The other day, Artificial Intelligence (AI) pioneer and Nobel Laureate Geoffrey Hinton, in the Financial Times, said that AI will make a few people rich and the rest of us poorer. In doing so, he hinted at an Engels’ pause in our modern AI economy.
But what is an Engels’ pause? The term refers to a paradox in economic history: in 19th century Britain, industrial output surged, yet ordinary living standards barely budged. It was first called so by Oxford economist Robert Allen, in a seminal paper, after Friedrich Engels, the German philosopher. In early 1800s Britain, wages stagnated, food consumed most household budgets, and inequality widened even as factories hummed, and Britain became the “workshop of the world”. Only decades later did sustained improvements in welfare reach the majority, as Allen wrote in his paper.
Today, as AI reshapes the global economy we face a hauntingly similar question: Are we entering a modern Engels’ pause, where productivity surges but broad-based prosperity stalls? This becomes particularly pertinent after observing how a recent Stanford paper, titled “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”, documents younger workers being more vulnerable to AI-induced shifts in the economy. It also comes on the heels of an Indian software giant shedding 12,000 jobs and making an AI pivot. And all of this at a time when a recent MIT study pointed out that 95% of AI pilots are not generating visible gains in organisations due to frictions in complementary capabilities.
So, what precisely gives? For answers we need to turn to the economics of innovation. AI bears the hallmarks of a general-purpose technology (GPT) such as steam power, electricity, and the Internet, and it has the potential to transform multiple industries.
According to Agrawal, Gans, and Goldfarb (2018), AI dramatically lowers the cost of prediction. And yet, GPTs historically unleash not just growth but also dislocation. Complementary innovations, institutional adjustments and new tasks and skills must emerge before benefits are widely shared. Sussex University economic historian Nicholas Crafts discussed this in a 2021 article building on Allen’s work, and so did Bojan Jovanović and coauthors (2005) with United States data. The pauses are likely because capital or technological deepening could create gains for oligarch entrepreneurs but not for the rest of us — like Prof. Hinton stated and as a student of this writer pointed out recently.
Some of the markers
What might be the empirical signs of a modern Engels’ pause? To start with, we can examine whether there are productivity gains but stagnant wages. In call centres in the Philippines, generative AI copilots have boosted productivity, some argue, by 30%-50%, with firms enjoying cost savings and faster service. Yet, worker wages have barely moved, and in some cases, workloads have intensified. It is also at a time when inflation is high and cost of living prices make workers feel poorer. It is very concurrent with a recent New Yorker cartoon too which showed how an individual is asking ChatGPT why her electricity bills are rising. The dark humour cannot be missed on AI’s Engels’ pauses.
A second sign of a modern Engels’ pause is to observe if there are rising costs of complements. AI productivity requires complements: cloud computing, retraining, data access, and cybersecurity. These are expensive. For workers, the “price of staying relevant” is rising, coding boot camps, new certifications and continuous learning. Much like 19th-century households where higher wages were offset by rising food prices, and more recently in the Philippines, where today’s workers may see modest wage growth eroded by the high costs of digital survival.
A third marker would be to observe unequal distribution of gains ultimately deepening global inequality. PwC estimates that AI could add $15.7 trillion to global GDP by 2030. But the benefits will be concentrated in the U.S., China, and a handful of firms controlling foundational models. The IMF (2024) estimates that 40% of jobs worldwide are exposed to AI — half in advanced economies, where high-skilled substitution is likely. This bifurcation suggests a delay, or even denial, of welfare gains for large swaths of the global workforce. This writer’s paper in the Journal of Development Economics shows that a technology race in India with stronger intellectual property laws caused deep wage inequality. Going forward, could this be the story ahead in much of the world?
Finally, job displacement and task transformation will also be early measures to ascertain a modern Engels’ pause. Doctors are being complemented with ChatGPT increasingly. A group of researchers from Tsinghua University in China have started the world’s first AI-powered hospital. Meanwhile in education, finance, public management or infrastructure management, AI is slowly making inroads and transforming tasks and displacing jobs, as this writer’s recent consulting work with GMR Airports also showed or when we see an Albania launch the first AI Minister, Diella.
Models to follow, steps to take
Overall, the historical resonance on an Engels’ pause is sobering. In the Gilded Age of the U.S., productivity soared. But so did inequality, leading to labour unrest and political upheaval. Only with reforms, trade unions, public schooling and welfare states did living standards broadly rise. The lesson is clear: without AI governance, the Engels’ pause may persist.
This brings us to the final point on public policy. How can governments extricate economies out of the malaise of an AI Engels’ pause? To start with, skills transitions programmes are going to be key. Singapore offers a promising model. Its SkillsFuture programme provides continuous education credits for workers to reskill. The world’s first AI University in Abu Dhabi, the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) is another example of the role of skilling and new generation AI-related human capital creation.
There also must be thought around redistribution of AI rents, through robot taxes or through Universal Basic Incomes (UBI). Experiments with UBI in the United Kingdom and the European Union, or philanthropic commitments such as the Chan-Zuckerberg Initiative, aim to channel AI gains toward public good.
Finally, AI infrastructure should be treated on a priority basis as a public good. Compute and data are the “food” of the AI economy. If these remain scarce and expensive, productivity gains will not translate into welfare improvements. The launch of K2Think.ai and Apertus from the United Arab Emirates and Switzerland, respectively, in September as public (not private) open AI reasoning models here are nice steps in this direction.
What the challenge is
Some may still argue, with merit, that the Engels’ pause analogy is overstated. Unlike in the 19th century, today’s societies have stronger welfare systems, democratic institutions (though evidence on democratic backsliding is now quite robust worldwide), and rapid diffusion of technology. Smartphones reached billions within a decade; AI assistants could follow. Moreover, AI’s potential to lower costs in health care, education and clean energy could deliver immediate welfare benefits, if governance accelerates deployment equitably. In this sense, the AI Engels’ pause might be shorter than its historical counterpart if policy aligns with innovation.
But we still need to be cautious about macro gains and micro stagnation. Political economy teaches us that the Engels’ pause is not destiny. Its erasure is also about political will.
The challenge for AI governance students and for policymakers worldwide is thus to ensure that AI is not just a productivity revolution, but a human welfare revolution wherein we ponder on a new theory of change. History warns us that progress delayed is progress denied. So, while the Engels’ pause is the ghost at the feast of AI optimism, whether it lingers, how long it lingers and how swiftly it may pass, is up to us.
Chirantan Chatterjee is a Professor of Economics at U-Sussex and 2025 Founding Fellow of the Royal Economic Society, U.K. He is also a Visiting Professor at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, UAE (the world’s first Artificial Intelligence University), the Max Planck Institute (MIPLC) and Ahmedabad-U, India
