Saffron Huang is co-founder and advisor at the Collective Intelligence Project, working on enabling more democratic emerging technologies, and a researcher on the Societal Impacts team at Anthropic.
Sam Manning is a senior research fellow at the Centre for the Governance of AI.
As AI advances, many have become worried about the technology’s potential to concentrate unprecedented wealth among a few, while eroding the economic value of human work for everyone else. A variety of thought leaders seem to have converged on a favored solution to AI-driven job loss: universal basic income (UBI), which would entail giving every citizen unconditional cash payments.
It’s a simple idea: If AI generates unprecedented wealth while displacing workers, why not just redistribute some of that bounty? And if we don’t want extreme economic gains to become overly concentrated, we could have AI companies pledge to redirect a slice of profits toward the common good if they capture too large a share of the economy — as encapsulated in policy proposals like the “Windfall Clause.”
But proposed solutions like UBI or the Windfall Clause — which rely entirely on the assumption that redistribution can not only address AI-driven inequality but also be implemented effectively — misread the nature of political and economic power. By the time such reactive approaches are necessary, those controlling the AI economy may be powerful enough to evade meaningful taxation — and citizens too weak to demand their share. History tells this story repeatedly: Once power concentrates, the powerful reshape the rules, especially around taxation and wealth redistribution. The Medici did it. The robber barons did it. Big Pharma does it. But no Gilded Age titan ever wielded the kind of power that transformative AI may create.
In the most extreme economic disruption scenarios, income redistribution would only address symptoms. It would do little to prevent the underlying forces generating inequality in a world where few would own the AI capital powering the economy, and most people who previously relied on their labor to contribute would be increasingly sidelined from economic relevance. Such a scenario would likely demand the provision of perpetual income support. But as environmental scientist and complexity theorist Donella Meadows noted in 1999, tweaking system parameters — like redistributing income — is the least effective way to change a system. Some amount of income redistribution in the face of extreme wealth concentration may be necessary, but are there actions we can take sooner to avoid such inequalities unfolding in the first place?
Redistribution Vs. Predistribution
Consider two scenarios for an AI-driven future. In the first, those positioned to leverage automation at scale become extremely wealthy, while those around the world without the necessary skills, capital or infrastructure access are unable to participate meaningfully in production. Costs plummet, but so do wages; most struggle to reap the benefits of the material abundance that advanced AI helps create. People live off UBI in order to meet their needs, hoping the payments aren’t reduced.
In the second scenario, productive capital is distributed well before advanced AI arrives: People worldwide have access to fundamental digital infrastructure (e.g., internet, stable electricity and affordable devices), AI tools and the skills to use them effectively. Some people start AI-powered businesses, working a few hours a week, but mostly delegating tasks to AI agents. More people globally have the opportunity to participate in an AI-powered economy, wielding this technology not just to contribute economically, but also to achieve their own goals outside of work. The first scenario makes people dependents; the second makes them participants.
The first approach is redistributive; the second is predistributive. Predistribution is the idea of preventing inequality in the first place by enabling equal access to income-generating resources. Enabling equality of opportunity, ex-ante, over enforcing equality of outcomes, ex-post, creates deeper structural change than after-the-fact corrections ever could.
UBI implemented early enough could itself be predistributive — if it enabled people to access relevant technology and invest in their capabilities before AI-driven disruption. But it’s unfortunately plausible that the political will to implement such a large-scale redistribution scheme won’t be built until, say, labor markets collapse. Implementing UBI under conditions of extreme inequality would mean that those who need it have minimal bargaining power to ensure its adequacy or stability.
The Case For Predistribution
The philosopher John Rawls championed a “property-owning democracy,” (a concept borrowed from Nobel Prize-winning economist James Meade) where productive capital — both capital assets and human capital, i.e. skills and education — are widely distributed across society. This is complemented by institutional arrangements that protect basic liberties and fair equality of opportunity. Productive resources can be dispersed through progressive taxation, inheritance reforms that promote greater equality of opportunity, and promoting access to education, healthcare and housing.
“Predistribution is the idea of preventing inequality in the first place by enabling equal access to income-generating resources.”
Widespread access to resources and property ownership would not only, as Rawls put it, “put all citizens in a position to manage their own affairs” — that is, empower them toward economic independence — but also enable social cooperation and relatively equal political influence. In short, this spreads economic and political power around. Contrast this with welfare-state capitalism, which redistributes income to maintain a minimum standard of living, but often leaves wealth concentration intact, perpetuating dependency and elite influence over political and social life.
Rooted in his principles in “A Theory of Justice,” Rawls concluded that property-owning democracy was superior, not only to welfare-state capitalism, but to every other system he analyzed (e.g. laissez-faire capitalism, various forms of socialism). Society is most just when capital is accessible to all citizens.
For us, predistribution is shorthand for the practical implementation of these Rawlsian ideals — a framework of policies that satisfies both justice and pragmatism. From educational access to capital endowments, predistributive measures champion human agency by enabling economic participation. Beyond ethical appeal, predistribution offers a more efficient mechanism for change. It sidesteps the political resistance that inevitably undermines large-scale redistribution when attempted after wealth has concentrated, and by intervening upstream in the economic process — before inequalities become entrenched — even modest initial investments can shift more power and resources over time than redistribution ever could.
The U.S. has done predistribution before, and it has worked. The GI Bill, also known as the Serviceman’s Readjustment Act of 1944, didn’t just compensate veterans — it provided them with access to education, housing and business loans, helping to create the largest middle class in American history. In the 19th century, the expansion of universal primary education and establishment of land-grant colleges created a skilled workforce that could participate in industrial growth. These initiatives didn’t just redistribute income — they gave people the tools to generate it themselves.
These capital-based approaches dovetail with the idea of universal basic capital (“UBC”), which focuses on distributing wealth-generating assets to all citizens, rather than regular cash payments like UBI. Joseph Stiglitz, Ray Dalio and Nathan Gardels have all previously advocated for UBC in Noema. Wealth transfers (UBC) are better at addressing inequality at its root than income transfers (UBI): Wealth transfers provide assets that can appreciate, generate returns and give recipients ownership stakes in the economy, while regular income transfers maintain dependence. The post-war GI Bill and Federal Housing Administration’s mortgage loans enabled significant homeownership, generating trillions in household wealth that continue to benefit the descendants of these beneficiaries today. One-time capital infusions can have multigenerational effects that monthly income streams cannot match.
We can be creative about it, too; capital can come in many forms. Diverse research from Nobel laureates like Amartya Sen and Elinor Ostrom, as well as studies on various forms of asset-based welfare, underscore the benefits of investing in human capital and ensuring broad-based ownership. Societies with widespread access to education, healthcare and capital generally exhibit higher social mobility and more robust democracies. There are clear political, social and economic benefits of investing in everyone, and of everyone having a stake.
Why AI Demands Predistribution
There are three reasons why we need an economic policy for transformative AI today that enables broadly distributed opportunity to contribute to AI value generation, whether through access to productive capital (e.g. the computing power and necessary hardware, accessible AI tools) or investment in human capital (e.g. AI literacy and skills programs).
First, AI represents an unprecedented opportunity for economic transformation. If AI progress leads to the material abundance its proponents envision, predistribution could be an inflection point for correcting — rather than exacerbating — existing economic inequalities and leveling the playing field between rich and poor globally. But without baseline technology and skills, many will be excluded from AI’s benefits.
Second, advanced AI is being pursued as a universal amplifier of cognitive ability, capable of disrupting not just one industry but potentially all of them. This heightens the challenge of relying on redistribution, which may be required on a never-before-seen scale. While there is significant disagreement among economists, some anticipate near-total unemployment from AI. But if even, say, 15% of people face job displacement or declining wages, while returns increasingly flow to AI capital owners, welfare-state capitalism — already intrinsically flawed in Rawls’s view — would likely lack the broad base of support necessary to fund itself, and could fail entirely. And even if welfare-state capitalism survives, it cannot address the root problem of continually generated inequality from an economy defined by unequal access to, or ownership of, AI resources.
“From educational access to capital endowments, predistributive measures champion human agency by enabling economic participation.”
Third, because the technology is advancing especially quickly, inequalities can be exacerbated and become entrenched just as quickly. By the time we realize we need serious redistribution, the power to implement it may be gone.
The Political Stakes
The political and social ramifications of AI automation are immense. It’s not just about wealth concentration; it’s about whether the interests of labor are represented. If business elites no longer need large workforces to generate profits, they may become less inclined to politically compromise with working people.
The same goes with labor interests — i.e. people’s interests — being represented in government. Today, democratic accountability partly comes from the fact that citizens are taxpayers. Governments need our productive output to function. But if AI systems generate most wealth and states shift from collecting taxes to distributing benefits, citizens become passive recipients or clients, not stakeholders. “No taxation without representation” also means “no representation without taxation.”
This kind of political-economic inequality may present most severely between countries. In the more extreme scenarios, traditional development models may become obsolete and less developed countries may have no alternative other than a position of permanent economic and political dependency — a new global feudalism. This presents a moral challenge for actors who have pledged to ensure that AI benefits all of humanity (see, for example, OpenAI’s mission statement).
If AI advancement continues to accelerate, and we don’t proactively enable inclusive progress, we could kick off a vicious cycle in which AI-fueled inequality steadily erodes both the economic and political power of almost everyone, everywhere. But if we all have a stake in AI, as AI’s share of the economy grows, we can create a virtuous cycle where economic and political inclusion mutually reinforce each other, leading to a more just and democratic world.
What Predistribution Looks Like
The predistributive solution space is broad. Many proposals are yet to be imagined. These range from pragmatic approaches like strategic AI infrastructure partnerships across countries to more ambitious ideas like regulating cloud computing as a public utility or creating sovereign wealth funds that give citizens direct stakes in AI.
Here are three examples:
- Investing in human capital
A predistributive approach begins with equipping people to participate in the AI economy. Global AI literacy programs and accessible AI tools are essential — accessibility means ensuring linguistic and cultural representation in large language models (LLMs) and designing intuitive user interfaces.
Additionally, AI could be applied to boost social mobility. Imagine AI used to improve education, skills training, global health diagnostics, job matching and access to government services — tools that augment and extend human capabilities, rather than replace them. Market-shaping mechanisms like prizes and advanced market commitments could steer AI development toward public goods that market forces might otherwise neglect.
- Investing in productive capital
In digitally underserved regions, multilateral funds could finance digital public infrastructure — AI computing resources, electricity, high-speed internet, payment systems, etc. This could be funded through partnerships between governments, international development agencies and technology companies. Such entities could also collaborate to treat compute (the processing power derived from specialized chips necessary to train and deploy AI models) as a public utility, with universal basic allocations for citizens and organizations, though this would require significant international coordination as countries compete over the scarce resource.
To broaden access to AI systems, developers could implement region-based differential pricing — offering more affordable access in lower-resource settings and charging higher prices in regions with greater ability or willingness to pay. In parallel, states at the forefront of AI development could provide export financing and technical assistance to promote the uptake of AI compute infrastructure, models and applications in emerging markets. These international support measures may prove especially important if the cost of running frontier-scale compute grows faster than improvements in compute efficiency.
- Financial instruments
Many financial instruments can enable greater ownership in the AI supply chain. A national AI sovereign wealth fund — modeled after Norway’s oil fund or Singapore’s global investment fund, Temasek — could turn automation into broad-based prosperity. The UK Day One Project’s proposed AI bond offers one blueprint: Citizens buy bonds to invest in the national AI ecosystem, which reaps both fixed returns and an “AI dividend” for citizens and public welfare projects.
But bonds are just one option. Funding could also come from taxing AI profits or through equity transfers. Much foundational U.S. AI research has been funded through the National Science Foundation, the Defense Advanced Research Projects Agency, etc; AI companies could contribute equity shares based on their use of publicly funded research and infrastructure.
“If we all have a stake in AI, as AI’s share of the economy grows, we can create a virtuous cycle where economic and political inclusion mutually reinforce each other, leading to a more just and democratic world.”
There may be incentives for contributing equity; perhaps a large trucking company seeking to transition to AI-enabled autonomous vehicles could contribute equity in return for regulatory approval for autonomous vehicle fleets, tax credits for retraining truck drivers or discounted loans for managing the technological transition. The fund could then fund general public dividends or services, or even specific transition assistance for displaced workers. Imagine truck drivers receiving both transition assistance and shares in the autonomous trucking companies replacing them.
Cities could also create local AI investment pools, similar to how the city of Austin’s public energy utility enables citizen ownership in renewable energy. Austin Energy has programs like Community Solar, Solar Standard Offer and Solar for All, which collectively enable citizens to invest in shared solar infrastructure, turn rooftops into revenue-generating assets, and build community wealth through distributed ownership models that democratize access to clean energy across economic backgrounds. Similar programs could be created for AI. On a smaller scale, transitioning to worker-owned companies or employee stock ownership plans can ensure that workers maintain a stake in their companies even as AI reduces human labor requirements. Worker-ownership models like Spain’s Mondragon Corporation, a federation of worker cooperatives operating in diverse sectors, have historically prioritized preserving employment and distributing profits equitably, principles that could help buffer the economic impacts of AI automation.
Our Current Crossroads
When ExxonMobil’s Esso first tapped the enormous North Sea oil reserves off Norway’s coast in 1967, Norway’s government stepped in to assert their sovereign rights over the oil. What to do with this discovery of such wealth? The Norwegian government wanted the wealth to be distributed broadly and across time, rather than creating a small class of ultra-wealthy individuals who happened to live in the extraction period, something that often occurs in resource-rich countries. Oil is now taxed at 78%, flowing into Norway’s sovereign wealth fund which supports extensive social policies and maintains one of the world’s lowest levels of inequality.
This is one way to deal with an economic transition. Another is the approach taken by factory owners and the British government during the Industrial Revolution. Where mechanized looms displaced textile workers, factory owners hoarded the gains, and the British government met protests with bullets and workhouses. The Luddites weren’t anti-technology — they were anti-exploitation. They resisted an unfair distribution of technology’s benefits, not the technology itself.
AI stands at a similar crossroads. Maybe its gains will build hospitals and safety nets; maybe it will replicate some of the Industrial Revolution’s worst chapters. Predistribution offers a way to realize the former. By ensuring productivity gains from AI automation are broadly shared, we can embrace technological advancement while mitigating the social upheaval that characterized the Industrial Revolution.
Some say that if AI automates all work, the idea of humans participating in wealth generation becomes meaningless. On the contrary, it is in such extreme circumstances that predistribution becomes even more important — people may not work, but they may own a share of AI assets, delegate work to AI, or use AI to fulfill non-work aspirations. Every person could have a slice of the (presumably enormous) pie.
This is not just about the hard levers of the economy and politics. It’s about broad-based inclusion in shaping the future generally — shaping what we actually want to see in the world. The iPhone was not a response to market demand; Steve Jobs, famously, was very good at deciding what people should want. Choices made by companies create new realities. But broader ownership means diverse decisions about what technologies define our world.
The window for action may be slim, and it may be now. Once AI-driven inequality takes hold, accumulated power may be used to circumvent any redistributive mechanisms. We need to act now to ensure broad participation in shaping how AI develops — not just through profit-sharing, but through genuine power-sharing that enables communities worldwide to use AI to rise together.
