The AI age requires inequality insurance

European Union lawmakers recently reached a provisional agreement on a landmark regulation to mitigate the risks that artificial intelligence (AI) poses to humanity, and other countries seem ready to follow the EU’s lead. But this regulation does not address one of the greatest disasters AI may bring – the prospect of mass unemployment and sharply increased income inequality. Regulation cannot eliminate these risks without precluding the world from enjoying AI’s potential benefits, namely dramatic increases in productivity and enormous wealth creation. That is why policymakers must also enact policies to compensate citizens if these disasters occur.
Let us be clear: we are not opposed to regulating AI. But just as we take a two-pronged approach to protect flood-vulnerable homes – building sea walls and providing flood insurance – so, too, must governments offer inequality insurance to ensure that AI does not widen the wealth gap. While future administrations could conceivably change the terms of such a program, cutting back on widely experienced benefits would be politically difficult.
The writing is already on the wall. Last year, Hollywood actors joined screenwriters in the first industry-wide strike in more than six decades, with safeguards against generative AI being one of their main demands. But AI will revolutionise the future of work for all types of professionals, from doctors and lawyers to taxi drivers and checkout clerks, and the subsequent increases in total output will not be shared equally. Those who make and own the inventions could amass immense wealth, much of which will come from economising on labour costs.
It is tempting to believe that AI will turn huge swaths of the workforce into hamburger flippers, but even that job is being automated. Perhaps there will be other low-value services for people to provide. If not, the ranks of the chronically unemployed will swell. Either way, income disparities will almost surely deepen.
Policymakers can limit, or even prevent, the resulting increases in inequality by reforming the tax system. For example, if inequality exceeds certain limits, federal income-tax rates on high earners could increase automatically. To stop inequality from rising further, taxes on the top 1% could be set each year to ensure that their share of total national after-tax income never rises above current levels.
To be sure, if AI causes truly catastrophic increases in inequality – say, if the top 1% were to receive all pretax income – there might be limits to what tax reforms could accomplish. Consider a country where the top 1% earns 20% of pretax income – roughly the current world average. If, owing to AI, this group eventually received all pretax income, it would need to be taxed at a rate of 80%, with the revenue redistributed as tax credits to the 99%, just to achieve today’s pretax income distribution; funding the government and achieving today’s post-tax income distribution would require an even higher rate. Given that such high rates could discourage work, we would likely have to settle for partial inequality insurance, analogous to having a deductible on a conventional insurance policy to reduce moral hazard.
Such a lopsided income distribution is unlikely, and in less extreme cases, full or nearly full inequality insurance would be possible and warranted. But while this thought experiment highlights a weakness of our plan – it cannot provide full insurance in the most catastrophic cases – it also points to the importance of building some form of automatic insurance into the tax system. After all, most people would agree that if economic inequality does spike, the 1% should be taxed at a substantially higher rate than they are today.
To deal with the massive unemployment that AI may bring, many – from Juliet Rhys-Williams in 1943 to US presidential candidate Andrew Yang in 2018 – have proposed a guaranteed minimum basic income for everyone, regardless of what they do. Other economists, such as the Nobel laureate Edmund Phelps, have instead suggested subsidising wages by expanding the negative income tax for low-income families or by adjusting corporate tax deductions. Either way, these plans require funding, and building inequality insurance into the tax system could be a long-term adjustment.
Even though our proposal does not cap the amount of money people can earn or save, we are under no illusions that establishing inequality insurance will be politically easy. But too much is at stake not to try.
Explicitly defining tax rates to provide insurance against extreme inequality was a good idea when one of us first proposed it two decades ago. But it is a much better idea today. To reap AI’s benefits, we must prepare for a potentially catastrophic increase in disparities of wealth and income.