Ask whether artificial intelligence is transforming the economy and you'll get confident answers in both directions. The honest answer, increasingly, is: we can't yet measure it well — a problem laid out by The New York Times.

What the data catches — and misses

When a company spends on AI, part of it shows up cleanly in the numbers: the data centers, chips and hardware land in the investment column of gross domestic product (GDP), the standard measure of everything an economy produces. That capital spending has surged, and it is visible.

What's far harder to see is whether all that spending is making the economy more productive. If an AI tool lets an engineer write code faster or a customer-service team resolve more cases, that gain often doesn't register in official statistics unless it turns into measurably higher output or revenue per worker — and many firms are still experimenting, not yet reorganized around the technology. (Productivity — output per hour of work — is what ultimately drives rising living standards; it's also notoriously hard to measure in real time.)

The productivity paradox, again

Economists have seen a version of this before. In the 1980s, as personal computers spread, the economist Robert Solow quipped that "you can see the computer age everywhere but in the productivity statistics." The gains eventually showed up — but only years later, after businesses reorganized around the new tools. This lag is now known as the productivity paradox.

AI may follow the same pattern, on a larger scale. The technology touches cognitive work across many industries at once, but each use is hard for statisticians to isolate. Labor data struggles too: it's difficult to tell whether a job disappeared because of AI or for ordinary reasons, so the technology's effect on employment blurs into the noise for months or years.

Why the gap is dangerous

The result is a widening gap between spending you can measure and impact you can't, and it matters well beyond academic debate.

For the Federal Reserve, uncertainty about productivity makes policy harder: strong productivity growth can let an economy run faster without inflation, while weak growth argues for caution. If the Fed can't tell which is happening, its decisions are less precise. For investors, the entire AI trade rests on a belief about future productivity: if the gains are real but hidden, patience pays; if the spending is huge but the payoff small — as happened after parts of the late-1990s tech boom — returns will disappoint. And for governments weighing worker retraining or tax treatment of AI investment, flying blind risks costly mistakes.

What's being done

Statistical agencies such as the U.S. Bureau of Labor Statistics and Bureau of Economic Analysis are adapting — adding AI-adoption questions to business surveys, refining how they classify jobs and capital — but such changes take time, as the BLS's own productivity work reflects. Recent U.S. productivity growth has been only moderate, neither confirming a boom nor ruling one out, which is precisely the ambiguity at issue.

Why it matters

The lesson of the last productivity paradox was patience: measurement eventually caught up, and the gains proved real. But the stakes are higher now, because the capital being committed is so large and the market's valuations lean so heavily on the promise of transformation. The longer official data stays foggy, the greater the risk that policymakers misjudge inflation or growth, and that investors misprice the boom in either direction. Boursel makes no forecast on AI's ultimate payoff; the takeaway is that one of the most important economic questions of the decade — is AI actually making us more productive? — is, for now, one the numbers cannot clearly answer — and acting as if they can is its own kind of risk.