The Day AI Starts Paying Humans

What happens to growth, work, and inequality when the fourth factor finally gets its check — and the road from here to there

Returns to Data — Article V, the final paper of a Vector1 Research Program

By Brian Curry · Vector1 Research

This is the closing paper of Returns to Data*. The series established that data is a factor of production (I); that its share of AI value is roughly 45% while its payment is roughly 2%, a $34-billion-a-year gap (II); that the technology to meter and pay individual contributors now exists (III); and that the claim is really two claims — a back-debt of $100–700 billion for data already taken, and an ongoing flow — best settled by a stack of six mechanisms rather than any one (IV). One question remains, and it is the one every skeptic reaches for last:* what happens to the economy if we actually do this? This paper answers with the best evidence available — and then draws the map from here to there.

1. The Last Objection Standing

Every argument in this series has faced the same final redoubt. Concede that data is a factor of production. Concede the share, the gap, the meter, the back-debt, the stack. There remains one objection with genuine force, because it gestures at catastrophe: you cannot pour tens of billions — eventually hundreds — into households without breaking something. People will stop working. Prices will surge. Investment will flee. The cure will be worse than the wedge.

These are empirical claims, and the unusual thing — the thing this paper exists to show — is that we do not have to speculate about them. A natural experiment has been running for forty-four years: a jurisdiction that declared a commonly-owned resource, captured a share of the windfall from its extraction, and mailed every resident a check, every year, forever. Alaska is the data dividend with the serial numbers filed off, and economists have studied it exhaustively. The macroeconomics of paying the fourth factor is not a leap into the dark. It is one of the better-lit rooms in empirical economics.

2. The Fourth Income Stream

Start with what changes for a household. Today, household income has two private engines — wages and capital returns — and most households own almost none of the second, which is precisely why AI-driven displacement is frightening: when the wage engine sputters, there is nothing else. The stack of Article IV adds a third private engine, and its size is set by the machinery this series built — the capture ratio Φ, data’s share γ̂, and the household’s attribution weight:

The deep property of this stream is where it comes from. It is generated by the same machine that threatens the wage stream. As AI substitutes for labor, AI revenue grows; as AI revenue grows, so does γ̂·Y; and the dividend rises because the displacement is happening. The fourth income stream is, structurally, a hedge against the third industrial transition — the first transfer in history that automatically scales with the disruption it cushions:

Figure 1. A scenario for the median household’s income composition as the stack phases in (Φ → 0.6 by 2036) along the revenue path of Article III. As AI compresses the wage share, the data dividend expands to replace it — funded by the same machine doing the compressing.

3. Will Anyone Still Work? The Alaska Answer

The oldest objection to unconditional payments is the income effect: pay people regardless of work and they will work less. It is a coherent theory. It is also, on the best evidence we have, approximately false at the scales that matter.

The cleanest test is the Alaska Permanent Fund Dividend — universal, unconditional, permanent, exactly the structure of a data dividend. Jones and Marinescu’s synthetic-control study compared four decades of Alaskan labor markets against matched states:

Employment effect: zero. Part-time work rose 17 percent — people at the margins joining the workforce flexibly, not leaving it — and the authors’ explanation is the general-equilibrium effect the simple theory misses: dividend money gets spent, spending is demand, and demand hires. The wider literature rhymes: negative-income-tax experiments found small effects; lottery studies found small effects; the Stockton RCT found none; the OpenResearch $1,000-a-month study found a modest two-point reduction concentrated in caregiving and education — time use most people would call investment, not idleness.

Figure 2. Estimated employment effects of unconditional cash programs. The only permanent, universal program ever run — the closest analogue to a data dividend — sits at zero. The catastrophist case rests on a theory the evidence keeps declining to confirm.

And the data dividend has a property no UBI shares, which should make its labor effects better than Alaska’s: at the attribution layer, it is not unconditional at all. It pays for contribution. Article III showed the meter rewards rare, expert, verified data — so the marginal incentive of the micropayment layer is to produce valuable knowledge, not to withdraw from production. A welfare check and a royalty check have different signs.

4. The Inflation Question

Will the dividend ignite prices? Decompose it, as Article IV insists, into stock and flow. The flow — Φ·γ̂·Y arriving annually — is bounded in its price impact by its sheer scale against the economy: even at full early-stack capture, the marginal-propensity-to-consume-weighted injection is a fraction of a tenth of a percent of GDP — within the Federal Reserve’s rounding error:

Two structural features push even that small number down. First, the dividend is not money creation — it is a transfer of existing revenue from firms (low marginal propensity to spend domestically) to households; redistributions move demand around far more than they add to it. Second, the stock — the $100–700B back-debt — is the part that looks alarming and is engineered not to be: paid substantially in equity (Article IV’s settlement design), it arrives as wealth, not as a spendable lump. Norway has run this exact play for thirty years — a sovereign windfall held as assets, with a fiscal rule metering the drawdown — and remains one of the world’s least inflationary rich economies. The design already exists; the stack simply inherits it.

5. The Growth Surprise

Here is the result that inverts the entire framing of “redistribution versus growth,” and it follows from the series’ own physics. Recall the data wall (Article I): the free supply of high-quality human data is exhausting, and as compute compounds against a fixed data stock, the marginal product of data rises while model progress strains. The binding constraint on AI growth is, increasingly, fresh human data.

Now ask what a functioning payment system does to that constraint. A price on data is a supply signal. The attribution layer pays most for exactly what is scarcest — rare expertise, verified observation, low-resource languages, novel human experience — and supply curves slope up:

Paying the fourth factor loosens the fourth factor’s own bottleneck. The dividend is not a drag coefficient on the AI economy; it is the procurement system for its scarcest input — the difference between strip-mining a fixed deposit and cultivating a renewable one:

Figure 3. AI-sector output under two regimes. Unpaid data (dashed): the wall binds and growth decays as the free stock exhausts. Paid data (solid): supply responds to price and growth compounds. The long-run growth case for the stack is stronger than the fairness case — and the fairness case was already strong.

This is the deepest answer to “the firms will flee”: flee to what? Every jurisdiction faces the same wall. The first economy to build a functioning data market is the economy where the scarce input keeps flowing — the same way the economies that organized resource markets, rather than exhausting commons, won the last century.

6. The Inequality Arithmetic

The distributional effect needs little modeling, because a uniform payment into an unequal distribution compresses it mechanically — every dollar of universal dividend is worth more, proportionally, at the bottom:

The precedents again oblige: the Alaska dividend measurably reduced poverty, with the largest effects among Alaska Native communities and the elderly — and, the detail worth a paper of its own, 81 percent of Alaskans say the dividend improves their lives. A payment of two thousand dollars purchased something money rarely buys: near-universal legitimacy for the system paying it. That is the real macroeconomic stake. The alternative to the stack is not the status quo; it is an AI economy whose gains concentrate in the narrowest ownership base in industrial history, contested by a public that knows the models were built from its own words. No growth model prices political rupture, but every investor should.

7. The Road From Here

The series closes with the map. Each step uses machinery that already exists; every date is a scenario midpoint, not a prophecy; and the sequence follows Article IV’s feasibility logic — litigation prices the debt, settlement converts it to equity, the flow layers phase in on infrastructure the settlement builds:

Figure 4. The transition, mapped. The litigation wave (running now) prices the back-debt case by case; a statutory settlement converts the stock claim into equity and seeds the fund; licensing and the training royalty lift the flow immediately; attribution micropayments switch on as the metering matures; and the fund pays its first true data dividend in the early 2030s. From Φ = 0.04 to the first check: roughly a decade, using no institution that hasn’t already worked somewhere.

What should each actor do now? Legislators: name the debt — the single highest-leverage act available is a statutory settlement framework that prices the stock claim, because every flow mechanism inherits its registry and its legitimacy. The labs: move first — the participation constraint (Article IV) says paying beats litigating, and the data wall says the first lab with a real procurement market for fresh human data corners the scarcest input. Creators and contributors: organize and register — every mechanism in the stack pays organized claimants first and unorganized ones late or never. Economists: the fourth factor is the most under-measured object in the discipline; Articles II and III are estimates begging to be beaten with better data.

8. Closing the Ledger

Five papers ago, this series began with an anomaly: a $3,000 book and a $380-billion company. It ends with an identity. The industrial economy learned — slowly, expensively, but durably — to pay its factors: capital earns its return, labor earns its wage, and the system’s legitimacy rests on the bargain being visible in every paycheck. The AI economy currently runs on three factors paid and one taken. Every result in this series — the share, the wedge, the meter, the back-debt, the stack, and now the macro evidence that paying is not only survivable but growth-positive — points at the same closing condition:

The data wall guarantees the question will be forced; the only choice is whether it is settled by design or by rupture. The arithmetic is done. The evidence is in. The institutions exist. What the fourth factor is owed is no longer a mystery — it is a number, a mechanism, and a date.

The ledger is open. The first economy to close it wins.

— End of the Returns to Data series.

References and Sources

1. Jones, D., & Marinescu, I. (2022). “The Labor Market Impacts of Universal and Permanent Cash Transfers: Evidence from the Alaska Permanent Fund.” AEJ: Economic Policy. No employment effect; +1.8pp (+17%) part-time; general-equilibrium demand channel.
2. West, S., & Castro, A. (2023). Stockton SEED RCT: no employment effects of unconditional transfers.
3. OpenResearch Unconditional Income Study (2024). $1,000/month: ~2pp employment reduction, concentrated in caregiving/education time use.
4. Cesarini, D., Lindqvist, E., Notowidigdo, M., & Östling, R. (2017). Lottery-winner labor supply: small effects.
5. Robins, P. (1985). Negative-income-tax experiments: small labor-supply responses.
6. Watson, B., et al. (2020). Alaska PFD: −8% property crime post-transfer.
7. Princeton Legal Journal (2025). Alaska PFD legal history; 81% of Alaskans report the dividend improves their lives; poverty reductions in Alaska Native and elderly populations.
8. Norges Bank Investment Management / Government Pension Fund Global: the >$2T sovereign-windfall-as-assets precedent and fiscal-rule drawdown design.
9. Epoch AI (2024–2026). Public-text exhaustion estimates (the data wall) underlying §5’s supply feedback.
10. Vector1 Research (2026). Returns to Data I–IV: the fourth-factor framework; γ̂ ≈ 0.45; Φ ≈ 0.04; the $34B/yr flow gap; the $100–700B back-debt; the six-mechanism stack.

Methodological note: Figures 1 and 3 are scenarios built on the series’ parameters (γ̂, Φ paths, revenue paths), not forecasts; Figure 2 summarizes point estimates from heterogeneous study designs and should be read for sign and magnitude, not precision; Figure 4’s dates are scenario midpoints. The inflation bound in Equation 3 is an upper bound under standard MPC assumptions, not a model of monetary policy response. Code and derivations: github.com/vector1-ai (forthcoming).

Brian Curry is a Kansas City–based data scientist, AI researcher, and founder of Vector1 Research. The Returns to Data program — Articles I–V — develops the economic theory of data ownership in the age of artificial intelligence. The full series, with code and figures, is collected at vector1.ai.

Contact: brian at vector1.ai · License: CC BY 4.0

#ReturnsToData #DataDividend #UBI #AlaskaPermanentFund #AIEconomics #DataOwnership #FourthFactor #FutureOfWork #Inequality #Vector1Research