AI并不认为人工智能是主要的通缩力量

Economics US Economic Perspectives AI is not convinced AI is a majordisinflationary force Matthew Luzzetti, Ph.D.Chief US Economist+1-212-250-6161 Introduction Brett RyanSenior US Economist+1-212-250-6294 In a recent note, we discussed how AI assesses the potential negative effects ofAI on the unemployment rate (see“Will AI drive unemployment higher this year?AI thinks not”). While several AI tools agree that the impact is likely to be limitedover the next year--all three tools we referenced assign a 5% probability AI willlift the unemployment rate by 0.5pp or more over the next 12 months--there wasdisagreement further out. The three tools assigned a 10-30% probability of moresubstantial disruption at the 5-year horizon. Justin WeidnerEconomist+1-212-469-1679 Amy YangEconomist+1-212-250-9959 In this notewetake the same approach to assess the impact of AI on inflation. Incontrast to the unemployment rate, where there appears to be considerableuncertaintyaboutthe size of the impact, our sense from client conversations isthat there isa strongconsensus that AI represents a substantial disinflationaryforce that willalsoexertmeaningful downward pressure on interest rates over thecoming years. Does AI agree with this consensus? Surprisingly not. Why AI is expected to be a significantdisinflationary force Thecase for AI to exert a meaningful drag on inflation is reasonablystraightforward. AI is expected to substitute cheap capital / technology formore-expensive labor, driving input costs lower. By leading to sustainably higherproductivity growth, it will push unitlabor costs substantially lower, weighing onprice pressures. To the extent that AI raises the unemployment rate, it will alsoreduce inflation throughincreasinglabor-market slack, reducing wage growth,andweakeningaggregate demand. Finally, AI represents an important equalizingforce for entrepreneurs, ushering in countless small businesses with lowerbarriersto entry that will compete and drive margins lower.This is ourunderstandingof the consensus view on why inflation and rates will besustainably lower due to AI. But there are countervailing forces to this narrative. In the near term at least, AIhas triggered a substantialpositiveaggregate demand shock, as investment hassurgedin anything AI-related,including software,chips,computers andperipheral equipment,and data centers. These investments are competing withalternative demands for resources. In addition, AIis likely torepresent an ongoingsource of demand that could create bottlenecks and price pressures in key pointsof the supply chain--the spillover to higher electricity prices is most obvious.Finally, while the initial impact of AI-driven higher unemployment would beweaker demand andsofterinflation pressures, we expect policy levers wouldbetriggered toultimatelylimit the rise in unemployment and put a floor under 30 March 2026US Economic Perspectives aggregate demand. Fiscal policy is best placed to serve this redistribution role--indeed strongerproductivity impliesfasterGDPgrowth whichmustalsoimplystrongeraggregate income growth,making redistribution possible--butmonetary policy could also be part of the initial response. So,while we find the disinflationary consensus to be compelling, there are forcesthat could at least limit the magnitude of these forces. What does AI say? What dbLumina says We begin by asking DB’s proprietary AI tool, dbLumina, how it assesses theprobabilities around inflation effects from AI overtwo different horizons--thenext 12 months and five years ahead.We use the following prompt: I’m interested in studying the different probabilities of outcomes for AI's impacton US inflation. I have in mind: the probabilities it raises inflation by 0.1pp or more,has minimal impact on inflation, slightly reduces inflation and meaningfullyreducesinflation. Please consider two different horizons: 1 year and 5 years.Define slightly reduces as-0.1ppto-0.4pp and meaningfully reduces as-0.5ppand more at both horizons. Please summarize the results in a table. Provide thereasoning for your findings as well as citations to academic papers in the table. Over the next year, dbLumina sees risks as tilted slightly towards higher inflationfrom AI. Thereareroughly equal probabilities assigned to the outcomes where AIhas a minimal impact on inflation(35%) and lifts inflation by any amount (40%).Incontrast,AI lowering inflationover the next yearhas a total probabilityof only25% (5% assigned to AI lowering inflation by0.5pp or more over the next 12monthsand 20% probability of a drag of-0.1pp to-0.4pp).dbLumina’srationaleis outlined withinthe table in the final column. As we noted in the prior section,dbLumina cites the positive impact on aggregate demand through AI-relatedinvestment as well as a potential boost to energy prices as rationalizing risks ofhigher inflation. Looking further out, the risks are better balanced. Five years ahead, dbLuminanow seesa roughly 50% probability that AI l