Stay informed with free updates
Simply sign up to the Global Economy myFT Digest — delivered directly to your inbox.
Most of the sell-side has remained hilariously silent at a mere Substacker seemingly shaking markets, even though the anguish and frustration is almost palpable.
Alphaville LLC was a notable exception, having issued an incisive analysis around the evolving narrative structure of market narratives last night. But Evercore ISI’s Krishna Guha has also offered his thoughts, and, given the heightened interest in the topic, we thought we’d share them here.
The summary is that he’s unimpressed with Citrini Research’s report — which definitely “isn’t bear porn or AI doomer fan-fiction”, according to Citrini itself. Guha, a former FT journalist, is unconvinced by this insistence, and writes that:
. . . even if the tech and microeconomics were to evolve in line with this scenario, it is highly unlikely that the macro would, as this would require a set of extreme and improbable conditions to hold.
Fortunately, Alphaville is fluent in sell-side, and this can be roughly translated as GTFO.
Guha does stress that he thinks Citrini’s research is interesting as a “thought-provoking example of scenario analysis”, which can be valuable if it spurs people into thinking through a variety of potential outcomes from AI and the disruption that a lot of technologists predict.
He also concedes that it’s impossible to completely discount the possibility of a science fiction-like world where AGI-powered robots make almost all human work obsolete. “The economics of such a world . . . would be unusual to say the least,” Guha notes.
But with his economist hat on, Guha can see a LOT of problems with the more near-term arguments made by Citrini. Given the current interest in the report, Alphaville thought we’d quote liberally from the Evercore ISI note and give you Guha’s main rejoinders in his own words:
The scenario analysis implies vast wealth creation accrues to the owners of AI and compute, but does not have this AI wealth generate new consumption (or indeed additional investment demand). There are some theories that propose the idea of satiated consumption — that at some point people stop wanting to consume any more — but we find these implausible. The AI wealthy may have a low marginal propensity to consume, but they will consume, and this will generate jobs, even if in occupations that are not common today. In the 19th century, for instance, each rich household employed dozens of personal servants. Even if there are limits to the consumption of current goods and services, new ones will be invented. The purest case is products or activities that extend a person’s healthy life: there is no limit to the amount of healthy life a wealthy person wishes to consume.
At the other end of the K-shaped economy blue collar workers would be less affected and would continue to consume, at least as long as robots are unable to fully replicate manual activities. Although some of these workers would suffer from the drop in the demand from white collar workers, and/or from competition from newly displaced white collar, others with discrete skills would in effect be complements to AI, and their activities would be in greater demand as collapsing cost of AI-led tasks increased the desired volume of consumption of these products. Because their productivity would not rise anything like as rapidly as AI-led productivity in cognitive tasks, these blue collar workers would see large relative wage gains. And they certainly would consume.
The scenario confuses the firm-level and macro implications of sharper agent-driven price discovery that destroys oligopoly rents and associated wealth in incumbent business models — but in any standard model increases efficiency and activity in the economy as a whole, all else equal. Consider two examples given in the paper. If Visa and Mastercard lose pricing power and have to cut fees, their shareholders lose, but merchants and consumers benefit, and economic activity increases rather than decreases. If realtors are no longer able to charge 6 per cent on home transactions, they lose, but homeowners benefit, and housing activity increases rather than decreases. There would be negative wealth effects for equity owners in the old oligopoly but positive wealth effects for those no longer paying monopoly rents.
The scenario lacks Schumpeterian creative destruction in which resources released from old businesses that fail to make the transition to an AI-led world are used to form new businesses that generate new jobs. Indeed (as others have noted) it is a high tech version of Marx’s thesis that capitalism would ultimately destroy itself by immiserating the petit bourgeois and working class until it had no consumers left, no additional profits to be earned on existing products produced and no reason to grow. Schumpeter’s basic insight was innovation of new products and services would drive activity and employment. (It is worth recalling that the Teamsters Union takes its name from its members original jobs as drivers of teams of horses or oxen.) This does not of course require that AI will at all times create new jobs at the same pace that old jobs are destroyed — there is a horse race here, and AI may well destroy more jobs than it creates to begin with.
Finally, Guha argues that even if Citrini’s scenario of swift and massive unemployment, wealth destruction and financial stress were to materialise, the report completely ignores the fact that governments and central banks would respond.
Even a Kevin Warsh-led Fed would restart quantitative easing, Evercore’s economist points out:
The scenario has no monetary or fiscal policy response, while in reality if AI does drive unemployment to anything like the 10 per cent level in the scenario — in particular if politically influential white collar workers are losing their jobs in droves — we would expect a vigorous policy response on both tracks. On the fiscal side, it is true that fiscal revenues could fall even if the economy grows rapidly if the share of income going to relatively highly taxed labor falls sharply while the share going to relatively lightly taxed capital rises. But (as the author acknowledges) this could and we think would be rapidly fixed by overwhelming political pressure to shift taxation from labor to compute and/or wealth. This would provide the resources needed to back aggressive fiscal support.
On the monetary policy side, the plausible tail risk version of this scenario is as noted earlier a very rapid shock to old business wealth and jobs that creates an old-school liquidity trap in which the natural rate of interest falls below zero. That could also be accompanied, as the scenario suggests, by a housing and banking crisis — but this is a generic observation: any big shock to labor income would generate similar stress dynamics. The Fed and other central banks have dealt with the zero bound trap and financial stress twice in recent decades and would respond with the familiar set of market stabilization and zero bound tools including QE and forward guidance, combined with fiscal support to recapitalize banks and back capital market interventions if needed.
This would happen under Warsh regardless of his prior preferences; the only concern is that market doubts about this could prevent the bond market from anticipating this fully ex ante in ways that are stabilizing. If instead we take a set of standard economic assumptions: that the rich continue to find new ways to consume, and are joined by those blue collar workers who gain in relative and absolute terms, that erosion of oligopoly rents stimulates rather than decreases economic activity, that Schumpeterian creative destruction generates new business models and jobs even if not at first at the pace that old jobs are destroyed, and that fiscal and monetary policy move aggressively against any incipient Keynesian liquidity trap caused by concentrated destruction of old wealth and old jobs, then macro evolves in a very different way — with continued and rapid economic growth, even through a potentially very difficult transition period for many firms and workers. That seems like a sound base case.
It feels like we’re all taking a good old-fashioned and fun bit of doomerism a bit too seriously right now. It’s almost as if people have forgotten that ZeroHedge still exists, even if it mostly just sells gold coins and vitamin supplements these days.
As Rob Armstrong noted earlier today, we should probably worry more about how the market is so jittery that a Substack can trigger a violent rout, than debunking the report itself.