Why the Fed's wait-and-see approach to AI can't last

Financial markets are obsessed with AI, and the broader public is aware of its looming impact on jobs and wages. Yet for the Federal Reserve, the concern has barely registered. This isn’t because its policymakers think AI doesn’t matter. It’s because they have no idea how its potentially vast repercussions will play out — and no good tools for shaping the outcome.

Last week, as Amazon cut thousands of corporate positions and Nvidia’s market capitalization climbed to $5 trillion, Chair Jerome Powell was asked about the subject during the press conference that followed the Fed’s latest policy announcement. A significant number of companies, he said, “are either announcing that they are not going to be doing much hiring, or actually doing layoffs, and much of the time they’re talking about AI and what it can do. So, we’re watching that very carefully.”

That was about it.

AI is already having conflicting macroeconomic effects. The stock market’s unshakable exuberance, which relaxes the “financial conditions” the Fed has to weigh when it sets its policy rate, rests on optimism about the big AI innovators. Massive outlays on AI software and equipment are powering overall demand. At the same time, the job market is limping thanks to layoffs and a marked reluctance to hire — most likely caused, in part, by companies waiting to see how AI will change their businesses and their need for workers.

As yet, these disturbances have no clear implication for monetary policy. On the face of it, a bubbly stock market and elevated spending on AI raise demand and call for higher interest rates. A sluggish labor market suggests the opposite. For now, doing nothing except “watching that very carefully” looks reasonable.

Before long, this stance will be hard to sustain. The likely effects of AI have what you might call quantitative and qualitative dimensions. AI could be a big deal or not such a big deal. And it could be a good thing or a bad thing. Small effects, good or bad, would complicate the Fed’s job by muddying the signals it uses to guide policy, but needn’t keep its policymakers up at night. Big effects, good or bad, could make their job — combining “maximum employment” with 2% inflation — much harder, and perhaps impossible.

Unsurprisingly, expert opinion is divided on where things will go, but suppose for the sake of argument that AI proves to be a truly transformative technology — on a par with the adoption of electric power. That would suggest a boost to productivity growth of maybe one percentage point a year for the next several decades. This supply-side push in turn implies a higher long-term neutral real interest rate (known as r-star), combined with lower inflation because of falling costs of production. The Fed’s appropriate response, given its inflation target of 2%, would be to cut the policy rate, but by less than the fall in inflation.

Getting that right is already tricky, but that’s just the start. So far, this scenario is one of good big effects: It assumes higher productivity plus the implicitly smooth adoption of displaced workers into different jobs, with little else changing. But big effects could easily turn bad.

The AI revolution could replace certain kinds of workers abruptly and en masse, as opposed to gradually making workers in general more productive. True, similarly dire predictions have been made about every wave of innovation, only to be proven wrong (eventually). Still, AI seems more directly aimed at saving labor through automation than many of its predecessors, and might arrive much faster and more forcefully.

Lower wages, higher unemployment and rising inequality as the labor force divides between workers empowered and disempowered by AI would pressure the Fed to do something — but what, exactly? Its only policy tool, the short-term interest rate, is beside the point in confronting a structural shift of this kind. The Fed can always engineer higher inflation; it can’t always raise real wages or find jobs for the unemployable.

Another plausible bad effect is greater price-raising power for the technology’s leaders and earliest adopters. As with collapsing demand for labor, this isn’t to be taken for granted: In the end, an open-source AI ecosystem might emerge to keep would-be AI monopolists in check and promote competition more broadly. At the moment, though, network effects and economies of scale appear to have the upper hand. Economic prospects hinge on how this turns out. The Fed can only stand and watch.

Short-term macro management is squarely in its wheelhouse, but the difficulties aren’t confined to ambiguous economic signals and a fluctuating neutral rate. AI could also amplify the short-term business cycle by making it easier for employers to cut jobs in a downturn. If supply chains are designed and managed by AI, and fail because of shocks not previously encountered, the technology might make things worse — and there’ll be fewer humans to put things right. The same goes for financial markets. Once they’re guided by models trained on history and opaque to human judgement, radical novelties (such as AI) could throw them for a loop, causing bigger, faster, self-reinforcing errors. The next crash could break records.

Call me an alarmist, but tweaking the federal funds rate will not suffice. Handling these possibilities demands smarter policy across an unnervingly wide front: a stronger safety net to cushion the blow for displaced workers; judicious regulation to temper the threat to competition; tax reform to guard against surging inequality; labor-market reform to improve occupational mobility; and, above all, schools and colleges capable of training students before their careers begin and throughout their working lives.

If things go badly, the Fed will doubtless get more than its fair share of the blame. But the truth is, AI is too big for any central bank to cope with on its own. The technology calls for politicians willing to think hard about these challenges and face them. Let me know if you see one.