In a provocative thought exercise posted on the Citrini Research Substack, an intriguing question is raised: what happens if AI progress stays bullish at the cost of macroeconomic trends turning bearish? It is purported that the bearish macroeconomic direction is driven primarily by an historic “intelligence displacement spiral”, in which abundant machine intelligence compresses the value of white-collar labour, upending many of the economic norms we currently take for granted.

Framed as a retrospective “macro memo” written from June 2028, the piece imagines an economy that can no longer rely on the traditional knowledge economy – i.e., production, value, and growth driven by human intellectual capital, information, and innovation. The centre of the spiral is layoffs, first rationalised through cost-cutting (see the most recent example of Block), which reinforces further reinvestment into AI, creating a negative feedback loop with “no natural brake”.

Since the knowledge economy represents a significant portion of the economy as a whole, as white-collar workers face structural displacement and, in some places, even existential obliteration, overall earnings – and hence spending – become impaired. As a result, household incomes and consumer demand are hollowed out, even as measured productivity and corporate margins surge. What’s worse, since white-collar income sustained the approximately $13 trillion (USD) mortgage market, the crisis in stable household cashflow that the mortgage market relied upon forces underwriters to reassess the entire notion of prime mortgages.

Think of it this way: in the context of factor automation, a robot generally changes how goods are produced. In a world where LLMs become the default, there is a possible future where an AI agent changes not only how white-collar work is done, but also who earns, which middlemen survive, and, ultimately, who buys. All the money spent on discretionary goods by white-collar workers won’t be replaced by machines, as the Citrini thought exercise describes; nor will they sustain the loop of demand on which modern economies depend. The shock is therefore no longer “just jobs”. Hence, the bearish macroeconomic trend worsens as the human intelligence displacement spiral widens. With decreased spending and widening cracks in credit, what begins as disruption in an isolated sector – starting with software – metastasises into systemic risk, and then into a cascading polycrisis. The divergent loop intensifies as further improvements in AI mean more firms require fewer people. Layoffs therefore spread throughout more and more sectors, while mortgages, private credit, and, ultimately, the engine of entire markets ceases to function. This is the idea.

The moral? AI can “work” exactly as promised and still threaten systemic collapse of the market economy.

What is so peculiar about this imagined world is how it is quite different from historical examples. Somewhat paradoxically, productivity explodes while wages collapse and unemployment rates skyrocket. The atrophy of the real human consumer economy creates a gap that shows up as “Ghost GDP”: output accrues in national accounts but doesn’t circulate through households.

***

At present, I am not especially sceptical about AI in the sense of the technology itself. My concern, rather, is with the political economy of adoption: who captures the gains, who bears the costs, and how institutions respond. What I take most from the Citrini thought exericse is precisely this same critique. Much as with the broader historical technological evolution of human society, from the slingshot to the atom bomb, AI can represent a positive potential as an epoch of technological transformation, or, its disruption and effects, could be realised in completely catestrophic ways.

All technology is socially mediated to greater or lesser degree, and there is a difference between conceptual form – the science of technology as an expression of the laws of nature – and social actuation, with the latter including a combination of factors. The internet offers a precedent. Here was a once enabling infrastructure whose social realisation was steered as much by incentives and power as by the technology itself. AI is no different. There are big questions about how machine intelligence is integrated into society and the economy – how it is realised socially and economiclly. In my opinion, the lessons of history, if nothing else, should intensify the urgency of the general public to demand unrelenting oversight and transparent discourse on the future use of AI.

***

On the topic of history, what is fascinating about the Citrini thought exercise – particularly its imagining of AI disruption and its consequences – is how closely it rhymes with the automation shocks to factory labour in the 1960s and 1970s. Except, as I see it, there is one significant difference to the automation of labour, and it has to do with the mechanics of the economy. Industrial automation substituted for muscle inside bounded physical systems, and, in a literal sense, was limited by the physics of human economy. AI substitutes for intelligence – an historically scarce input positioned upstream of wages – constrained only by energy input and the maximal efficiency of data centres. In principle, we already have the technological means to ensure such constraints basically become redundant.

This is where an historical analogy snaps into focus. If factory automation primarily replaced “just jobs” on the shop floor, the 1980s economy could isolate and, in some sense, quarentine itself against the shock and disruption. Even if one were to argue about the value of human relationships, it’s clear that economically this was viewed merely as friction. Indeed, as the Citrini article notes, the established mantra has long been technological innovation destroys jobs and then creates even more. The argument has, until now, been convincing because it has been right for two centuries: even if we could not yet imagine the future jobs, the assumption was that they would surely arrive. This made sense when the majority of new jobs used to require humans to perform them. Unlike with the automation of factories, in the Citrini scenario, once intelligence becomes abundant and the intelligence premium collapses, the shock ripples throughout the financial and institutional scaffolding built on the assumption that human cognition is irreplaceable at scale.

If history is anything to go by, governments cannot be relied upon to plan for such crises. As evidence, see the example of the many towns and communities previously of the American and British Heartlands, since effaced and transformed into rustbelts, respectively. Although a very different kettle of fish, an example that immmediately comes to mind is close to home: British miners from the 70s and 80s and sudden deindustrialisation. Whatever one’s view of the Thatcher governments’ shift away from a nationalised industrial base, few accounts make the retraining response feel commensurate with the scale of the upheaval. Training initiatives existed, sure, but they were vastly inadequate for the scale of the crisis. As a result, many communities faced high unemployment and long-term decline as alternative industries failed to arrive at anything like the necessary pace. To compound matters, the prevailing neoliberal economic policy emphasised a lack of state investment in regional development to create alternative industries or sectors of growth that maximise the skills of previous labour.

It’s difficult to imagine how an analogous crisis unfolds beyond June 2028 in the Citrini article. Although very different to the cases I’ve compared, perhaps a hint is in their one crucial similarity. In the 1980s, during rapid deindustrialisation, the UK’s problem was closer to aggregate growth with locally missing prosperity. National GDP eventually recovered and shifted toward services, while many industrial regions – with “old” labour skills – experienced long-lived scarring. In other words, there was real growth, just unevenly distributed. The production system still relied on human wages and household spending. Even as manufacturing and mining employment collapsed, output and demand were still tied to other sector jobs and incomes – whether through new service jobs, public-sector employment, self-employment, or transfers, to varying degrees. It is difficult to imagine how the same proves true in the extended Citrini outlook, especially as non-human “labour” expands. Somewhat ironically, perhaps wellspring for future jobs is the traditional trades, establishing a new hard labour class.

In whatever way the future plays out, what I think is clear is that, just as in the many examples we might pull from history, there will be prosperity in the aggregates and decay in the periphery. It is just that, in 2029, the decay in the periphery may be unlike anything seen before.

*Cover image: Rose Willis & Kathryn Conrad / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/