Will AI be an economic net positive? Probably not

THE boom in artificial-intelligence (AI) investment is undoubtedly boosting both the US stock market and the broader economy right now. But what about the longer term? Will AI be a big net positive, delivering prosperity and solving the nation’s fiscal problems?

Most likely, I’m sorry to say, it’ll be closer to a wash.

To assess AI’s longer-run implications for the economic outlook, one must consider three questions:

How much will it lift productivity and growth? How will it change the demand for labour and the equilibrium unemployment rate? How will it affect interest rates?

On productivity, there’s no consensus. It’s still hard to say what kinds of work AI can handle and how quickly businesses can reconfigure their processes to take full advantage.

The advent of electricity, for example, took several decades to transform manufacturing; of the computer age, the economist Robert Solow famously said in 1987 that one could see it “everywhere but in the productivity statistics”.

Nobel laureate Daron Acemoglu argues that AI won’t have a significant impact on most tasks performed by most workers and hence will boost the level of productivity by less than 1 per cent over the next decade. Economists at Goldman Sachs, by contrast, estimate a 15 per cent boost when it’s “fully adopted and incorporated into regular production”, which might take considerably longer.

Most likely, the impact will be very modest at first and then grow over time as businesses put the technology to use – first internally, where the risks of hallucinations and bias can better be managed, then externally in client-facing products and services, where reputation risk and legal liability are more of a concern.

With regard to the labour market, one key issue is whether AI will replace workers or just help them do their jobs better. Probably a bit of both. Consider coding: AI will speed it up, reducing the number of workers required to do a given project; but it also should boost the demand for software by lowering the cost of development. The total demand for coders will depend on which of those two effects dominates.

A second issue is friction. If coders lose their jobs, how easily can they retrain and find employment elsewhere? The pace of change matters a lot. If AI displaces many workers in a short period, more of them will be in limbo as they seek to move into new jobs and professions. In that scenario, equilibrium unemployment – the rate consistent with stable inflation – will be higher.

AI’s effect on interest rates is more straightforward. Historically, technological innovation has both increased the prospective return on investment and required new kinds of investment to adapt to new ways of doing business. The added demand for capital, in turn, pushes up real interest rates.

During the internet boom of the late 1990s, for example, spending to build the national fibre optic network pushed real interest rates well above 3 per cent. In the case of AI, the impact will likely be frontloaded, given the scale and speed at which computing and energy infrastructure must be built.

It’s hard to see how AI’s effects, taken together, will cure the US government’s unsustainable fiscal trajectory.

Higher productivity and growth will be positive to the extent they happen, generating more tax revenue and boosting the denominator in the country’s debt-to-gross domestic product ratio. But higher interest rates will increase debt service costs, which will push in the opposite direction.

No doubt, AI is important. Yet it’s unlikely to dramatically change the nation’s long-term economic outlook. One can hope for the best, but hope is not a strategy. BLOOMBERG