AI Rewards Sharing(ARS): Modeling an Economy Where AI Does the Work

Introduction

Advances in artificial intelligence (AI) are poised to transform the economy by taking over a large share of human labor. As AI systems drive productivity to new heights, the cost of producing goods and services could plummet. In such a scenario, traditional jobs may be scarce and human labor will be far less central to economic output. How will people earn a livelihood, and how will they afford living expenses if AI does most of the work? One proposal is to implement an “AI rewards sharing” system — a form of universal dividend that distributes the fruits of AI’s productivity gains to all people. This concept, akin to (but pointedly not called) universal basic income, suggests that since AI is built on collective human knowledge and data, its economic benefits should be shared with the humans who helped create it. In this article, we explore the historical context for such an idea, examine how an AI-dominated economy might function, and use a simple model to project the outcomes of an AI rewards sharing scenario for the United States.

Historical Lessons: Technology, Productivity, and Prices

Historically, technological progress has boosted productivity and often led to lower costs and prices for consumer goods. For example, the automation of agriculture and manufacturing in the 20th century dramatically reduced the cost of essentials. In 1947, American households spent about 23% of their income on food; by 2023, this had fallen to only 7% as food became far cheaper relative to incomes[¹^][³^]. Similarly, the share of household budgets devoted to clothing and home furnishings fell by multiple-fold over the same period[³^]. These shifts were driven by efficiencies and innovation — higher crop yields, factory automation, global supply chains — which made goods more abundant and affordable. Rising productivity meant consumers could buy more with each dollar, effectively raising living standards even without equivalent wage increases.

While consumers benefited from lower prices, productivity gains did not always translate into broadly shared wage growth for workers. Since the 1980s, a gap has opened between productivity and typical pay. In the U.S., net productivity rose about 60% from 1979 to 2019, but median worker compensation rose only ~16% in real terms[²^]. In large part, this reflects a declining labor share of national income and increasing inequality — a growing portion of economic output has been accruing to business owners and capital rather than workers[²^]. Automation has enabled companies to produce more with fewer employees, shifting income from workers to owners of capital[¹^]. As a Brookings analysis notes, workers who can complement new machines often see gains, but those who are substituted by automation lose out; overall, automation tends to raise profits while reducing the need for labor, unless countervailing policies exist[¹^]. In sum, technology has made goods cheaper and societies richer on average, but without deliberate intervention, the distribution of those gains can become very skewed.

The Coming AI Disruption to Labor

The next few decades could bring an automation wave unlike any before, as advanced AI and robotics enable the automation of not just routine factory tasks, but also service and cognitive jobs. AI systems are already writing code, drafting documents, driving vehicles, and performing customer service. Unlike past technological shifts that mostly affected manual and repetitive jobs, modern AI can potentially perform a wide array of tasks, including complex decision-making and creative work. Economists and technologists are debating: Will this time be different? There is a real possibility that AI could displace workers across many industries faster than new job roles emerge, leading to net job losses on an unprecedented scale.

If AI takes over the bulk of productive work, human labor may become a much smaller factor in production. In economic terms, the labor share of output could shrink dramatically from its current ~60% of GDP to much lower levels. In a thought experiment “worst-case” scenario (purely for modeling purposes), one could imagine an economy where almost all tasks are automated and human labor contributes only, say, 10–20% of output. In such a scenario, the traditional link between employment and income breaks down for most people. We would have an economy of immense output — since AI-driven productivity could increase GDP many times over — but most individuals would have little or no labor income to claim a share of that output.

At the same time, with AI and robots performing work more efficiently than humans, we’d expect the cost of goods and services to fall significantly. Just as earlier automation made food, clothing, and appliances cheap, AI could do the same for a wider range of products — possibly even services like healthcare and education, areas that until now suffered from “Baumol’s cost disease” (low productivity and rising costs). If AI can handle tasks in medical diagnosis, tutoring, construction, etc., it might reverse the relentless price inflation in those sectors, making even today’s expensive services far more affordable. In an AI-dominated economy, many goods could become near-zero marginal cost: abundant energy (via AI-optimized systems), highly efficient manufacturing, and automated logistics might yield extremely low prices for consumers. Essentially, society could approach a post-scarcity condition for many basic needs; the challenge would be less about producing enough, and more about distributing the wealth and output that AI produces.

However, even if everything is cheap, people still need an income or credit to obtain goods and services. If few have traditional jobs, we must rethink the income distribution mechanism. This is where AI rewards sharing enters the picture: the idea that every person should receive a regular stipend or dividend funded by the enormous wealth generated by AI-driven productivity. This concept goes by many names — data dividends, AI dividends, universal basic income, social wealth funds — but the core principle is the same: ensure that everyone has purchasing power and benefits from the AI revolution, not just the owners of AI systems.

Sharing the Fruits of AI: From Robot Taxes to Data Dividends

How might an AI-driven dividend be funded and distributed? Several mechanisms have been proposed by economists, technologists, and policymakers:

• Taxing AI and Automation: One idea is to levy taxes on companies that replace human workers with AI or robots, and use the revenue to support those displaced. Even Bill Gates famously suggested a “robot tax” — essentially, if a robot takes a human’s job, it should pay taxes like the human worker would have, to fund social benefits. This could be implemented as a tax on automation equipment or on the extra profits firms earn from AI-driven efficiency.
• Data/AI Dividend: AI systems derive their intelligence from vast datasets of human-produced content (text, images, code, etc.). In that sense, AI is built on a collective resource: human knowledge. There’s a growing argument that companies profiting from AI should pay a “data dividend” to the public. Former presidential candidate Andrew Yang, for example, proposed that internet users should be paid for the data that tech companies harvest. Expanding this concept, some suggest an AI dividend for all citizens, reflecting society’s stake in the knowledge fuel that powers AI. As one commentator put it, “AI models get their value by ingesting articles, art, books, and other human-created works… It’s only fair that these companies share the benefits more widely.”[⁶^]
• Sovereign Wealth Fund for AI: Drawing inspiration from resource-rich economies, another approach is to treat AI’s profits like oil revenue — something to be partially socialized for the public good. For instance, the U.S. could impose an AI revenue tax and funnel it into a national wealth fund. This fund would invest in the economy and pay out dividends to citizens. The analogy here is the Alaska Permanent Fund: since 1982, Alaska has taxed oil companies and invested the proceeds, paying every Alaskan an annual dividend (approximately $1,700 per resident in 2024)[⁶^]. The idea is that AI, like oil, is a resource (in fact, AI’s resource is human-generated data, which we collectively “own”) and thus its gains should be shared. A recent proposal suggests implementing a “severance tax” on AI based on the size or capability of AI models (for example, a tiny fraction of a cent per AI model parameter), with the tax revenue going into a public investment fund[⁶^]. Over time, this could yield a sizable dividend for all citizens, just as Norway’s oil fund or Alaska’s fund have done for their populations.
• Public Stakeholding in AI Firms: Yet another route is for the government (on behalf of citizens) to hold equity or ownership stakes in the most profitable AI enterprises. If AI-driven companies become the new engines of the economy, paying large dividends and commanding high valuations, a public ownership share would ensure a portion of those profits flow to the people. This could be done via a sovereign wealth fund that buys shares, or through requiring companies to issue some equity to a public trust.

Each of these mechanisms faces practical questions — how to implement the taxes without stifling innovation, how to identify which companies or AI activities to target, how large the dividend should be, etc. But they all align with a common vision: the benefits of AI automation must be broadly shared to maintain economic balance. Importantly, capturing even a modest fraction of AI’s value for public benefit could fund a substantial basic income. One recent economic modeling study found that if AI-driven automation became roughly 5–6 times more productive than current levels, it could finance a UBI of around 11% of GDP (roughly equivalent to, say, ~$12,000 per person annually in the U.S.), without requiring new taxes beyond capturing the AI-generated rents[⁵^]. That analysis assumed a “worst-case” scenario of mass job loss (no new jobs created), and showed that with sufficient AI productivity and some mechanism to socialize its profits, a sizeable universal dividend is economically feasible. Moreover, if the government increases its public revenue share of AI profits (for example, via higher corporate taxes on AI earnings or equity stakes), the required AI capability threshold for funding a given basic income would be even lower[⁵^]. In plain terms: AI doesn’t have to achieve sci-fi levels of superintelligence to fund a generous social dividend — provided we implement policies to capture part of AI’s windfall for society.

Modeling an AI-Dominated Economy with Rewards Sharing

To better understand these ideas, let’s walk through a simplified simulation of a future U.S. economy where AI plays a dominant role. This is a thought experiment — not a precise prediction — but it will illustrate the quantitative impact of AI-driven productivity and the introduction of an AI dividend. We make the following high-level assumptions for our model:

• Baseline (no AI revolution) Scenario: GDP grows at a modest 2% per year (roughly in line with historical averages). Labor’s share of income remains around its current level (~60% of GDP in the U.S.), meaning wages and salaries grow in tandem with productivity. In this scenario, most people continue to earn income through jobs as usual.
• AI Automation Scenario: AI causes a productivity boom starting in the 2020s. By 2030, annual GDP growth accelerates to 5% and stays high thereafter (reflecting AI’s contribution to output). Meanwhile, as automation spreads, labor’s share of GDP steadily falls — we assume it declines from 60% in 2025 to only about 10% by 2060, indicating that the vast majority of output by then comes from AI capital rather than human labor. Importantly, we introduce an AI rewards sharing policy: the government taxes 30% of the AI-driven capital income (the profits/rents accruing to AI owners) and redistributes it equally to all citizens as an “AI dividend.” This is analogous to a robust data dividend or sovereign fund payout.
• Population: The U.S. population is assumed to grow slightly (about 0.3% per year), from 330 million in 2025 to around 366 million in 2060. The dividend is shared among all residents.

Below, we include a Python code snippet that implements this scenario. This code simulates GDP growth, tracks the division of income between labor and capital, and calculates the annual AI dividend per person over time. (For simplicity, all figures are in inflation-adjusted 2025 dollars.)

import numpy as np

# Simulation years
years = np.arange(2025, 2061)

# Initial values
gdp = 25.0  # $25 trillion GDP in 2025 (approximate)
labor_share = 0.60  # 60% of GDP to labor in 2025
population = 330.0  # 330 million people in 2025

# Arrays to store results
labor_income_pc = []   # labor income per capita
ai_dividend_pc = []    # AI dividend per capita

for year in years:
    # Record current labor income per person
    labor_income_pc.append((labor_share * gdp * 1e12) / (population * 1e6))
    # Calculate AI dividend for this year (30% tax on capital income, distributed per person)
    capital_share = 1 - labor_share
    dividend_total = 0.30 * capital_share * gdp * 1e12  # total dividend $ 
    ai_dividend_pc.append(dividend_total / (population * 1e6))
    # Project next year
    if year < 2030:
        growth_rate = 0.02 + (0.05 - 0.02) * ((year - 2025) / (2030 - 2025))
    else:
        growth_rate = 0.05  # 5% annual GDP growth after 2030
    gdp *= (1 + growth_rate)
    # Labor share declines linearly from 60% in 2025 to 10% by 2060
    labor_share -= (0.60 - 0.10) / (2060 - 2025)
    # Modest population growth
    population *= 1.003

In this code, we start with 2025 values and iterate year by year to 2060. The labor share is reduced a bit each year (eventually reaching 0.10 = 10%), representing AI gradually taking over tasks. The AI dividend per capita is calculated as 30% of the capital income, divided by the population. Now let’s discuss the outcomes of this simulation, and then present a chart of the results.

By construction, the baseline scenario (no AI revolution) would simply show GDP per person and labor income per person growing gradually ~2% annually, with no special dividend — essentially a continuation of current trends. The interesting results come from the AI scenario with rewards sharing. In that scenario, total GDP grows explosively due to AI: by 2060, GDP in our simulation reaches about $130 trillion (in 2025 dollars), roughly 5 times the 2025 level. However, because human labor now contributes only 10% of output, the labor income going to workers in 2060 is around $13 trillion (10% of 130) — which, spread among a larger population, actually reduces the average labor earnings per person. In fact, in this scenario the average labor-based income per person peaks around the 2040s and then declines. By the 2050s, the typical person would earn less from labor than they would have in a world without AI (since relatively few people are needed or employed in the workforce).

This is where the AI dividend makes a dramatic difference. The AI dividend per person grows exponentially as AI-driven output expands. Initially, in the 2020s when automation is only moderate, the dividend is small — a few thousand dollars per year. But as we approach 2050 and beyond, the dividend eclipses labor income. By our 2060 endpoint, the annual AI dividend per American reaches nearly $100,000 (2025 dollars) per person under these assumptions, far exceeding what the average labor earnings would be. In other words, in a mature AI economy, people could receive more income from the AI dividend than from wages. This makes intuitive sense: if AI and machines are doing, say, 90% of valuable work, then humans must draw their livelihood from that automated output rather than from their own labor.

To visualize these dynamics, the following chart plots the simulated average income per person under different sources over time:

Simulation of an AI-dominated economy (2025–2060). Blue line: average labor income per person in a baseline scenario with modest growth. Orange line: average labor income per person in the AI automation scenario (labor share declining). Red line: AI dividend per person in the AI scenario (30% of AI profits redistributed). All values are in inflation-adjusted 2025 dollars.

In the baseline (blue line), average labor income per person grows from about $45k in 2025 to around $80k by 2060, reflecting slow and steady economic growth with stable labor share. In the AI scenario without any redistribution (orange line), average labor income per person initially follows a similar path, but then tops out around the 2040s at roughly $55–60k and declines to about $40k by 2060. This decline happens because even though total output is much higher, workers are getting a shrinking slice of that pie (only 10% by 2060). Finally, the AI dividend (red line) starts near zero and skyrockets in the latter decades. By the mid-2050s, the red line (dividend) overtakes the orange line (labor income in AI world), indicating the dividend has become the primary source of income for the average person. By 2060, the dividend reaches roughly $95k per person per year in this scenario, more than double the labor earnings in the AI scenario, and even higher than the baseline-world labor income.

The takeaway is striking: with a robust AI reward sharing system, individuals could be substantially better off in terms of income, despite hardly working at all — provided that the dividend captures a healthy share of AI’s output. This would enable people to consume the abundant products and services produced by AI. In essence, AI replaces the paycheck. Instead of wages for work, people receive “AI income” funded by the productivity of machines. And because AI makes goods cheap, that income goes very far in terms of purchasing power, enabling a high standard of living.

It’s important to stress that this scenario is optimistic and stylized. The real world may impose constraints that slow down such progress or limit the dividend. For instance, political will might cap tax rates or dividends at lower levels; AI’s productivity gains might not be as rapid or might stall due to regulatory/societal limits; concentrated corporate power could resist wealth redistribution, etc. Nonetheless, the exercise shows that from a purely economic output perspective, an AI-dominated economy can generate more than enough wealth to provide everyone a comfortable income — if we choose to structure it that way.

Challenges and Considerations

Designing and implementing an AI rewards sharing system will come with many challenges:

• Measurement and Taxation: Identifying what and how to tax in an AI-driven economy is non-trivial. Do we tax companies based on the number of human jobs they automate away? Or on their AI-generated profits? How do we account for AI systems that are diffused across many industries? Some proposals, like the parameter-based tax mentioned earlier, are novel and untested. Governments will need new tools and legal definitions to capture AI’s value without impeding innovation.
• Global Competition: If one country heavily taxes AI to fund a basic income but others do not, there could be competitive disadvantages or incentives for companies to relocate. This suggests that an international approach or at least coordination may be needed for taxing AI corporately, similar to global agreements on corporate taxation. On the flip side, if AI-driven unemployment becomes a worldwide issue, we might see multiple countries converging on some form of universal dividend or basic income as a solution.
• Amount and Timing: The level of dividend (how much per person) and when to start it will be key policy questions. Start too low, and it may not fully offset job losses; start too late, and you risk social upheaval from masses of unemployed workers with no support. If started too early or too generous without adequate AI productivity, it could strain public finances. Policymakers might phase in an AI dividend gradually — for example, beginning with a small “data dividend” today (some states have even considered requiring tech companies to pay users for data) and scaling it up as AI becomes more ubiquitous.
• Inequality and Ownership: Without interventions, AI could massively enrich a small group of tech firms and their owners, leading to extreme inequality. An AI dividend directly addresses income inequality, but not wealth inequality. There is a deeper question: should the AI platforms and algorithms themselves be publicly owned or open-sourced to some degree? Public ownership of AI infrastructure (like data commons or government-developed AI) could complement taxation as a way to ensure the technology benefits everyone. Also, care must be taken that a universal dividend truly reaches everyone and doesn’t get captured or eroded by other rising costs.
• Social and Psychological Factors: A transition to a post-work economy raises societal questions. Work provides meaning, purpose, and structure for many people. If AI frees humans from the necessity of work, how will individuals spend their time, and will they find fulfillment? Investing in education, arts, community, and volunteer opportunities could become important so that people can pursue passions instead of jobs. Additionally, there may be political resistance to giving “free money” to all — UBI pilots have faced skepticism — so framing the dividend as “everyone’s rightful share of AI prosperity” will be crucial for public acceptance. It emphasizes reciprocity: we all contributed (directly or indirectly) to AI’s training data and to the society that enabled these innovations, so we all deserve to benefit.
• Inflation and Macroeconomics: Interestingly, an AI-dominated economy could be deflationary (falling prices) because of tech efficiency. This is unlike typical scenarios where giving out lots of money (UBI) could cause inflation by boosting demand. If supply (production) is super-abundant thanks to AI, increased consumer spending won’t necessarily drive prices up — producers can just ramp up AI-driven output to meet demand at stable low prices. The bigger concern might be ensuring that the AI dividend is sufficient to stimulate demand. If goods are extremely cheap, a moderate dividend might be enough; but if certain goods like housing or land remain scarce and expensive, policy must also address those (e.g., building more housing, or providing housing vouchers in addition to cash). We might also see a shift in what constitutes a “good life” in material terms — perhaps basic goods are cheap but people spend dividend money on experiences, travel, or artisan products that AI can’t easily supply.

Conclusion

We stand on the brink of a new economic era. If AI indeed takes over the lion’s share of work, humanity could face a future reminiscent of science fiction: plenty of output with little labor. This could be a utopia of leisure and creativity — if we manage the transition right. The key will be decoupling livelihood from traditional jobs by ensuring everyone has access to “AI-earned” income. An AI rewards sharing program (whether one calls it AI dividends, universal basic income, or something else) may become not just a compassionate policy but an economic necessity, to maintain consumption and social stability. History has shown that productivity gains can enrich society, but without deliberate policy they can also concentrate wealth and leave many behind. With AI, that lesson looms larger than ever.

The encouraging news is that the numbers do add up: even a partial capture of AI’s economic gains can fund a substantial income for all, given how powerful AI productivity could be. The real challenges are societal and political. We must recognize that AI’s knowledge base was built by us — by countless people generating data and content — and thus its fruits rightfully belong to everyone. By setting up mechanisms to share AI’s bounty, we can reduce fear of automation and welcome AI as a tool that liberates humanity from drudgery. In such a world, the role of human beings can evolve beyond labor for survival. Our economic mantra could shift from “you must work to eat” to “AI works so that everyone can eat (and live well).” Achieving this will require ingenuity in policy equal to our ingenuity in technology, but it is a future worth striving for.

About the Author:

Brian Curry is a Kansas City–based Data Scientist, Artificial Intelligence expert and founder of Vector1 Research, a part time passion project, where he explores the frontier of AI, SEO, Machine Learning, Economic Systems, Marketing Modeling, and Econometrics. His work focuses on building intelligent systems that reveal the causal structures driving performance across marketing, content, and economics.

As the creator of Papilon, an open-source framework for complex system optimization, and Daedalus, a platform for causal and economic analysis.

References

1. Harry J. Holzer (2022). “Understanding the impact of automation on workers, jobs, and wages.” Brookings Institution. Online Article (Discusses how automation reduces costs/prices and shifts income from workers to owners.)
2. Lawrence Mishel (2021). “Productivity has grown 3.5 times as much as pay for the typical worker (1979–2019).” Economic Policy Institute, Working Economics Blog. Online (Documents the divergence between productivity growth and median compensation, and the erosion of labor’s share of income.)
3. Eliana Zeballos and Wilson Sinclair (2020). “Average share of income spent on food in the U.S. remained relatively steady from 2000 to 2019.” Amber Waves Magazine, USDA Economic Research Service. Online (Notes that from 1960 to 2000, the share of income spent on food fell from 17.0% to 9.9%, due to rising incomes and efficiencies that kept food prices low.)
4. Dean Baker (2024). “In the Good Old Days, One Fourth of Income Went to Food.” CEPR Blog (Apr 17, 2024). Online (Shows that in 1947 Americans spent 23% of income on food at home, versus just 7.1% in 2023, illustrating long-term cost reductions in necessities.)
5. Aran Nayebi (2025). “An AI Capability Threshold for Rent-Funded Universal Basic Income in an AI-Automated Economy.” arXiv preprint arXiv:2505.18687. Online (Finds that AI systems ~5–6× more productive than current automation could fund a UBI ~11% of GDP if AI rents are captured, and explores how higher public revenue share lowers the required AI capability level.)
6. Andrew Leahey (2024). “Alaska Is a Model for Taxing AI as an Extracted Natural Resource.” Bloomberg Tax (Oct 1, 2024). Online (Argues that AI’s value comes from human knowledge, analogous to a natural resource, and proposes taxing AI to fund a sovereign wealth fund that pays out dividends, similar to the Alaska Permanent Fund’s oil-based dividends.)