Agricultural machinery allowed fewer workers to produce more food. Industrial machinery multiplied the output of factories. Computers reduced the cost of information processing. The internet reduced the cost of communication and distribution.

Artificial intelligence takes this principle into a much larger domain.

It can potentially reduce the amount of human labor required for:

information processing,
software development,
documentation,
design,
customer service,
analysis,
translation,
research,
administration,
forecasting,
scheduling,
quality control,
and increasingly complex knowledge work.

That creates a straightforward engineering relationship:

more useful output per unit of input → higher productivity → lower unit cost → potentially higher economic output.

But there is a missing variable.

Who receives the economic value created by the productivity improvement?

That question cannot be answered by productivity statistics alone.

Consider a simplified production system:

Y=F(K,L,A)

where:

(Y) = economic output,
(K) = capital,
(L) = labor,
(A) = technology/productivity.

Suppose AI increases (A). Output can rise substantially without requiring a proportional increase in (L). That is economically beneficial. But the resulting income can flow through several channels:

ΔY → { higher wages; higher profits; lower prices; higher investment; higher asset values; higher tax revenue; new employment }

There is no economic law requiring these channels to receive equal shares.

This is the central productivity paradox:

An economy can become substantially more productive without becoming proportionally more prosperous for every person participating in it.

The distinction matters because AI is simultaneously a productivity technology, automation technology, capital technology and knowledge technology.

The distributional consequences therefore depend not simply on how capable AI becomes, but on how AI interacts with labor, capital, markets and ownership.

Recent evidence makes the issue less hypothetical. The International Labour Organization’s 2026 review finds that productivity gains from generative AI are real in several settings but uneven, while large-scale displacement remains limited so far. It also identifies inequality, reduced employment opportunities for younger workers and changes in work organization as significant concerns.

The IMF’s 2026 analysis of Asia reaches an equally important conclusion: countries with greater structural preparedness may obtain earlier productivity gains, while weaker adoption capacity can produce divergence; within countries, AI can increase inequality when gains are concentrated among high-skilled workers and capital owners.

The problem, therefore, is not whether AI can increase productivity. It can. The engineering problem is determining how the productivity gain propagates through the economic system.

Productivity Does Not Determine Who Gets Paid

Productivity and income distribution are related, but they are not the same variable. A worker producing twice as much output per hour does not automatically receive twice the wage. Why?

Because the market value of the additional output depends on:

demand,
competition,
bargaining power,
ownership,
substitutability,
scarcity,
institutional rules,
and the structure of the production process.

A useful simplified relationship is:

Productivity Gain ≠ Wage Gain

More specifically:

ΔW + ΔΠ + ΔC + ΔI + ΔT + ΔR

where the additional economic value can ultimately appear as:

- ΔW = higher labor compensation
- ΔΠ = higher profits
- ΔC = consumer surplus/lower prices
- ΔI = additional investment
- ΔT = government revenue
- ΔR = returns to asset ownership

The distribution depends on the production system.

A simple example

Imagine a company employing 1,000 workers.

Before AI:

Variable	Before AI
Workers	1,000
Output	100,000 units
Output/worker	100
Wage/worker	$50,000
Total wages	$50 million

Now assume AI increases output per worker by 40%.

If employment remains constant: 100 → 140

Output rises to 140,000 units. That is an obvious productivity success. But several different outcomes are possible.

Scenario	Employment	Wage effect	Profit effect	Productivity
Augmentation	1,000	Higher	Moderate	High
Cost reduction	800	Mixed	High	High
Strong demand expansion	1,200	Higher	High	High
Automation without expansion	600	Lower aggregate wages	Very high	High
Competitive price reduction	1,000	Moderate	Moderate	High

The same technological improvement can therefore produce radically different social outcomes. This is why saying “AI will increase productivity” is economically incomplete.

The more important questions are:

How much?

In which tasks?

For whom?

Who owns the technology?

What happens to the displaced labor?

Does demand expand sufficiently to absorb released labor?

Does competition force productivity gains into lower prices?

Or does market concentration allow firms to retain the gains?

The answer determines who gets richer.

Automation, Augmentation and the Engineering of Labor

The most important distinction in analyzing AI is not simply AI versus humans.

It is:

AI substituting for a task versus AI complementing a worker performing a task.

An occupation is not a single activity. It is a collection of tasks.

A civil engineer, for example, may perform:

calculations,
field inspections,
technical writing,
design review,
coordination,
judgment,
contract interpretation,
stakeholder communication,
risk assessment,
site supervision.

AI may automate some of these tasks while improving the productivity of others. The occupation therefore does not necessarily disappear. Its task composition changes.

The ILO’s refined 2025 global analysis estimates that roughly one in four workers globally are in occupations with some degree of GenAI exposure. However, it emphasizes that transformation is more likely than complete replacement because most occupations contain tasks requiring continuing human involvement.

The distribution is highly uneven. The ILO estimates exposure at approximately:

Country-income group	Employment with some GenAI exposure
High-income countries	~34%
Low-income countries	~11%
Global	~25%

These are exposure measures, not predicted job losses. That distinction is critical.

The task-engineering model

Consider a worker performing 100 tasks.

Before AI:

100 tasks = 100 units of labor effort

After AI:

40 tasks automated

30 tasks augmented

30 tasks largely unchanged

The worker’s job has not necessarily disappeared. But the economic value of different skills has changed. The worker who knows how to supervise AI, verify outputs, integrate information and make decisions may become more productive. The worker whose principal value was executing routine information-processing tasks may become less valuable. This produces two opposing forces.

Substitution effect

AI performs tasks previously performed by humans:

L_d ↓

where (L_d) is labor demand.

Complementarity effect

AI increases the productivity of human workers:

MP_L ↑

where (MP_L) is the marginal product of labor.

Demand/productivity effect

Lower production costs can increase output and demand:

C_u ↓ → P ↓ → Q_d ↑ → Y ↑

where:

(C_u) = unit cost,
(P) = price,
(Q_d) = quantity demanded.

These mechanisms operate simultaneously.

Research by Acemoglu and Restrepo formalizes this task-based distinction: automation creates a displacement effect by moving tasks from labor to capital, while new tasks can create a reinstatement effect that generates new demand for labor. Their framework also shows why automation can increase productivity while reducing labor’s share of value added.

That is the fundamental reason technological unemployment is neither guaranteed nor impossible. The outcome depends on which effect dominates.

Why Higher Output Can Produce Lower Labor Demand

The popular argument is simple:

If AI makes workers more productive, firms will need fewer workers.

Sometimes that is correct. But it is not the whole system. Suppose a firm needs 1,000 labor-hours to produce 1,000 units. AI reduces the requirement to: 0.6 labor-hours/unit. The firm can respond in at least four ways.

Case A — Produce the same quantity

Q=1,000

Labor requirement: 600 hours

Employment falls.

Case B — Reduce price and expand output

Suppose lower unit cost increases demand sufficiently that:

Q=1,667

Then:

1,667(0.6) = 1,000 hours

Employment returns to approximately its original level.

Case C — Expand into new markets

The firm uses the cost savings to create new products. Labor is redeployed rather than simply eliminated.

Case D — Retain the savings as profit

Output rises, employment falls, and the additional value primarily accrues to owners.

All four are economically possible. This is why technological progress historically produces complicated employment effects rather than a single predictable result.

Acemoglu’s macroeconomic analysis of AI makes another important point: if AI’s effects primarily arise through task-level cost savings and productivity improvements, aggregate gains depend on both the fraction of tasks affected and the size of the productivity improvement. His 2024 analysis found that plausible near-term aggregate productivity effects could be meaningful but substantially smaller than some extreme AI forecasts.

This matters because micro-productivity is not automatically macro-productivity. A worker may save two hours using AI. That does not mean the economy automatically produces two hours’ worth of additional GDP.

The saved time must be converted into:

additional output,
higher quality,
new products,
lower prices,
additional demand,
innovation,
or some other economically measured benefit.

The ILO’s 2026 review finds precisely this gap in emerging evidence: workers report time savings, but those savings have not yet consistently translated into higher measured output, earnings or employment.

This is one of the most important engineering lessons of AI economics:

Efficiency at one component of a system does not guarantee proportional efficiency at system level.

A faster pump does not necessarily increase the flow rate of an entire pipeline if another section is the bottleneck. AI can be the same.

The New Distribution of Income: Skills, Capital and Ownership

The most consequential distributional question may not be simply:

“Will AI replace workers?”

It may instead be:

“Which workers become more valuable, which become less valuable, and who owns the capital complementary to AI?”

AI can create strong complementarities with workers who possess:

technical expertise,
domain knowledge,
decision-making authority,
management capability,
communication skills,
AI literacy,
data skills,
and the ability to verify machine outputs.

This can increase the skill premium.

But AI can also compress some skill differences. The 2023 NBER study Generative AI at Work provides an instructive example. Using data from 5,179 customer-support agents, researchers found that access to an AI assistant increased productivity by about 14% on average, with approximately a 34% improvement for novice and lower-skilled workers and little measurable effect for experienced, highly skilled workers.

That is a powerful result. AI did not simply reward the most skilled worker. In that particular setting, it helped less-experienced workers catch up. This suggests that AI can sometimes function as a skill equalizer.

But the opposite mechanism is also possible. If highly skilled workers are especially complementary with AI, their output may rise dramatically.

Then:

MP_high-skill ↑↑

while:

MP_low-skill ↑

The wage distribution can widen.

The IMF’s 2025 analysis explicitly identifies this ambiguity. AI could reduce some wage inequality by affecting high-income occupations, but high-skilled workers may simultaneously benefit from strong complementarity with AI and from higher returns to capital, creating greater wealth inequality.

Labor income versus capital income

This leads to the deeper issue.

If AI increases the importance of:

GPUs,
data centers,
software,
intellectual property,
cloud infrastructure,
proprietary datasets,
algorithms,
energy infrastructure,
and financial capital,

then the income generated by AI can increasingly flow toward owners of those assets.

A simplified national-income relationship is:

Y=WL+rK
where:

(WL) = labor income,
(rK) = capital income.

If AI causes the effective capital contribution to rise faster than labor compensation:

(rK)/Y ↑

then national output can rise while labor’s share falls.

That creates the uncomfortable possibility that:

The economy becomes richer because machines are doing more, while workers become relatively less important to the production process that creates the wealth.

This is not an argument against AI. It is an ownership problem. If ownership of productive AI capital is widely distributed, productivity gains can become broadly distributed. If ownership is concentrated, productivity gains can become concentrated. The technology does not determine this distribution by itself. The economic architecture does.

AI Infrastructure Creates a New Concentration Problem

AI is frequently described as “software.” That is misleading. Frontier AI is a physical industrial system.

It requires:

semiconductor manufacturing,
advanced accelerators,
memory,
data centers,
electricity,
cooling,
networking,
cloud infrastructure,
data,
engineering talent,
and enormous financial investment.

The concentration of these inputs matters because ownership of bottleneck infrastructure creates economic power.

The 2026 Stanford AI Index reports that global AI compute capacity has grown extremely rapidly since 2022 and that Nvidia accounts for more than 60% of total compute capacity measured in H100-equivalents. It also reports that the United States hosts thousands of data centers and that the leading AI hardware supply chain remains heavily dependent on a small number of firms and locations.

This creates an economic architecture resembling a bottlenecked engineering system.

Consider:

f(chips, compute, energy, data, talent, capital)

If several inputs are highly concentrated, entry becomes difficult.

That can create:

High fixed costs → few competitors → market power → higher returns to capital

This is fundamentally different from a technology that can be copied at almost zero cost using ordinary infrastructure.

The AI concentration chain

Layer	Scarce resource	Potential economic power
Semiconductor	Advanced chips	Hardware bottleneck
Compute	GPU/accelerator clusters	Access bottleneck
Data centers	Physical infrastructure	Capacity bottleneck
Energy	Reliable electricity	Operating bottleneck
Models	Frontier intellectual property	Technology bottleneck
Data	High-quality proprietary data	Information bottleneck
Talent	Specialized researchers	Human-capital bottleneck
Capital	Massive investment	Financing bottleneck
Distribution	Cloud/platform ecosystems	Customer-access bottleneck

This is why AI could generate a new form of capital concentration even if the underlying software becomes increasingly accessible.

The World Bank’s 2025 assessment makes the global dimension explicit: high-income countries dominate AI innovation, compute infrastructure and startup funding, while developing countries face gaps in connectivity, compute, locally relevant data and skills. It identifies four foundations—connectivity, compute, context and competency—as essential to inclusive AI adoption.

The productivity machine therefore has a physical geography. And geography affects who gets rich.

Why Countries May Diverge Instead of Converge

Technology is often expected to reduce international inequality. Digital technology should, in theory, allow a worker anywhere to access the same information. AI strengthens that possibility. A small company in a developing country can potentially use a world-class language model without building the model itself. That is a genuine opportunity.

But the opposite force is equally important. AI adoption requires infrastructure.

A country without:

reliable electricity,
broadband connectivity,
cloud access,
data infrastructure,
technical education,
digital institutions,
investment capital,
and capable firms

cannot capture the same productivity gains as a country possessing them.

The IMF’s 2026 analysis of Asia finds that structurally prepared advanced economies tend to adopt AI earlier and experience earlier growth gains, while emerging and developing economies can face delayed adoption and higher costs of capital. It also concludes that reforms improving productivity and human capital can accelerate AI adoption and amplify growth gains.

The IMF’s AI Preparedness Index illustrates the structural gap. Its published index values include approximately 0.68 for advanced economies, 0.46 for emerging-market economies and 0.32 for low-income countries.

Simplified Divergence Mechanism:

Higher preparedness
        ↓
Earlier AI adoption
        ↓
Higher productivity
        ↓
Higher investment
        ↓
More AI infrastructure
        ↓
Further productivity gains
        ↺

The reverse can also occur:

Low preparedness
        ↓
Delayed adoption
        ↓
Lower productivity growth
        ↓
Lower investment capacity
        ↓
Weak infrastructure
        ↓
Further adoption delay
        ↺

This is a positive-feedback problem.

It means AI could either: reduce development gaps,

or: reinforce development gaps

depending on whether access to the productivity technology becomes sufficiently broad.

The ILO has similarly warned that high-income countries have substantially greater occupational exposure to GenAI, while low-income economies have less exposure but also face infrastructure constraints that can limit their ability to capture augmentation benefits.

This produces a paradox. Countries with the greatest immediate exposure to AI may also have the greatest capacity to benefit from it. Countries with lower exposure may appear protected from automation while simultaneously being excluded from the productivity gains. That is not technological convergence. It is potentially technological divergence.

Why GDP Can Misread the AI Economy

There is another problem. Even if AI makes people materially better off, GDP may not capture the entire improvement. Suppose an AI assistant provides:

free tutoring,
free translation,
free programming assistance,
free research support,
free writing assistance.

A user receives substantial economic value. But if the service is free, the measured market transaction may be small or zero. That creates a measurement problem. A simplified welfare relationship is:

Economic welfare ≠ GDP

GDP measures production within a defined national-accounting framework. It does not directly measure every dimension of consumer surplus.

The 2026 Stanford AI Index estimates substantial consumer value from generative AI services in the United States, including an estimated annual consumer surplus of about $172 billion by early 2026. The important implication is that users can receive considerable value even when many AI services are free or inexpensive.

This creates an unusual situation. AI can simultaneously produce: higher real welfare

while: modest measured GDP effects

at least during periods when services are heavily subsidized, priced near zero or difficult to measure.

Productivity measurement problem

Suppose an engineer previously spent 8 hours, producing a technical report. AI reduces this to 3 hours. The engineer now has five hours available. What happened to GDP? If the engineer produces another report:

GDP ↑

If the engineer uses the time for internal learning:

GDP ≈ unchanged

If the engineer improves project quality:

economic welfare ↑

but the GDP effect may be difficult to measure.

If the engineer works fewer hours:

GDP ↓

even though productivity per hour may have increased.

Thus:

GDP/hour ↑

does not necessarily imply:

GDP ↑

And:

GDP ↑

does not necessarily imply:

median household welfare ↑

This distinction becomes increasingly important as AI produces more intangible output.

What Governments Can Actually Do

If AI productivity gains are distributed unevenly, governments face a difficult design problem. They cannot simply prevent automation. That would sacrifice productivity gains and potentially leave domestic firms less competitive. Nor can they assume that market forces will automatically distribute the gains.

The policy objective should therefore be:

maximize productivity + maintain competitive markets + expand access to opportunity + redistribute where necessary

Several mechanisms matter.

1. Education and reskilling

Education should move away from merely teaching people to perform tasks that machines increasingly perform well.

The more durable objective is developing capabilities that complement automation:

problem formulation,
engineering judgment,
systems thinking,
verification,
interdisciplinary reasoning,
communication,
physical-world execution,
leadership,
creativity,
and domain expertise.

But “reskilling” should not become a slogan. Training is useful only when there is actual demand for the resulting skills. A worker cannot be economically protected simply by being given a certificate. The system must connect:

training → skills → jobs → income

2. Competition policy

If AI infrastructure becomes concentrated, competition policy becomes increasingly important. The objective is not to punish successful companies. It is to prevent control of critical bottlenecks from becoming an enduring barrier to entry.

Potential policy areas include:

interoperability,
cloud competition,
data portability,
anti-competitive conduct,
infrastructure access,
merger scrutiny,
open standards,
and support for open-source ecosystems.

3. Taxation

A government can capture part of the additional economic surplus through taxation. But taxation must be designed carefully. Taxing productive investment too heavily can discourage the very capital formation required for productivity. The better target may be excessive rents rather than productive investment itself.

The basic distinction is:

productive return ≠ economic rent

4. Social insurance

If automation produces temporary displacement, workers may need:

unemployment insurance,
wage insurance,
portable benefits,
retraining support,
relocation assistance,
and income support.

These mechanisms reduce the transition cost.

5. Universal basic income

Universal basic income is often presented as the obvious response to automation.

It is not.

UBI can redistribute purchasing power, but it does not solve:

housing shortages,
healthcare costs,
market concentration,
education quality,
regional inequality,
ownership concentration,
or the loss of meaningful employment.

It can also be fiscally expensive.

A better framing is: UBI is one possible distribution mechanism, not a technological necessity.

If AI productivity becomes extraordinarily high, however, some form of unconditional or near-unconditional income transfer could become more economically feasible because the taxable surplus would be larger.

6. Broadening capital ownership

This may be the most structurally important option. If AI increasingly behaves like productive capital, then broad ownership of productive assets becomes more important.

The economy could distribute AI gains through:

pension funds,
employee ownership,
profit-sharing,
sovereign wealth funds,
broad-based investment vehicles,
citizen dividends,
or other mechanisms.

The core principle is simple:

AI ownership → AI returns → household income

If households own a portion of the productivity-generating capital, they can receive income from productivity growth even if labor’s share declines.

This changes the problem from:

“How do we preserve every existing job?”

to:

“How do we ensure that people participate in the ownership of the productive system?”

That is a much more durable question.

The Real Test: Who Owns the Productivity Machine?

The deepest mistake in discussions about AI economics is treating technology as though it determines the distribution of its own benefits. It does not. Technology changes the production function. Institutions determine much of what happens afterward.

A useful systems representation is:

                    AI TECHNOLOGY
                         │
          ┌──────────────┼──────────────┐
          ↓              ↓              ↓
      Automation     Augmentation    Innovation
          │              │              │
          ↓              ↓              ↓
      Labor demand    Labor output    New markets
          │              │              │
          └──────────────┼──────────────┘
                         ↓
                  TOTAL PRODUCTIVITY
                         │
        ┌────────────────┼────────────────┐
        ↓                ↓                ↓
      Wages           Profits         Lower prices
        │                │                │
        ↓                ↓                ↓
    Households        Owners        Consumers
                         │
                         ↓
                  Wealth distribution

The final distribution therefore depends heavily on:

Technology + Ownership + Competition + Labor institutions + Human capital + Public policy

That is why the question “Will AI make us richer?” is too crude.

The technically meaningful questions are:

How much additional output can AI generate?
Which tasks are automated?
Which tasks become more productive?
How elastic is demand for the resulting products?
How quickly can workers move between tasks?
How concentrated is AI capital?
Who owns that capital?
How competitive are AI markets?
How quickly can countries adopt the technology?
How effectively can governments redistribute part of the resulting surplus?

The answers determine the distribution of the productivity dividend.

The 3 Possible AI Economies

Economic structure	Productivity	Ownership	Likely distribution
Broad-access AI	High	Broad	Broad prosperity
Competitive but capital-heavy AI	Very high	Concentrated	Mixed prosperity
Highly concentrated AI	Very high	Highly concentrated	High inequality

The most important variable may therefore not be the intelligence of the machine. It may be the ownership structure surrounding the machine.

Conclusion: A Richer Economy Is Not Automatically a Richer Society

The historical promise of technology has always been simple:

less effort → more output → higher living standards

AI could push that relationship further than previous digital technologies because it attacks a much broader category of production: cognitive and information-intensive work. The productivity opportunity is genuine.

AI can reduce production time, increase output, improve access to expertise, lower information costs and allow small organizations to perform functions that previously required large teams. Empirical research already demonstrates measurable productivity improvements in specific work environments.

But productivity is only the first stage. The second stage is distribution. And distribution is not determined by the production function alone.

If AI replaces labor-intensive tasks while ownership of the resulting productive capital remains concentrated, national output can rise while labor’s relative share declines.

If AI complements workers, productivity gains can raise wages.

If AI reduces costs and stimulates demand, new employment can emerge.

If new tasks are created faster than old tasks disappear, technological change can expand rather than contract labor demand.

If market power becomes concentrated around compute, data, infrastructure and proprietary models, however, a large portion of the productivity surplus can flow toward capital owners.

And if access to AI infrastructure differs dramatically between countries, the technology could increase rather than reduce international economic divergence.

The emerging evidence does not justify either extreme. There is no sound basis for claiming that AI will inevitably eliminate most employment. There is equally little basis for assuming that every productivity gain will automatically become higher wages. The evidence points toward something more complicated: task transformation, heterogeneous productivity gains, changing skill premiums, potential capital-income concentration and significant differences in national preparedness.

The central economic relationship is therefore not:

AI → productivity → prosperity

It is:

AI → productivity → surplus → distribution → prosperity

The last two stages determine the answer to the original question. If machines become more productive, who actually gets richer? The answer will not be determined by how intelligent the machines become.

It will be determined by who can use them, who works alongside them, who competes with them, who controls the infrastructure, who owns the capital, and how society distributes the enormous surplus that increasingly productive machines can create.

That is the real AI productivity paradox. A society can solve the engineering problem of producing more. The harder problem is deciding who gets to own, earn and benefit from what is produced.

Selected Research and Data Sources

IMF — AI and Economic Divergence in Asia, 2026 — Evidence on productivity gains, country divergence, capital income and within-country inequality.
IMF — AI Adoption and Inequality — Task-level analysis of AI adoption, wages, high-skilled complementarity and wealth inequality.
IMF — Gen-AI: Artificial Intelligence and the Future of Work — Global AI exposure and the IMF’s AI Preparedness Index.
ILO — The Impact of GenAI on Jobs, Productivity and Work Organization, 2026 — Review of emerging empirical evidence on productivity, employment and workplace effects.
ILO — Generative AI and Jobs: A Refined Global Index of Occupational Exposure — Global occupational exposure analysis covering nearly 30,000 tasks.
World Bank — Digital Progress and Trends Report 2025: AI Foundations — Evidence on the global AI divide and the importance of connectivity, compute, context and competency.
Stanford HAI — 2026 AI Index: Economy — AI investment, adoption, productivity, consumer value and labor-market indicators.
Stanford HAI — 2026 AI Index: Research and Development — Compute concentration, data-center infrastructure and AI hardware.
NBER — Generative AI at Work — Field evidence from 5,179 customer-support workers showing heterogeneous productivity gains.
NBER — Automation and New Tasks: How Technology Displaces and Reinstates Labor — Task-based framework explaining displacement and creation of new labor demand.
NBER — The Simple Macroeconomics of AI — Analysis connecting task-level AI productivity effects to aggregate productivity and GDP.

How AI Productivity Can Make the Economy Richer, But Without Making Everyone Richer

Table of Contents

Quick Links

The Productivity Paradox

Productivity Does Not Determine Who Gets Paid

A simple example

Automation, Augmentation and the Engineering of Labor

The task-engineering model

Substitution effect

Complementarity effect

Demand/productivity effect

Why Higher Output Can Produce Lower Labor Demand

Case A — Produce the same quantity

Case B — Reduce price and expand output

Case C — Expand into new markets

Case D — Retain the savings as profit

The New Distribution of Income: Skills, Capital and Ownership

Labor income versus capital income

AI Infrastructure Creates a New Concentration Problem

The AI concentration chain

Why Countries May Diverge Instead of Converge

Simplified Divergence Mechanism:

Why GDP Can Misread the AI Economy

Productivity measurement problem

What Governments Can Actually Do

1. Education and reskilling

2. Competition policy

3. Taxation

4. Social insurance

5. Universal basic income

6. Broadening capital ownership

The Real Test: Who Owns the Productivity Machine?

The 3 Possible AI Economies

Conclusion: A Richer Economy Is Not Automatically a Richer Society

Selected Research and Data Sources

Leave a Comment Cancel Reply