"Data Dividends" by Nathan In

Data Dividends: Bridging the Divergence between Productivity and Wage Growth

For much of modern economic history, productivity carried an implicit promise. When workers became more productive, their wages would eventually rise with them. New machines could eliminate certain jobs, but they also created industries, raised output, and increased the value of human labor. Economic growth and individual prosperity were never perfectly aligned, but they were connected closely enough to sustain a basic expectation: when an economy produced more, the people producing it would share in the gains. Artificial intelligence is beginning to test that expectation. The problem is not simply that AI makes work faster. It can produce value while requiring less human labor to do so. A company can increase output by investing in algorithms, computing power, and data without making a comparable investment in workers. As more of the productive process moves from labor to capital, those who own these systems may capture a growing share of the gains. Productivity can rise even while wages stagnate.

This changes the question we should be asking about automation. It is no longer only What happens to jobs when machines become more capable? It is also Who owns the value those machines create? The release of ChatGPT in late 2022 made this question difficult to ignore. Previous forms of automation had already replaced repetitive physical and administrative tasks, from factory assembly to cashier checkout. Generative AI extended automation into writing, analysis, coding, translation, and other tasks once assumed to depend on distinctly human abilities. A kiosk replacing a cashier is easy to see. An algorithm absorbing fragments of a researcher’s, translator’s, or customer-service worker’s job is less visible. Yet the economic logic is similar. If a task can be performed more cheaply by a machine, businesses have an incentive to automate it. The difference is that AI does not remain confined to routine work. It can substitute for some forms of labor while making highly skilled workers considerably more productive.

This does not mean human work will disappear. Technology has repeatedly destroyed some occupations while creating others. But even if AI creates new employment, the transition may alter how income is distributed. A worker earns wages by selling labor; an owner earns returns by possessing an asset. If increasingly productive assets are algorithms, models, computing infrastructure, and proprietary datasets, an economy can become more productive without making its workers proportionally wealthier.

One response is Universal Basic Income. Its appeal is straightforward: if automation weakens employment as the primary way people receive income, governments can establish another floor beneath them. Everyone receives a regular payment, regardless of employment status. But security and participation are not quite the same thing. UBI redistributes money after economic value has already been created. It does not fundamentally change who owns that value. The automated economy continues to produce wealth through privately owned technologies, and the government later redistributes part of it through taxation and transfers.

Data dividends begin from a different premise. They ask whether individuals should have an economic claim on something that helps make the digital economy productive: their data. Every search, purchase, route, click, photograph, preference, and interaction can become part of a larger informational system. Individually, these traces may appear insignificant. Aggregated across millions of people, they can become commercially valuable, helping companies understand behavior, improve products, predict demand, and refine algorithms. Yet the person generating the data rarely participates directly in the value that follows from it. A data-dividend system would attempt to change this relationship by treating data not merely as something companies collect, but as an asset to which its producers retain some economic claim. When an asset produces value, those with a recognized stake in it receive part of the return.

Compensation could also change the quality of what people contribute. Free data is abundant, but abundance does not guarantee usefulness. Datasets can be incomplete, mislabeled, duplicated, or stripped of context. If individuals had a reason to provide accurate, structured, voluntarily shared data, they might become deliberate contributors rather than passive sources from which information is extracted. This does not make data dividends simple. Determining what a piece of data is worth would be difficult. One person’s information may have little value alone but substantial value when combined with millions of other records. A workable system would require rules for valuation, consent, privacy, portability, and collective ownership. These complications reveal something important. The debate is not merely about finding another source of cash for people displaced by automation. It is about deciding what counts as economic contribution in a digital society.

This is where UBI and data dividends part ways most clearly. UBI separates income from contribution by design. That is also its strength: a person does not need to prove productivity to deserve basic economic security. But universality raises difficult questions of cost. Funding meaningful payments for an entire population requires substantial taxation, cuts elsewhere, new public revenue, or some combination of the three. The problem becomes especially difficult in countries with narrow tax bases, large informal economies, weak administrative systems, or rapidly ageing populations. Data dividends frame the problem differently. Instead of asking the state to redistribute income solely because citizens exist, they ask whether people should be compensated because they contribute an increasingly important input into digital production.

Consider the strange economics of an ordinary day online. A person checks the weather, searches for a restaurant, walks through a city with a phone in a pocket, watches a video, buys a pair of shoes, and sends messages to friends. None of these actions feels like work. Yet together they generate information about location, preferences, language, consumption, attention, and social behavior. At sufficient scale, these patterns become economically useful. We have built an economy in which ordinary human behavior can be transformed into productive capital without the person producing that behavior necessarily recognizing the transaction. The question is not whether every click deserves a check. It is whether the economic architecture of AI should continue to assume that human-generated data is effectively free.

By establishing some form of property or participation right over data, data dividends could preserve a relationship between contribution and compensation even as traditional labor becomes less central to production. People would not simply receive money because automation had displaced them; they would retain a stake in one of the resources from which automated systems derive value. There is also a deeper difference between the two approaches. UBI primarily treats automation as a problem of lost income. Data dividends treat it as a problem of ownership.

That distinction will matter if AI continues to raise productivity without producing equivalent wage growth. The central economic challenge of automation may not be a shortage of wealth. AI could make societies extraordinarily productive. The harder question is how that wealth travels from the systems that generate it to the people whose lives, choices, language, behavior, and knowledge helped make those systems possible. For the industrial economy, labor law established rules around the ownership of human work. For the digital economy, we may need an equivalent conversation about the ownership of human data.

The goal should not be to slow technological progress until wages catch up, nor to accept a future in which productivity belongs increasingly to machines and income must simply be handed back to people afterward. A more durable response would reconsider where economic value begins. If data has become a factor of production, then the people who generate it should not remain invisible within the production process. The widening distance between productivity and wages is ultimately not only a question about artificial intelligence. It is a question about what we decide belongs to us in an economy increasingly built from traces of ourselves.