Can artificial intelligence write a song about hamburgers in the style of Taylor Swift? Yes, in mere seconds. Can it pay for the substantial dislocations it’s likely to impose on global workforces? Probably not for years. Policymakers and economists are already debating how to regulate AI, and companies are rushing to understand how they can profitably put technologies like ChatGPT to use. A less racy question is also the important one: how to tax it.

The chances of “generative AI” being put back in its box are very small. Because it can do cognitive tasks quickly that humans do slowly, and because it is rapidly honing its skills, it will increasingly enhance the capabilities of workers, or replace them. Goldman Sachs economists estimate that 18% of work could be automated globally, and that 7% of the U.S. workforce might be substituted by AI. A paper led by OpenAI scholar Tyna Eloundou suggests that half of workers could find half their tasks “exposed to” so-called large language models.

Over time, the nightmare scenario of workers made obsolete by machines is unlikely to come to pass. The “lump of labor” idea, that there is only so much work to go round and therefore what’s given to a machine must be taken from a human, isn’t borne out by the experience of previous technological leaps, like the automobile or the internet. New jobs have been created; workers have retrained or relocated.

But retooling takes time – and possibly many years. While job creation roughly matched destruction from new innovations for decades from the 1950s, that changed in the 1980s, Goldman found. A stark example of the frictions that occur is the “China shock”, where cheap Chinese exports in the early 2000’s idled millions of American workers and hollowed out their communities. The effect basically plateaued by 2010, and in the long run, cheaper goods and Chinese demand helped create millions of new service-sector posts. But resentment over American jobs lost to the People’s Republic was still on view at the ballot box a decade later.

The challenge, then, is to cushion the impact for workers who find themselves out in the cold, even if temporarily. It’s a new kind of dilemma, because those who will bump up against AI aren’t factory workers but knowledge-economy employees, like lawyers and other professional services staff. Geographically, they may be more centered on big cities than small communities. They are also wealthier in relative terms, which gives them political clout.

The answer to this is likely to be money: unemployment benefits, healthcare, and even straight-up cash. Universal basic income – unconditional, non-means-tested cash transfers – may come back onto the agenda. But if government largesse is needed, the timing couldn’t be worse. Politicians in the United States are already squabbling over the self-imposed debt ceiling, and federal debt is projected to hit nearly 120% of GDP in a decade. European states like Spain, France and Italy have debt surpassing 100% of their GDP. Interest rates are rising almost everywhere, making it more expensive to borrow.

Moreover, while AI will create profit windfalls, many countries don’t tax those as effectively as they ought to. A dollar of lost earnings for a well-paid worker deprives U.S. governments, state and local, of around 30 cents in taxes, based on the Organisation for Economic Co-operation Development's calculations. But a dollar of saved costs for a company creates only 21 cents in corporate income tax revenue. Dividends are taxable too, but only around a quarter of investors in U.S. companies are actually subject to that tax, Brookings Institution fellow Steven Rosenthal calculated in 2020. And share buybacks, which globally hit a record $1.3 trillion in 2022 according to Janus Henderson, are even less conducive to topping up the public purse because of their favorable tax treatment.

There are other glaring holes that AI could lay bare. Capital gains are still taxed below the level of income in most countries. So an entrepreneur whose business swells in value because of AI-related activities will pay relatively little if they sell shares in their company. A more egregious loophole is the American “step-up in basis”, which President Joe Biden tried and failed to scrap. When a company owner passes their empire on to an heir, the embedded capital gain resets to zero, slashing the recipient’s tax bill if they sell later. Raising the capital gains rate to 28%, imposing the tax at the point of inheritance and making other tweaks could raise $185 billion over a decade, according to a 2022 analysis by the Wharton School of the University of Pennsylvania.

Finally, there’s the question of how to stop profits from AI being whisked to lower-tax jurisdictions. The OECD has struck a deal that intends to make sure every large multinational company pays a minimum of 15% tax, regardless of where it’s based. But that 15% is still a lot lower than the worldwide average of 23% levied on corporate profit, not to mention the taxes that many white-collar workers pay on their earnings. Prior ideas for a “robot tax”, whose advocates include Microsoft (MSFT.O) founder Bill Gates, have mostly been written off as too complex. For one, there’s the challenge of defining what exactly constitutes a robot.

All these raise the specter of a tech boom that concentrates gains and socializes losses. True, from the invention of the printing press to the cellphone, humans have found new ways to thrive and use disruptive technologies in life-enhancing ways. But the process isn’t always smooth or equitable. Revisiting the way profit and capital gains are taxed might sound like the least exciting part of the AI story, but it’s one that deserves urgent attention to make sure an economic good doesn’t leave a trail of social carnage.

LeackStat 2023