Pro-Worker Artificial Intelligence: NBER Research on Building AI That Empowers Workers

Why Pro-Worker Artificial Intelligence Matters Now

Pro-worker artificial intelligence represents a fundamental shift in how we think about AI development and deployment. In a landmark NBER working paper published in February 2026, MIT economists Daron Acemoglu, David Autor, and Simon Johnson—two of whom received the Nobel Prize in Economics in 2024—present a rigorous framework for understanding when AI helps workers versus when it harms them. Their central argument challenges the dominant narrative that artificial intelligence inevitably displaces human labor.

The paper arrives at a critical juncture. Generative AI tools are being rapidly deployed across industries, and the prevailing discourse focuses almost exclusively on automation and job displacement. Yet Acemoglu, Autor, and Johnson argue that AI’s capacity to serve as a collaborator—extending human judgment, enabling new tasks, and accelerating skill acquisition—is equally transformative and currently underexploited. This distinction between AI that replaces workers and AI that empowers them has profound consequences for wages, employment, inequality, and who ultimately benefits from technological progress.

Understanding pro-worker artificial intelligence is not merely an academic exercise. As organizations worldwide adopt AI systems, the design choices made today will shape labor markets for decades. The researchers demonstrate that these outcomes are not predetermined by the technology itself but by the incentives, policies, and institutional frameworks that guide its development. For a deeper exploration of how AI research shapes business strategy, see our analysis of AI enterprise transformation trends.

Five Categories of AI Technological Change

The NBER framework introduces a five-category taxonomy that advances beyond the simplistic automation-versus-augmentation dichotomy. Each category carries distinct implications for workers, wages, and employment patterns. Understanding these distinctions is essential for policymakers, business leaders, and workers navigating the pro-worker artificial intelligence landscape.

Labor-augmenting technology increases the effective productivity of human labor in existing tasks. Workers produce more output per hour, which can raise wages—but the effects on total employment remain ambiguous because higher productivity per worker may reduce the number of workers needed for a given output level. A factory line tool that helps an assembler work faster fits this category.

Capital-augmenting technology makes machines, equipment, and infrastructure more productive. This primarily benefits capital owners, with indirect and often uncertain effects on workers depending on how substitutable capital and labor are in specific contexts.

Automating technology enables machines or algorithms to perform tasks previously done by humans, directly substituting capital for labor. The authors emphasize that much of current AI development—from robotic process automation to large language models performing routine writing—falls into this category. Automating technology can reduce wages and employment, particularly for workers whose skills concentrate in automatable tasks.

Expertise-leveling technology commodifies or democratizes existing expert knowledge, enabling less-skilled workers or machines to perform tasks that previously required significant human expertise. While this appears beneficial on the surface, it erodes the wage premium associated with expertise and can reduce the economic returns to skill acquisition. According to researchers at the National Bureau of Economic Research, this category requires particularly careful evaluation because it can look pro-worker while actually undermining the economic value of human skills.

New task-creating technology generates entirely new tasks, roles, and categories of work that require human involvement—particularly novel forms of human expertise, judgment, and creativity. This is the only category the authors identify as unambiguously pro-worker, because it creates new demand for human skills rather than commodifying or replacing existing ones.

New Task-Creating Technology as the Pro-Worker Path

The concept of new task-creating technology is the intellectual cornerstone of the pro-worker artificial intelligence framework. Unlike other categories that may have mixed or negative effects on workers, new task-creating AI expands the frontier of what humans do rather than shrinking it. This distinction has enormous practical significance for how organizations should approach AI adoption.

Consider the difference between an AI system that automates customer service entirely—replacing human agents with chatbots—and one that creates an entirely new role: AI-human collaboration specialists who manage complex cases that neither humans nor machines can handle alone. The former eliminates jobs; the latter creates a new category of work that demands novel human expertise in understanding both AI capabilities and human needs.

Acemoglu, Autor, and Johnson build on their influential task-based framework, which reconceptualized how technology affects labor markets through task displacement and reinstatement. Their previous work, including the book Power and Progress (2023), argued that technological progress does not automatically benefit the majority and depends on institutional and political choices. The pro-worker AI framework operationalizes this insight by providing a clear taxonomy for evaluating specific AI applications. As the Massachusetts Institute of Technology continues to lead research in this area, their framework offers actionable guidance for technology governance.

The key insight is that most AI applications do not fall neatly into a single category. A given system may simultaneously automate some tasks, level expertise in others, and create new tasks. The critical question is which effect dominates—and whether deliberate choices can tilt the balance toward the pro-worker dimension. Our coverage of digital transformation and workforce impact explores how leading organizations are making these choices.

Real-World Pro-Worker AI Applications Across Industries

The NBER paper illustrates its framework through compelling hypothetical and real-world examples spanning multiple industries. These cases demonstrate that the same underlying AI technology can be deployed in fundamentally different ways depending on design choices, institutional context, and the incentives facing firms and developers.

In aviation maintenance, AI systems can either replace human inspection and diagnostic judgment—an automating approach—or augment the capabilities of maintenance technicians by providing enhanced diagnostic tools, real-time data overlays, and predictive analytics that extend their expertise to more complex and novel situations. The pro-worker path creates new task-creating roles where experienced technicians handle increasingly sophisticated challenges with AI assistance.

In education, the contrast is particularly vivid. Automating AI would replace teachers with fully automated tutoring and content delivery systems. Pro-worker AI instead empowers educators with real-time data on student learning patterns, enables personalized instruction at unprecedented scale, and frees teachers to focus on higher-order mentoring, critical thinking development, and socio-emotional support—tasks that require distinctly human capabilities. The U.S. Department of Education has increasingly recognized AI’s potential to augment rather than replace teaching.

In electrical services, the paper examines whether AI commodifies electrical expertise—enabling less-trained workers to perform tasks with AI guidance (expertise-leveling)—or expands what electricians can do, such as managing complex smart-grid systems, integrating renewable energy technologies, and performing novel diagnostic tasks that never existed before.

The gig economy example reveals how AI-driven algorithmic management can either intensify surveillance and control over workers—automating managerial functions in ways that reduce autonomy—or enhance worker decision-making through better route optimization, transparent earnings data, and tools that increase worker agency. In patent examination, AI tools can either automate prior-art searches (reducing examiner roles) or augment examiners’ ability to evaluate increasingly complex, interdisciplinary applications.

These examples span a wide range of skill levels, industries, and demographics, underscoring the paper’s core message: whether AI helps or harms workers is a design and policy choice, not a fixed property of the technology.

Market Failures Driving Underinvestment in Worker Augmentation

Perhaps the most consequential contribution of the pro-worker artificial intelligence framework is its identification of systematic market failures that lead to chronic underinvestment in worker-empowering AI. The researchers argue that without deliberate intervention, market forces will consistently favor automation over augmentation—not because automation is always superior, but because of structural incentive misalignments.

Misaligned firm incentives constitute the first major failure. Firms adopting AI face strong pressure to reduce labor costs, which is directly measurable and immediately visible on balance sheets. The benefits of pro-worker AI—higher service quality, improved worker retention, greater innovation capacity, and expanded service offerings—are harder to quantify and accrue over longer time horizons. Additionally, firms with monopsony power in labor markets have reduced incentive to invest in technologies that raise worker productivity and wages, since they can extract value from workers without improving conditions.

Misaligned developer incentives create a second bottleneck. AI developers and technology companies are rewarded for building general-purpose automation tools with large addressable markets. Venture capital and the broader technology ecosystem prioritize scalable solutions that can replace workers across many industries over context-specific augmentation tools tailored to the needs of workers in specific occupations. A tool that replaces 10,000 customer service agents has a clearer business model than one that makes 10,000 customer service agents significantly better at their jobs.

Path dependence compounds these problems. Once the AI development ecosystem is oriented toward automation—in research agendas, training data, benchmarks, talent pipelines, and business models—it becomes self-reinforcing. The accumulated infrastructure and expertise in automation creates technological lock-in, making it progressively more expensive to pivot toward pro-worker applications even when they would deliver greater social value.

The Pro-Automation Ideology Problem

Among the market failures identified in the NBER research, the concept of a pervasive pro-automation ideology is perhaps the most provocative and important. Acemoglu, Autor, and Johnson argue that a cultural and ideological bias within the technology industry and among business leaders equates technological progress with human replacement. This ideology shapes R&D priorities, investment decisions, and public discourse in ways that crowd out serious consideration of pro-worker alternatives.

This is not merely a neutral market outcome. The pro-automation ideology reflects specific cultural values, power dynamics, and narratives within the tech sector—what the authors in Power and Progress called the “vision of the tech elite.” The belief that full automation is inevitable, desirable, and represents the highest form of innovation has become a self-fulfilling prophecy: developers build automation tools because that is what the culture celebrates, and the resulting wave of automation tools reinforces the narrative that replacement is the natural trajectory of technology.

The implications are far-reaching. When AI conferences, academic publications, and industry benchmarks primarily reward systems that perform tasks without human involvement, the entire research agenda shifts away from studying how AI can make humans more capable. When venture capitalists and corporate boards evaluate AI investments primarily through the lens of labor cost savings, projects that would expand worker capabilities receive less funding. This ideology also shapes public expectations, leading workers and policymakers to accept displacement as inevitable rather than mobilizing to demand pro-worker alternatives.

Challenging this ideology requires more than technical innovation—it demands a cultural shift in how we define technological progress. As our guide to AI collaboration in the future of work examines, forward-thinking organizations are already adopting frameworks that measure AI success by how much it enhances human contribution rather than how much human labor it eliminates.

Nine Policy Directions for Pro-Worker Artificial Intelligence

The NBER paper moves beyond diagnosis to propose nine concrete policy directions that would change incentives and redirect AI development toward pro-worker outcomes. These recommendations represent a structural approach that goes far beyond the usual calls for worker retraining and safety nets.

Targeted investment in healthcare AI would direct public funding and procurement toward applications that augment healthcare workers—nurses, primary care physicians, community health workers—rather than replacing them. AI diagnostic tools that extend a nurse practitioner’s capabilities to underserved communities create new task-creating value while expanding access to care.

Targeted investment in education AI follows a similar logic, prioritizing tools designed to empower teachers with real-time learning analytics and personalized instruction capabilities over systems that automate teaching itself. Public investment signals can shape the direction of private R&D by demonstrating viable markets for augmentation tools.

Tax code reform addresses a fundamental structural bias. Current tax systems in many countries, including the United States, effectively subsidize capital investment—including automation technology—through accelerated depreciation and investment tax credits while taxing labor heavily through payroll taxes. This creates an artificial incentive to replace workers with machines even when automation is not inherently more efficient. Leveling this playing field could significantly redirect AI investment.

Additional policy directions include R&D subsidies specifically targeting pro-worker AI research, workforce development programs redesigned around AI tools that accelerate skill acquisition, procurement standards requiring government agencies to prioritize pro-worker AI, and data governance frameworks that give workers ownership of the data generated by their labor. The White House Office of Science and Technology Policy has begun exploring several of these directions in its AI governance framework.

Tax Reform and Antitrust as Pro-Worker AI Levers

Two of the most powerful policy instruments in the pro-worker artificial intelligence toolkit are tax reform and antitrust enforcement—areas where government action can fundamentally alter the incentive landscape for AI development without requiring direct regulation of specific technologies.

Antitrust enforcement targets the concentration of power among large technology companies that dominate AI development and deployment. When a handful of corporations control the AI platforms, models, and data infrastructure, they effectively determine the direction of technological change for entire economies. Greater competition would diversify the types of AI applications developed—including more pro-worker tools—and reduce the market power that enables dominant firms to impose automation-centric solutions across industries.

The tax reform argument is particularly compelling because it addresses an often-invisible structural bias. The effective marginal tax rate on capital income (including automation investments) is substantially lower than the effective rate on labor income in most advanced economies. This means that firms face a fiscal incentive to replace workers with machines that has nothing to do with the relative productivity or social value of workers versus machines. Acemoglu and his co-authors have calculated that eliminating this bias alone could meaningfully slow the pace of purely cost-driven automation.

Intellectual property protections for worker expertise represent a third critical lever. Current frameworks allow AI systems to be trained on vast quantities of human expertise—accumulated through years of education, apprenticeship, and practice—and then used to replace the very workers whose knowledge made the AI possible. Creating legal protections that give workers rights over their expertise data would both slow harmful automation and generate revenue streams that fund transition support. The International Labour Organization has called for exactly this kind of framework in its AI governance recommendations.

Together, these three instruments—tax reform, antitrust enforcement, and IP protections—address the structural underpinnings of the pro-automation bias rather than merely treating its symptoms. They change the incentive calculus for firms and developers at the most fundamental level, making pro-worker AI development not just socially desirable but economically rational.

Building a Pro-Worker AI Future

The vision articulated in this NBER research is not anti-technology—it is pro-human. Acemoglu, Autor, and Johnson are not arguing against AI development but for a redirection of AI development toward applications that expand human capabilities, create new categories of meaningful work, and distribute the gains from technological progress more broadly. This vision requires action from multiple stakeholders.

Policymakers must recognize that the direction of technological change is not predetermined. Tax reform, antitrust enforcement, targeted public investment, and IP protections for worker expertise can fundamentally alter the incentive landscape. The nine policy directions proposed in the paper provide a comprehensive roadmap that goes beyond Band-Aid solutions to address structural causes of pro-automation bias.

Business leaders must challenge the assumption that AI adoption means worker replacement. The most innovative applications of AI create new forms of value by combining machine capabilities with human judgment, creativity, and social intelligence in ways that neither could achieve alone. Companies that invest in pro-worker AI may find that they build more resilient organizations with higher-quality outputs, stronger employee retention, and greater capacity for innovation.

Workers and their representatives must engage in shaping how AI is deployed in their workplaces and industries. Collective voice—through unions, professional associations, and democratic participation—is essential for ensuring that technological change serves broad social interests rather than narrow profit maximization.

The stakes could not be higher. As AI capabilities continue to expand at an extraordinary pace, the choices made in the next few years about the direction of AI development will shape labor markets, wage distributions, and economic opportunity for generations. The NBER framework provides the analytical tools and policy roadmap needed to build an AI future that works for workers—not just for the firms and developers that deploy AI systems. Explore more research-driven insights in our economic policy and AI workforce analysis.