AI and inflation: how worried should central bankers be?

I resent paying more for annoying software features. I don’t want an AI-generated summary of a long document, especially if that document is a book I spent considerable effort writing. Working out how to switch off AI assistants takes minutes of my life I’ll never get back. Pity the policymakers gripped by fear of a similar, economy-wide phenomenon: could AI cause inflation without delivering productivity benefits?

For now at least, AI doesn’t seem to be the primary driver of overall inflation dynamics. Figures from Capital Economics put the annual rate in August across major developed markets at around 3 per cent, uncomfortably high. But of that, energy prices contributed almost one percentage point. That may be partly because of the energy needed to power ChatGPT requests for “easy cupcakes bananas yoghurt no eggs”, but the war in Iran is probably playing a bigger role.

Although there are some signs of narrower, AI-related price pressure, they’re not always particularly strong. Despite hot global demand for chips, British prices of accessories for information-processing equipment and mobile phones haven’t risen over the past year. ECB chief economist Philip Lane recently noted that euro-area inflation was protected as affected phones, laptops and cars were such a small share of spending.

AI’s effects are more obvious in the US, where producer prices of the chips that furnish data centres have been soaring. US prices of computer software and accessories have been climbing too, though again, the category makes up a small share of spending. And some of the price increase could reflect mismeasurement, if it does actually reflect software becoming better.

Earlier this month, economists at Goldman Sachs took a more systematic approach, totting up estimates of AI’s effects on different components of spending in the US, both direct and indirect. They counted four: on components and software directly relevant to AI; on electronics competing for those components; on electricity; and on the construction materials that go into making data centres.

Overall, they estimated that AI-related price pressures were adding around 0.5 percentage points to annual headline inflation. That fell to around 0.3 percentage points after they stripped out some mismeasurement related to the way different spending categories were being combined in the official statistics. That’s meaningful, but not massive. And the analysts noted that since they couldn’t remove all tariff effects, the true effect could be lower.

The true effect of AI on inflation might be smaller still, if the technological revolution is squashing prices indirectly, through its effects on interest rates and spending. Think of a protein craze that pushes up the price of some snacks relative to others. If households cut back on other foods, retailers might lower their prices, leaving the overall price of food unchanged. Similarly, AI could affect relative prices without raising the overall price level.

Don’t feel too soothed. In theory, industry-specific shocks can raise the overall level of inflation, if the industries experiencing an increase in demand are unable to raise production quickly to match that demand and so have to raise prices a lot instead. One study suggests that it is relatively common. Elisa Rubbo of the University of Chicago estimates that between 2000 and 2019, roughly half of inflation fluctuations were explained by relative price shifts.

There would be precedent, then, for AI raising inflation, complicating life for central bankers trying to keep prices under control. They don’t want to raise interest rates too far, only to delay the rollout of a technology that could deliver productivity gains — and price cuts. Other research suggests that within industries, inflation tends to rise in the run-up to productivity booms, but then falls soon after. (I look forward to software becoming cheaper, or a competent AI assistant.)

So how worried central bankers should be depends on the answer to a trillion-dollar question: will AI actually deliver the hoped-for productivity gains? Business executives seem increasingly positive, though this could reinforce yet another pathway through which AI could raise prices.

In theory, households and business executives might become so excited about an AI-fuelled productivity boom that they raise spending and investment in anticipation, heating up the economy and boosting inflation that way. That should be temporary — unless the expected productivity gains never materialise. In which case, the economy could end up sluggish, with an inflationary hangover anyway. And with hordes of annoying AI assistants distracting us from books.

soumaya.keynes@ft.com, @SoumayaKeynes

The Economics Show with Soumaya Keynes is a podcast from the FT bringing listeners a deeper understanding of the most complex global economic issues in easy-to-digest weekly episodes. Listen to new episodes every Friday on Apple, Spotify, Pocket Casts or wherever you get your podcasts