Citrini: The Market’s Permission to Plunge

Gain entry to the Editor’s Overview at no charge

Roula Khalaf, the FT’s Editor, curates her preferred articles for this periodical publication.

AI-catastrophe narratives triumphant

In line with the recent trend, yesterday saw a decline in the stock values of numerous firms potentially susceptible to AI-driven disruption — including software, wealth management, and, entering the fray for the first time, financial institutions. The predominant rationale for this fresh apprehension stemmed from a weblog entry by Citrini Research, detailing how AI could lead to the dismissal of many high-income professionals and destabilize the economy. As reported by the FT: 

Recently, investors have latched onto social media speculation and minor advancements by nascent AI enterprises to warrant additional divestment. A widely disseminated weblog entry from Citrini Research published over the weekend, outlining how AI could theoretically elevate the US jobless rate beyond 10 percent by 2028, served as the most recent impetus.

The paramount aspect of the article is not its content. Rather, it signifies that the equity market has arrived at a juncture where weblog entries trigger substantial share fluctuations, or at minimum, where observers believe they do. Frequently, markets experience downturns due to reasons that are vague or nearly incomprehensible, compelling individuals such as myself to seek clarification. This could represent such an instance. Regardless, the Citrini commotion provides additional proof that we are operating within an overvalued market, which appears poised for a decline, motivated by factors likely extending beyond solely AI.

However, which part of the article captured the market’s focus? It adopts a viewpoint set several years ahead, from which it contemplates an economic downturn and monetary upheaval that have already transpired. It commences by stating: “What is presented hereafter is a hypothetical situation, not a forecast . . . The singular aim of this composition is to construct a model for a situation that has been comparatively neglected.” Yet, the inverse is accurate. The article gained traction because it embodies an intensified iteration of one of history’s most frequently examined scenarios: automated machines rebelling against their inventors. This concept has been exhaustively discussed since at least when Karel Čapek coined the term ‘robot’ and, simultaneously, foresaw that robots could engender significant difficulties. That occurred over a century prior. Even more than a hundred years before that, certain textile workers in Nottingham, operating under the guise of the fictitious General Ned Ludd, delved into this subject by demolishing various fabric factories. 

Moving on to the article’s premise. The core situation posits that AI eradicates employment and earnings, yet fails to substitute them, thereby pulling the economy into a slump and financial markets into turmoil:

The proprietors of computational power witnessed their assets skyrocket as personnel expenses disappeared. Concurrently, genuine wage increases stagnated . . . professional workers forfeited positions to automated systems and were compelled into less lucrative occupations . . . As fissures emerged within the consumer sector, financial commentators widely adopted the term “Phantom GDP”: production registered in national ledgers but never permeating the tangible economy . . . The rate of monetary circulation ceased. The consumption-driven economy, then comprising 70 percent of Gross Domestic Product, deteriorated . . . AI functionalities advanced, enterprises required fewer employees, professional redundancies rose, displaced individuals curtailed spending, profit margin pressures compelled businesses to allocate more into AI, AI functionalities advanced . . . This constituted a detrimental self-reinforcing cycle without any inherent restraint . . . 

The paradox herein was that the AI infrastructural ecosystem maintained its operations even as the economic system it was unsettling commenced its decline. NVDA continued to report unprecedented earnings. TSM sustained an operational capacity exceeding 95 percent. The large-scale cloud providers persistently allocated $150-200 billion each quarter towards data center capital expenditures. Nations intrinsically aligned with this trajectory, such as Taiwan and Korea, exhibited superior performance to a significant extent. 

A particular aspect of this portrayal seemed peculiar to me, though (given my lack of formal economic schooling) I found it difficult to verbalize. It resembles fiscal analysis that solely examines one facet of a transaction (“equities decreased because vendors surpassed purchasers,” and so forth). Citrini depicts an environment of exponentially rising efficiency coupled with a severe reduction in spending. Is this coherent?

Joseph Steinberg, an economics professor at the University of Toronto, aided me in developing my hunch that something was incorrect. He stated, “The initial segment of the argument I would expect my economics pupils to highlight pertains to ‘phantom GDP’.” What signifies when production “appears in governmental financial records but never permeates the actual economy”? Should Gross Domestic Product be escalating — with numerous automated devices producing goods ever more rapidly — then a corresponding element on the opposing side of the national income equation must also be increasing. The potential factors include expenditure, capital allocation, public sector outlays, or foreign trade balances. Within Citrini’s hypothetical, consumption is rapidly diminishing. Is governmental expenditure then escalating (supported by taxation or loans from whom, precisely)? Or exports — directed towards other nations experiencing an identical predicament?

None of these explanations appear logical, thereby leaving capital expenditure as the sole remaining option. However, Steinberg highlights that capital allocation is only viable when predicated on anticipated future consumption; “it is only rational to invest in AI if there exists sufficient purchasing power for the items that AI is creating”.

From my perspective, a component related to distribution must be present in the argument, which Citrini fails to elaborate completely. Earnings and expenditure escalate, but this transpires in areas that do not serve the typical American professional. Possibly, it is entirely concentrated within the “nations that were exclusively aligned with this trajectory, such as Taiwan and Korea.” Alternatively, it might all accrue to a handful of exceedingly irksome magnates in Silicon Valley. Under this situation, everyone ultimately serves exorbitant coffees to Peter Thiel. Or perhaps Thiel’s riches are confiscated by the state and reallocated. Nevertheless, it is challenging to fathom how individuals like Thiel would retain their affluence for an extended period in the world Citrini conceptualizes, given that their AI ventures would cease to yield returns once spending plummets.

This is not to imply that I hold the conviction that the AI transformation, akin to previous shifts in technological paradigms, will ultimately generate additional employment and enrich everyone. I do not. My assertion is that Citrini’s portrayal of the hazards appears illogical (perhaps those with a superior understanding of macroeconomics can interpret it more clearly?). The article achieved widespread attention not due to its clarification of the current circumstances, but rather because it resonates with age-old anxieties. 

A noteworthy publication

“While a fire marshal might possess the prerogative to cordon off thoroughfares during a crisis, this does not suggest that the fire marshal holds the power to levy road fees.”