Not a technology problem. A timing problem.

Narrow AI already writes code, diagnoses disease, and drafts contracts. Artificial General Intelligence is categorically different — a system capable of performing any intellectual task a human can perform, across domains, without retraining, with recursive self-improvement. The definitions remain contested. The forecasting consensus has moved sharply in one direction.

Sam Altman, CEO of OpenAI, puts AGI arrival at 2029. Elon Musk named 2026. Demis Hassabis of DeepMind assigns a 50% probability to 2030. Shane Legg, DeepMind's co-founder, gives 50% odds on "minimal AGI" by 2028. The forecasting platform Metaculus places the median at late 2027. Superforecasters, as of early 2026, put a 25% probability on AGI before 2029, and 50% before 2033.

Every one of those estimates lands inside the current or next election cycle in most democracies.

Democratic institutions do not move in years. They move in decades. The United States' unemployment insurance framework was designed in 1935. It was built for an economy where jobs disappeared slowly — sector by sector, over generations — allowing retraining, relocation, and adaptation. AGI does not operate at generational speed. It operates at the speed of a software update.

The gap between arrival date and institutional response capacity is not a gap. It is a chasm.

AGI is not here. The jobs are already going.

On SWE-Bench — the standardised coding benchmark — AI systems solved 4.4% of problems in 2023. By 2024, that number was 71.7%. Microsoft CEO Satya Nadella disclosed that 30% of company code is now AI-written. Over 40% of Microsoft's recent layoffs targeted software engineers. Big Tech reduced new graduate hiring by 25% in 2024. Among workers aged 20–30 in tech-exposed occupations, unemployment has risen almost 3 percentage points since early 2025.

This is before AGI. This is narrow AI, still accelerating.

The International Monetary Fund's 2024 assessment found that roughly 40% of jobs globally face meaningful AI exposure. In high-income countries, that figure rises to 60%. The World Economic Forum's Future of Jobs Report 2025 projects 92 million roles displaced by 2030 — and 170 million new roles created. A net gain of 78 million. That headline is cited constantly to dismiss the urgency.

Here is what the headline hides.

The 92 million displaced roles are concentrated: writers, analysts, coders, administrative workers, paralegal staff, radiographers, call centre operators. The 170 million new roles require different skills, in different locations, with different education requirements. The displaced analyst in Youngstown, Ohio does not automatically become a machine learning engineer in San Jose. The lag between displacement and re-employment is measured in years. During that lag, families collapse. Communities hollow out. Political structures destabilise.

That sequence — mass displacement, slow institutional response, political rupture — is the precondition for authoritarianism, not adaptation.

Dario Amodei, CEO of Anthropic, has said plainly that AI could eliminate half of all entry-level white-collar jobs within five years. Not eventually. Within five years.

The truck-driving sector alone represents 3.8 million jobs in the United States. Autonomous trucking projections estimate that falls to 2.3 million by 2030. A 38% reduction in per-mile costs creates an overwhelming commercial incentive. No government has a coherent plan for the 1.5 million displaced drivers.

What was welfare designed for?

Cyclical unemployment. People who lose jobs in a downturn and find new ones in the recovery. The architecture assumes labor market reallocation — that workers can be retrained and repositioned, that the economy will absorb them if given time.

That assumption holds when technology displaces sectors slowly. It does not hold when one general-purpose intelligence displaces across all sectors simultaneously.

The United States' unemployment insurance system requires prior work history. It pays for weeks, not years. It demands active job searching. It is conditional, temporary, and means-tested. None of those features function when the jobs being searched for no longer exist.

The Nordic model is more generous. But even Nordic-style systems assume workers can be retrained and repositioned into labor markets that still want human labor. When the AI performing a radiologist's role is also writing the code, analysing the law, and managing the hospital's operations, the retraining pathway ends before it begins.

There is a structural problem beneath the design problem. Welfare states are funded by payroll taxes tied to employment. As automation displaces workers, those tax revenues fall. Welfare costs rise. The fiscal base that funds the safety net shrinks at the precise moment demand for it explodes. The system is not just inadequate. It is mechanically pro-cyclical in the wrong direction.

What is needed is a floor independent of labor market participation. That is the definition of Universal Basic Income — an unconditional, regular cash payment to every citizen, regardless of employment status.

UBI is not a new idea. Thomas Paine proposed a citizen's dividend in 1797. The Belgian socialist Joseph Charlier outlined a "territorial dividend" in 1848. Milton Friedman — no socialist — advocated a Negative Income Tax achieving similar ends. The contemporary philosophical framework was built by Philippe Van Parijs in the 1990s.

The real-world evidence base is now substantial.

Finland, 2017–2018. The world's first nationwide randomised UBI trial gave 2,000 unemployed citizens €560 per month for two years, unconditionally. Employment increased slightly. Mental health improved significantly. Recipients reported higher life satisfaction, lower stress, and greater trust in public institutions than the control group. The welfare-dependency thesis failed to appear.

Kenya — GiveDirectly, 2018–ongoing. The world's largest and longest UBI study distributed funds to 23,000 people across 195 villages in three formats: a 12-year monthly payment, a 2-year monthly payment, and a lump sum. As of 2023, recipients did not stop working — they shifted from agricultural wage labour to self-employment and small business formation. Food security improved. Psychological wellbeing improved. The 12-year cohort showed the strongest investment behaviour: knowing income would continue allowed people to take risks that short-term recipients could not afford.

Stockton, California — SEED, 2019–2021. 125 low-income adults received $500 per month for two years, unconditionally. Financial stress fell. Mental health improved. Spending went primarily to groceries and utilities. Full-time employment among recipients increased compared to the control group.

Alaska Permanent Fund Dividend, 1982–present. Every Alaskan resident receives an annual payment from oil revenue invested in a sovereign wealth fund — between $331 and $3,284 per year. A 2024 study found the dividend reduced the number of Alaskans below the poverty threshold by 20–40%. Rural Indigenous poverty fell from 28% to below 22%. A 2018 paper found the dividend had "no effect on employment, and increased part-time work by 17 percent." In 2024, 81% of Alaskans said it improves their quality of life.

The work-disincentive objection is not a theory awaiting a test. It has been tested, repeatedly, across four continents, and repeatedly failed.

Three objections dominate the anti-UBI case. All three deserve precision rather than dismissal.

The cost objection. A $1,000-per-month UBI to every American adult would cost roughly $2.8–3 trillion per year. This is a real number. It is not, however, an argument against UBI. It is an argument about funding mechanisms. Andrew Yang's 2020 Freedom Dividend funded the payment through a 10% Value Added Tax on major corporations, projected to yield $800 billion annually, alongside consolidation of existing welfare programmes. The cost objection mistakes gross expenditure for net cost. The more direct question: is mass unemployment without an income floor cheaper — factoring in healthcare, mental health services, incarceration, addiction, political instability, and emergency welfare provision at scale?

The inflation objection. More money in circulation, the argument goes, produces higher prices. The evidence does not support this at the scale of existing pilots. Basic income transfers in India saw prices fall, not rise — assured demand created economies of scale and supply-side investment. Four decades of Alaskan dividends have not produced inflationary spikes. The inflation risk is theoretically coherent; it has not materialised in practice.

The work-disincentive objection. The 1970s Negative Income Tax experiments showed a 5% reduction in hours worked — concentrated in secondary earners, typically mothers who reduced hours to provide childcare. Full-time primary earners barely changed behaviour. Finland, Kenya, Stockton, Alaska: no mass work withdrawal. What the objectors consistently understate is the distinction between working less and not working. A parent who reduces hours to raise children is not a policy failure. That is the policy functioning.

What unites the opposition across the political spectrum is an implicit shared assumption: the labor economy will persist. The factory will reopen. The new sector will absorb the displaced. The market will solve it.

The people building the technology do not share that assumption.

Sam Altman has been publicly advocating for UBI since 2016 — when OpenAI began getting serious about the trajectory of AI capabilities. He funded a three-year, 3,000-person study through OpenResearch, giving 1,000 low-income people in Texas and Illinois $1,000 per month while 2,000 received $50. Published in 2024, the study found recipients worked slightly less, gained more flexibility and autonomy, and showed consistent improvements in wellbeing.

Altman also co-founded Tools for Humanity, which operates Worldcoin — a cryptocurrency distributing a monthly token allowance to any human who verifies their identity. The structure mirrors UBI precisely. More recently, he has floated "universal basic compute" — giving every person a share of AI compute capacity they can use, resell, or donate. Whether framed as cash, token, or compute share, the underlying argument is identical: the proceeds of AI must be distributed broadly, not captured by those who own the infrastructure.

Andrew Yang took this argument to the national political stage in 2020. His Freedom Dividend — $1,000 per month to every American adult — was the central plank of his presidential campaign. Yang argued that automation would eliminate one in three American jobs within twelve years. He cited factory closures, self-driving trucks, call centre workers replaced by automated voice systems. He lost the primary. He did not lose the argument.

Yang's campaign did something economic papers cannot. It made the AGI displacement argument legible to people who do not read economic papers. The political establishment called him a fringe candidate. Three years later, the fringe argument became Anthropic's official public position.

The loudest advocates for UBI share a specific characteristic. They built the technology. They know what it does.

This is the question beneath the policy question.

Work is not just income. It is identity, structure, social recognition, and daily purpose. Economists Anne Case and Angus Deaton documented "deaths of despair" — rising mortality from suicide, drug overdose, and alcohol-related illness — in communities experiencing economic displacement. These are not communities that ran out of money. In many cases, disability payments and family support provided subsistence. What they lost was work.

Viktor Frankl's concept of the existential vacuum describes exactly this condition: material needs met, existential needs unmet. A guaranteed income addresses the material dimension. It does not address the existential one.

UBI advocates who speak only about cash transfers are solving one dimension of a multi-dimensional problem. The transition to a post-labor economy requires not just income redistribution but a cultural reorientation around what constitutes meaningful contribution. Some argue that this reorientation is impossible — that humans require scarcity and competition to find meaning. That is an empirical claim, not a moral truth.

Finland's UBI recipients did not descend into nihilism. Alaska's dividend recipients did not withdraw from their communities. The evidence from every pilot suggests that when existential threat is removed, people pursue connection, creativity, and contribution more readily, not less.

The question is not whether people will find meaning without paid work. The question is whether any society will design the conditions that make meaningful activity possible for everyone.

The argument for UBI before AGI is not primarily economic. It is structural.

Policy infrastructure cannot be built in crisis. The United States spent twelve years designing and implementing the Affordable Care Act. The UK's Universal Credit took a decade to roll out and remains contested. A UBI at national scale requires legislative consensus, administrative infrastructure, funding mechanisms, and public legitimacy. None of those can be assembled in the months between AGI arrival and mass unemployment.

Once AGI displaces labor at scale, two political dynamics activate simultaneously. The displaced majority demands immediate income support. The productive minority — those owning AI infrastructure — resists redistribution with the full force of concentrated wealth. The political conditions for designing a fair, adequate UBI will be at their worst precisely when they are most needed.

Acting before AGI changes this calculus entirely. A UBI established during near-full employment, when the fiscal base is intact and the threat is visible but not yet catastrophic, can be designed properly. It can be tested, adjusted, and legitimised. It does not have to be improvised under emergency conditions against a destabilised democracy.

There is also a signal function. A society that implements UBI before AGI is declaring a value: the proceeds of intelligence — human or artificial — belong to everyone. Human worth is not contingent on market productivity. The social contract does not terminate when the economy no longer requires your labour.

A society that waits until the crisis hits is making a different declaration. It is saying that human welfare is a negotiating position, not a baseline. In that negotiation, the party that owns the AGI holds all the leverage.

The machinery of control is economic before it is anything else. Decoupling human survival from market productivity is not a soft proposal. It is the precondition for a functioning democracy, for meaningful AI governance, for any social contract worth the name in a world where the machine can perform your job better than you can.

The window for doing this well is measured in years, not decades.