How to share the AI windfall

Should artificial intelligence cause mass unemployment, workers will not be thrilled. But neither will the taxman, even if he hasn’t been automated. For most of the past century, rich countries have had simple rules for sharing prosperity: raise money mostly by taxing work and consumption, sprinkle in some borrowing and hand out the proceeds. That model may collapse if AI advances as quickly as its boosters suggest. Hence, many say, a new approach is needed, in which government makes its money primarily from the new technology.

Exactly how God-like AI would hit employment is widely debated. It might be that human workers would simply reshuffle and do things that AI cannot, in a much wealthier economy. After a painful adjustment, in this scenario, most people would still have good jobs. But you can also imagine a trickier outcome in which AI diverts to capital holders income that currently flows to workers. That would be a huge upheaval. Throughout modern history, the ratio between labour and capital has been a remarkably stable two-to-one, noted Nicholas Kaldor, an economist at Cambridge University, in the 1960s. America’s labour share has since fallen on some measures, but Kaldor’s observation has mostly held up.

If the pattern broke, governments would still need money—probably much more, if swathes of the population were jobless and needy. Whether states supported them with cash handouts, job training or something more imaginative, the funding would need to come from somewhere. But if the labour-share of income was low, the tax base would have collapsed. The average member of the OECD, a club of mostly rich countries, raises about half of its tax revenue from labour. Another 30% comes from consumption taxes, and the rest from a smattering of levies on corporations, capital and property. America, which lacks a national consumption tax, is especially reliant on labour.

This broadly follows the advice of tax economists. They tend to prescribe flat levies on consumption to raise revenue and differential income taxes (higher for higher earners) for redistribution. They frown on corporation taxes (which discourage investment) and other levies on returns to capital (which distort choices between current and future consumption).

Less income tax would mean both less revenue and less redistribution. The obvious replacement would be to substantially raise consumption taxes. People, including rich-as-Croesus AI capitalists, would still buy things after all, and governments could pass the proceeds off to others.

Capital taxes might also become more important. These would still distort the economy but could help with redistribution, since most capital returns accrue to the rich. An AI world might also offer more sources of super-normal rents (like land in the richest cities today), which could be taxed without twisting incentives to invest and thereby damaging growth.

Plenty of the flashier AI tax proposals that are floating around fit into this traditional taxonomy. Robot taxes, for instance, are levies on one specific type of capital. Token taxes, on the volume of text passing through AI models, are a form of targeted consumption tax. Both would give rise to unhelpful distortions, nudging investment and spending away from automation and so hurting growth (though proponents tend to argue that such distortions would helpfully prompt innovation in technologies with more human involvement).

A more radical option is to offer workers direct stakes in the progress of AI. One way to do this would be to spread ownership of the companies developing the technology more widely, whether by handing out shares or through sovereign-wealth funds. On the face of it, this seems a world away from taxation. But it would be similar in principle to redistributing the proceeds from a well-designed corporation tax. In both cases, after all, governments in effect grant their citizens stakes in corporate earnings.

A big advantage of conducting redistribution through taxation is that governments need not make guesses about which companies will benefit most from AI, which they would have to when buying equity stakes. And yet owning shares in AI’s winners might mollify displaced workers a lot more than a circuitous system of tax-and-spend. Equity stakes would also sidestep the issue of which country taxes rich companies—a boon given tech giants’ knack for shuffling profits to lower-tax jurisdictions.

In fact, this radical system of redistribution is not so far from reality. Many workers are already capitalists, with shares in their retirement or savings accounts. Any American with holdings in a fund tracking the S&P 500 share index is a fractional owner in Nvidia, Alphabet, Apple and 497 other big companies. Savers elsewhere in the world are in on the action, too. None needs to pick winners: in such funds, the most successful shares gain value and become a larger share of the portfolio, while the losers shrink.

Ideally, the biggest winners would be publicly listed, so outsiders could invest as easily as insiders. There is progress on this front, with both OpenAI and Anthropic preparing to list their shares soon. Even private markets are more democratic than many assume. The ultimate investors in many big venture-capital funds are Canadian savers (via vast pension schemes) or Singaporean taxpayers (via a sovereign-wealth fund). If governments are concerned about AI raising inequality, they could try boosting the shareholdings of the less well-off.

For all the upheaval that AI promises to bring, then, economics offers a comforting lesson: there are plenty of tools for sharing the prosperity that mass automation would create. Will politicians take them up? That is another matter entirely.■