Acemoglu and Johnson on the Past and Future of Work

A sort of review of Power and Progress

The 2024 Nobel Prize in economics went to Daron Acemoglu and Simon Johnson. Their joint work on automation is an extension of long debates that first arose during the early Industrial Revolution. Their book, Power and Progress, alongside their other writings, confronts a deceptively simple question: why does some automation create jobs while others eliminate them? This question has ricocheted through centuries, from David Ricardo’s grappling with machinery’s dual-edged sword to today’s speculations on artificial intelligence (AI).

The Genesis of Their Thought

The book explores three foundational questions: What determines when new machines and production techniques increase wages? What would it take to redirect technology toward building a better future? Why is current thinking among tech entrepreneurs and visionaries pushing in a different, more worrying direction?

On that third point, they emphasize power. In the first chapter they outine H.G. Wells in the Time Machine, which they quote at the beginning of the first chapter. Wells wrote,

‘Well, you thought technology is about controlling nature, but it’s as much about humans controlling humans.’

This is what the authors consider as the double role of technology

With respect to wages and work, Acemoglu and Johnson draw on a long lineage of economists who sought to reconcile the promise of technological progress with its potential to devastate labor markets. Their intellectual framework owes much to David Ricardo’s transformation treaties during the early 1800’s. Ricardo initially viewed machines as unalloyed contributors to prosperity. Yet, as power looms displaced skilled handloom weavers, Ricardo revised his view, conceding that automation could diminish demand for labor. The cotton mills of industrial England exemplified this dual reality, raising overall productivity while plunging thousands of workers into destitution. Similarly, E.P. Thompson chronicled how this transition destroyed autonomy, subordinating labor to the drudgery of factories.

Technological improvements have often not led to shared prosperity, but instead benefited a narrow elite.

Acemoglu and Johnson extend these historical debates into the age of AI, blending Ricardo’s economic perspective with Thompson’s sociopolitical lens. However, their framework invites critical reflection on its adaptability to unforeseen developments. For instance, how does it account for the potential of AI to generate entirely new industries and job categories that are difficult to predict? These possibilities could challenge assumptions about displacement and reinstatement effects, raising questions about the scope and scale of technological transformation in the AI era. They argue that automation’s effects hinge on whether it displaces existing tasks or creates entirely new ones. Displacement, as with handloom weavers, tends to concentrate benefits among capital owners, exacerbating inequality.

For instance, the advent of automated customer service systems will displace call center employees, while basic data entry automation will replaced administrative workers. These modern examples highlight how automation can reduce demand for labor in certain sectors without proportionately, or immediately, creating new opportunities elsewhere. In sectors like healthcare, AI has the potential to create roles such as data analysts specializing in medical records, AI model trainers for diagnostic systems, and technicians for managing AI-enabled medical devices. Similarly, in education, AI-driven personalized learning systems may lead to new roles in AI content development, system customization for different curricula, and training educators to effectively integrate these technologies into teaching. In contrast, creating new tasks, a phenomenon they term “reinstatement effects”, can integrate labor into emerging economic structures which has a lag. They show that the introduction of computers led to roles like IT support specialists, software developers, games developers and cybersecurity analysts, which did not exist before but now form critical parts of the economy.

New tasks introduced by technological advances have been a major driver of employment growth and productivity growth, as they help launch new products and lead to more efficient reorganization of the production process.

Automation: Threat or Opportunity?

Their critique zeroes in on what they call "so-so automation", technologies like self-checkout machines, automated customer service systems, and basic data entry automation that replace workers without significantly boosting productivity or creating complementary roles. These innovations, they argue, merely shift costs to consumers (as unpaid labor) while diminishing workers’ bargaining power. AI, they warn, risks deepening this pattern unless harnessed to empower rather than displace labor. Showing historical parallels, they note that technologies creating “good jobs,” those combining meaningful tasks with decent wages, require deliberate intervention. However, their emphasis on 'machine usefulness' raises important questions. Could this focus inadvertently stifle innovation in areas where automation might fully replace human tasks but lead to substantial efficiency gains? Furthermore, how might this approach balance the pursuit of disruptive advancements with the goal of equitable job creation? They acknowledge the work of Erik Brynjolfsson who states that the challenge is not to slow down technology but to race with the machine to create shared prosperity. The authors show that fears about technological unemployment did not come to pass, so far, because automation was accompanied by improvements and reorganizations that produced new activities and tasks for workers. However, the authors view the current technological narrative as increasingly elitist and blindly optimistic, similar to the attitudes prevalent 250 years ago.

Policy Agenda

Acemoglu and Johnson champion a multidimensional reform agenda to steer AI development. For example, they propose tax incentives for companies investing in labor-complementary innovations, such as credits for firms that integrate AI to enhance rather than replace human roles. Practically, this could involve offering tax breaks to companies that demonstrate measurable job creation linked to new technologies or investing in upskilling initiatives. Additionally, enforcement mechanisms, such as periodic audits and public reporting requirements, could ensure these incentives lead to tangible benefits for workers and align with broader economic goals. Furthermore, democratizing technological decision-making could involve mechanisms like worker representation in AI deployment decisions or public deliberation forums to shape technology policy. They critique the prevailing shareholder-value orthodoxy, an heir to the industrialists of yore, that prioritizes cost-cutting over shared prosperity. This critique aligns with their insistence on democratizing technological decision-making, a theme reminiscent of Thompson’s calls for empowering labor.

Automation, Marginal Productivity, and Task Creation

The authors look into the intricate ways automation impacts employment, focusing on why some forms of automation lead to job creation while others do not. Automation can increase a company's average productivity, calculated by dividing total output by the total number of employees. However, hiring decisions are based on marginal productivity, the added value generated by one more employee, not average productivity.

New technologies, such as industrial robots, automate tasks previously performed by humans, raising average productivity while keeping marginal productivity stagnant or even reducing it. Automation that significantly increases productivity can, however, lead to job growth. Lower production costs may allow automating firms to hire more workers for non-automated tasks, while increased demand for their products can drive employment in other industries.

For example, the growth of car manufacturing illustrates this dynamic. Automating car production increased demand for non-automated technical and clerical roles. Additionally, as my research showed seven auto manufacturers increased their number of employees by more than 132,000 people between 2009 and the end of 2014 while also adding tens of thousands of new robots to their factory floors.

Moreover, productivity gains in car factories fueled the expansion of related industries such as oil, steel, and chemicals while transforming transportation. This revolution in mobility also spurred growth in retail, entertainment, and service sectors.

Conversely, "so-so automation" often fails to create such growth. Self-checkout kiosks in grocery stores, for example, shift work from employees to customers, achieving minimal productivity gains that do not offset the reduction in cashier jobs.

The authors stress that creating new tasks is essential for boosting marginal productivity and employment. Henry Ford’s reorganization of the car industry in the 1910s exemplifies this principle. Alongside introducing mass production and assembly lines, Ford created new roles in design, technical operations, and clerical tasks, which increased labor demand. Similar patterns emerged with the advent of electricity, telephony, and computing, where technological advancements spawned entirely new industries and professions. This is aligned with the work of Nobel Prize Professor Robert Solow, who indicated that technological development will be the engine for economic growth in the long run. In the Solow-Swan model, if continuous technological progress can be assumed, growth in real incomes will be exclusively determined by technological progress.

Despite these historical successes, the authors highlight a troubling trend, since 1980, automation has accelerated while the creation of new tasks has slowed. This imbalance, they argue, has undermined workers’ economic positions, reflecting a technology portfolio overly focused on displacement rather than complementarity. To rectify this, they advocate for policies fostering innovation in technologies that complement and enhance human abilities, ensuring a balanced and inclusive approach to technological progress. Such as that in Singapore where the government subsidizes re-skilling of adults over the age of 40 years old.

AI’s Diverging Futures

The implications of their work should be considered carefully with the advent of AI-driven transformations. As seen during the Industrial Revolution, the trajectory of technology is not predetermined. However, key differences between then and now must be acknowledged. Today’s technological changes occur at an accelerated pace, transforming industries within years rather than decades. Unlike the Industrial Revolution, this rapid pace demands policies that can adapt swiftly to evolving technological landscapes. Moreover, the shift toward knowledge-based and service-oriented work fundamentally differs from the manual labor focus of the past, necessitating tailored approaches to education, workforce training, and equitable access to new opportunities. These differences underline the need for proactive and flexible policy frameworks to address the unique challenges of modern technological transformations.

Acemoglu and Johnson underline that AI’s potential for harm or benefit lies in human hands. However, shaping AI’s trajectory involves significant challenges, particularly in fostering international cooperation. AI development is a global endeavor, with differing national priorities and regulatory frameworks. This raises questions about how to align efforts across borders to ensure equitable and ethical outcomes. Without a unified approach, disparities in governance could exacerbate inequality and limit the effectiveness of policy interventions. Without intervention, they predict rising inequality, job precarity, and erosion of work’s intrinsic value. Yet, with thoughtful governance, AI could catalyze a reinvigorated social contract, enabling more equitable prosperity.

Why Some Automation Creates Jobs and Others Do Not

Central to their argument is the distinction between automation that substitutes for human labor and automation that complements it. Historical evidence supports their thesis, the spinning jenny initially displaced spinners but also expanded downstream weaving jobs. In contrast, power looms, introduced under conditions of weakened labor rights, precipitated widespread impoverishment. By contrast, new software tools that aid the tasks of car mechanics, enable greater precision work and increase worker marginal productivity, and wages. Today, AI’s potential to augment human capabilities hinges on similar factors, power dynamics, institutional frameworks, and the distribution of technological benefits.

Acemoglu and Johnson emphasize the need to shift away from "automation for its own sake." Instead, they advocate for “machine usefulness,” where AI supplements human expertise, as in medical diagnostics, rather than displacing it wholesale.

Their insights also extend to emergent industries, such as renewable energy and AI ethics, where policies could incentivize job creation through deliberate task design.

Lessons for Contemporary Policy

The broader lesson is that technology’s consequences are not dictated by its intrinsic properties but by the sociopolitical contexts shaping its deployment. This aligns with the principles of Utilitarianism, which emphasizes maximizing the combined welfare of all people in society. Ensuring technological progress contributes to collective well-being requires deliberate policy choices and institutional frameworks. During the Industrial Revolution, laissez-faire policies enabled capital’s dominance, exacerbating inequality. Today, a similarly hands-off approach to AI risks repeating those mistakes. Acemoglu and Johnson’s work thus serves as both a historical lens and a policy roadmap, urging societies to actively shape AI’s trajectory.

The authors suggest that shared prosperity does not automatically result from technological progress, it requires conscious efforts to ensure that advancements benefit a wider population. They say we need to reorient technology to a “socially beneficial trajectory.”

The relevance of Acemoglu and Johnson’s ideas lies in their dual emphasis on technological malleability and institutional reform. They show that progress is never automatic and must involve efforts from society to ensure equitable benefits. By bridging historical analysis with contemporary challenges, they suggest policy pathways to ensure AI enriches rather than impoverishes. As with Ricardo’s belated pivot, the key lies in recognizing that automation’s future is not inevitable, it is ours to determine.

Stay curious, get off the hamster-wheel and build for the future

Colin