stable position weakens. (Inference)
• Politics. The question moves from whether AI destroys jobs to who receives its gains. Cross-party coalitions form around taxing AI firms, who pays for electricity, and public AI infrastructure. (Inference. Evidence: Regulation and law rose from 5.04% to 7.52%)

Daily life

• Home. Most adults use an AI assistant every day and hand it government forms, taxes, medical appointments and household budgets. (Inference)
• Learning. Children receive individual AI tutoring. The role of schools shifts from delivering content to building the ability to check information, work with others and keep daily routines. (Inference. Evidence: Education and skills at 5.88% to 8.32% of coverage each month)
• Care and relationships. Many elderly and isolated people use AI companions every day. Rules to protect children and isolated people become stricter. (Inference. Evidence: Personal life and relationships at 9.0% to 12.16% of coverage each month)

3. What is happening now

• Total employment is growing while specific sectors shrink. In September 2026 total nonfarm payrolls were 159,044,000, up 496,000 on a year earlier; unemployment was 4.2% (down 0.2 points) and U-6 was 7.6%. Information employment was 2,739,000 (down 120,000), financial activities 9,082,000 (down 107,000) and computer systems design 2,355,400 (down 32,200) (BLS).
• Reported AI-attributed job cuts have risen but remain small. This site's count rose from 7,800 in 2025-Q1 to 107,900 in 2026-Q2 (17 cases), then 17,183 in 2026-Q3 and 4,430 so far in 2026-Q4 (a partial quarter). This is small next to total US layoffs and discharges (JOLTS: 1,641,000 in August 2026 alone).
• Coverage is shifting from adoption to safety and law. Enterprise adoption fell from 19.0% to 15.92% of AI news, while Safety and incidents rose from 6.8% to 11.32% and Regulation and law from 5.04% to 7.52%.
• Power and environment move together. Power and data centres and Environment and resources have a 12-month correlation of 0.8, with reports of a possible 32% rise in US electricity demand by 2030 and a survey in which about 70% of AI users worry about higher power demand [C05].
• Attention to robots is not growing, attention to capital is. Robotics fell from 5.72% to 3.72% of coverage, while Investment and capital rose from 6.28% to 8.32%.

4. What history suggests

The First Industrial Revolution (c. 1760-1840, Britain: textiles, steam, factories)

• What changed. The incomes of skilled hand workers such as handloom weavers fell sharply in the early nineteenth century, and work moved into factories. E. P. Thompson (1963, The Making of the English Working Class) documented this process and the Luddite protests of 1811-1816. Thompson (1967, 'Time, Work-Discipline, and Industrial Capitalism') showed how factories brought clock-time discipline into daily life. The 1851 census showed that more than half of the British population lived in towns.
• Wages. Robert Allen (2009) showed that from about 1800 to the 1830s output per worker grew while real wages barely rose, and the profit share increased: the 'Engels' pause'. Wages began to catch up with output only from the 1840s.
• Institutions. After the Factory Acts of 1802 and 1819, the Factory Act of 1833 set age and hour limits for child labour in textile mills and created factory inspectors. Children continued to work in mills long afterwards, but the system of regulation grew from this point. Institutions responded decades after the technology changed.
• What is similar and different. Similar: gains from higher output may reach the owners of capital first, with wages lagging. Different in three ways: speed (AI spreads through existing internet and cloud networks in years, not decades), which tasks (the people affected are college-educated office workers, not manual craft workers), and ownership (many mill owners held factories, while frontier AI is concentrated in a few firms).

The Second Industrial Revolution (c. 1870-1914: electricity, chemicals, steel, mass production)

• What changed. Paul David (1990, 'The Dynamo and the Computer') showed that productivity gains from electric motors did not appear in the statistics until the 1920s, decades after they entered factories, because factory layouts and the organisation of work had to be rebuilt. Henry Ford introduced the moving assembly line in 1913 and the five-dollar day in 1914.
• Work and education. Claudia Goldin (1990) showed how the growth of clerical work transformed women's employment. Goldin and Lawrence Katz (2008, The Race between Education and Technology) showed that the US high school movement of 1910-1940 supplied the skills the new jobs required on a broad scale.
• Daily life. Robert Gordon (2016, The Rise and Fall of American Growth) showed that household electricity, running water and appliances changed housework and time use over several decades.
• What is similar and different. Similar: rebuilding organisations takes time, and gains arrive late. The years to 2041 fall within the range of delay David described. Different: the Second Industrial Revolution created a large new category of work, clerical jobs, while AI targets exactly that office work. What the next large category of new jobs will be is not yet visible.

The Third Industrial Revolution (1970s onward: computers, internet)

• What changed. Robert Solow (1987) wrote that 'you can see the computer age everywhere but in the productivity statistics'. US productivity growth picked up in about 1995-2004. David Autor, Frank Levy and Richard Murnane (2003) showed that computers replaced routine tasks and complemented non-routine analytical and interpersonal tasks. Autor and David Dorn (2013) documented job polarisation: middle-wage jobs declined while high-wage and low-wage jobs grew. Daron Acemoglu and Pascual Restrepo (2020) found that industrial robots reduced employment and wages in local US labour markets.
• What is similar and different. Similar: losses concentrate in particular occupations and places while total employment holds up. Different: AI reaches the non-routine cognitive tasks that computers had protected.

Where history stops being a guide

None of the three revolutions replaced the people who invented, organised and set strategy. Machines complemented human judgment, and demand for people who exercise judgment grew. ASI removes that assumption. In scenario (a), the historical regularity that new kinds of work appear is no longer assured (Inference). For that reason, the description of scenario (a) relies on analysis of ownership and governance institutions, not on history.

5. Forecasts

Social structure

• Combined employment in information and financial activities in 2041 is at least 10% below September 2026 (11,821,000), that is, below 10.64 million. Probability: 60%. Reasoning: the two sectors lost 227,000 jobs together in one year to 2026; as Autor and colleagues showed, substitution starts with routine office tasks, and AI extends it to non-routine cognitive tasks (Inference). Check: BLS CES annual averages for 2041.
• The US nonfarm business labour share (BLS) in 2041 is lower than in 2026. Probability: 65%. Reasoning: ownership is concentrated, and in Allen's 'Engels' pause' output gains went first to profits (Inference). Check: compare the BLS Labor Productivity and Costs labour share for 2026 and 2041.

Employment and work

• Total nonfarm payrolls in 2041 exceed 159.0 million (September 2026). Probability: 75%. Reasoning: in none of the three revolutions did total employment fall over the long run, and payrolls are still growing in 2026; scenario (a) is the main reason this is not higher (Inference). Check: BLS CES annual average for 2041.
• The 2041 annual average unemployment rate is below 6.5%. Probability: 70%. Reasoning: the shift is gradual and unemployment was 4.2% in 2026, but rapid ASI deployment could push it above this level (Inference). Check: BLS seasonally adjusted unemployment rate.
• Health care and social assistance accounts for a larger share of total payrolls in 2041 than in 2026. Probability: 85%. Reasoning: the population is ageing, physical care is hard for robots to replace, and AI replaces office tasks before in-person care (Inference). Check: BLS CES health care and social assistance employment divided by total nonfarm payrolls.

Attitudes and values

• In a major national survey (Pew Research Center or Gallup) in 2040-2041, a majority of US adults say they are more concerned than excited about AI. Probability: 60%. Reasoning: the shares of Safety and incidents and of Trust and public opinion are rising, and concern persists when gains concentrate among owners (Inference). Check: Pew or Gallup surveys in 2040-2041.
• In at least one month of 2040-2041, 5% or more of respondents to Gallup's 'most important problem' question name AI or technology. Probability: 55%. Reasoning: in the First Industrial Revolution factory conditions became a political issue and led to legislation; ownership and distribution of AI will become politicised in the same way (Inference). Check: Gallup monthly series.

Daily life

• Fewer than 10% of US households own a humanoid robot in 2041. Probability: 80%. Reasoning: as Gordon showed, household appliances took decades to spread; price, safety and accident liability slow adoption in homes; attention to robotics is not rising now (Inference). Check: US consumer surveys (Census Bureau, industry associations, Pew).
• A majority of US adults use an AI assistant every day in 2041. Probability: 80%. Reasoning: software spreads through existing devices and networks, faster than electrification, which needed physical installation; Personal life and relationships holds about a tenth of AI coverage (Inference). Check: Pew or similar usage surveys.
• Average weekly hours of all private-sector employees (BLS CES) in 2041 are at least 0.5 hours lower than in September 2026. Probability: 50%. Reasoning: working hours fell over the long run after the Second Industrial Revolution, but through labour movements and law; shorter hours will not follow automatically from AI (Inference). Check: BLS CES 'average weekly hours of all employees, total private'.

6. Scenarios

Base case: AGI-level but not ASI (45%)

As in Section 2. The office middle class shrinks while health care, care work and trades grow. Ownership is concentrated in a few firms, and governments govern through licensing, audits and liability rules. Humanoid robots spread through warehouses, factories and lifting and carrying in care homes, but are rare in homes. Income is still mainly wages, supplemented by taxes and transfers. The citizenship issue is the right to an explanation of, and to challenge, decisions made by AI (Inference).

Alternative 1: ASI exists and is widely deployed (20%)

• Social structure. The few firms that hold ASI, and the governments that regulate them, gain great power. The gap between people who own AI capital, through shares or public funds, and those who do not, matters more than the gap in education (Inference).
• Work. ASI performs most scientific research, software and management decisions. Human work concentrates in signing as the accountable person, in-person care, jobs reserved for humans by law, and work valued because a human does it (sport, art, religion, parts of teaching). Humanoid robots become cheap and enter care homes and some households. Unemployment exceeds its usual range and working hours fall sharply (Inference).
• Ownership and governance. The most capable systems are treated as matters of national security and are controlled jointly by governments and a very small number of firms. Who governs them is the largest political question (Inference).
• Meaning, income and citizenship. Part of income comes from dividends on AI profits or public funds. Citizenship is defined less by working and paying taxes and more by the right to take part in governing AI. Work stops being the main source of self-respect, and community, family and faith gain weight (Inference). In this case history offers little guidance.

Alternative 2: Slower progress (35%)

• Social structure and work. AI is a strong tool but needs detailed human direction and checking. Information and finance employment falls, but less than in the base case. Graduate hiring recovers and the office middle class stays near its 2026 size (Inference).
• Robots and daily life. Robots stay mainly in warehouses and factories, with a limited role in care. Home AI is a useful tool, but people still handle many government and medical procedures themselves (Inference).
• Attitudes and governance. Politics focuses on electricity bills, data centre siting, copyright and privacy. Ownership is concentrated but less central to politics than in the base case. Polarisation of the kind seen in the Third Industrial Revolution continues (Inference).

7. Indicators to watch

Current values are refreshed from this site's data (BLS fetched daily) each time the page is built.

8. What would change this view

• By 2030, AI systems repeatedly produce major scientific discoveries on their own, verified in peer-reviewed papers. We would raise scenario (a) from 20% to at least 35%.
• By 2030, AI investment falls sharply and information and finance employment returns to 2026 levels. We would make scenario (c) the base case.
• By 2032, the United States enacts a law that distributes AI profits to citizens (a public fund or dividend). We would revise the forecasts on labour share and attitudes (2, 6 and 7).