What work has shaped
the AI economy?

find evidence of racial bias in one widely used algorithm, such that Black patients assigned the same level of risk by the algorithm are sicker than White patients (see the Perspective by Benjamin). The authors estimated that this racial bias reduces the number of Black patients identified for extra care by more than half.

The paper classifies occupations by their technical susceptibility to computerisation and estimates that 47 percent of U.S. employment is in high-risk occupations. It is an exposure exercise—not a forecast of realized job loss—and became a benchmark for later task-based measures.

Greater exposure to industrial robots reduced employment and wages in affected U.S. commuting zones. The estimates imply that each additional robot per thousand workers lowered the employment-to-population ratio and average wages.

In this essay, I begin by identifying the reasons that automation has not wiped out a majority of jobs over the decades and centuries. Journalists and even expert commentators tend to overstate the extent of machine substitution for human labor and ignore the strong complementarities between automation and labor that increase productivity, raise earnings, and augment demand for labor.

Orthogonal scores and sample splitting allow flexible machine-learning estimates of nuisance functions without invalidating inference on causal parameters. The resulting estimators remain approximately unbiased and normally distributed under broad conditions.

In a static version where capital is fixed and technology is exogenous, automation reduces employment and the labor share, and may even reduce wages, while the creation of new tasks has the opposite effects. Stability is a consequence of the fact that automation reduces the cost of producing using labor, and thus discourages further automation and encourages the creation of new tasks.

The paper introduces causal forests for estimating how treatment effects vary across people or settings. It establishes asymptotic theory and shows how forest estimates can support confidence intervals and heterogeneity analysis.

We present a framework for understanding the effects of automation and other types of technological changes on labor demand, and use it to interpret changes in US employment over the recent past. Automation, which enables capital to replace labor in tasks it was previously engaged in, shifts the task content of production against labor because of a displacement effect.

Across industries in seventeen countries, greater robot use raised labor productivity and total factor productivity while lowering output prices. The estimates do not show a significant fall in total employment, but they do show a declining employment share for lower-skilled workers.

We examined the productivity effects of a generative artificial intelligence (AI) technology, the assistive chatbot ChatGPT, in the context of midlevel professional writing tasks. Our results show that ChatGPT substantially raised productivity: The average time taken decreased by 40% and output quality rose by 18%.

Machines are increasingly doing “intelligent” things. Face recognition algorithms use a large dataset of photos labeled as having a face or not to estimate a function that predicts the presence y of a face from pixels x.

Generalized random forests extend random forests beyond prediction to estimate heterogeneous economic parameters such as treatment effects. The method provides consistent estimates and valid confidence intervals while adapting flexibly to local variation.

Firms investing in AI subsequently experience faster growth in sales, employment, and product innovation. The gains are concentrated among larger firms, suggesting AI can reinforce differences between leading firms and the rest.

Computers are now involved in many economic transactions and can capture data associated with these transactions, which can then be manipulated and analyzed. Conventional statistical and econometric techniques such as regression often work well, but there are issues unique to big datasets that may require different tools.

An ever-increasing share of human interaction, communication, and culture is recorded as digital text. We provide an introduction to the use of text as an input to economic research.

We discuss the relevance of the recent machine learning (ML) literature for economics and econometrics. Finally, we highlight newly developed methods at the intersection of ML and econometrics that typically perform better than either off-the-shelf ML or more traditional econometric methods when applied to particular classes of problems, including causal inference for average treatment effects, optimal policy estimation, and estimation of the…

Access to AI assistance increases worker productivity, as measured by issues resolved per hour, by 15% on average, with substantial heterogeneity across workers. The effects vary significantly across different agents.

Data is nonrival: a person’s location history, medical records, and driving data can be used by many firms simultaneously. Nonrivalry leads to increasing returns.

We present evidence from various industries, products, and firms showing that research effort is rising substantially while research productivity is declining sharply. More generally, everywhere we look we find that ideas, and the exponential growth they imply, are getting harder to find.

wage structure over the last four decades are accounted for by relative wage declines of worker groups specialized in routine tasks in industries experiencing rapid automation. We report robust evidence in favor of this relationship and show that regression models incorporating task displacement explain much of the changes in education wage differentials between 1980 and 2016.

Increasingly, algorithms are supplanting human decision-makers in pricing goods and services. We find that the algorithms consistently learn to charge supracompetitive prices, without communicating with one another.

Rapid advances in artificial intelligence (AI) and automation technologies have the potential to significantly disrupt labor markets. In this paper we discuss the barriers that inhibit scientists from measuring the effects of AI and automation on the future of work.

Our task-based framework emphasizes the displacement effect that automation creates as machines and AI replace labor in tasks that it used to perform. This displacement effect tends to reduce the demand for labor and wages.

Abstract Artificial intelligence (AI) is set to influence every aspect of our lives, not least the way production is organised. AI, as a technology platform, can automate tasks previously performed by labour or create new tasks and activities in which humans can be productively employed.

Among Taiwanese electronics firms, AI patenting is positively associated with productivity and employment, with effects similar in magnitude to other patenting. AI invention also shifts workforce composition away from workers with college-level education or less.

Abstract We use detailed administrative data to study the adjustment of local labor markets to industrial robots in Germany. Robot exposure, as predicted by a shift-share variable, is associated with displacement effects in manufacturing, but those are fully offset by new jobs in services.

The review documents rapid growth in AI activity and synthesizes evidence on productivity, labor, inequality, and competition. It argues that labor and antitrust policy will materially shape whether AI's gains are broad-based.

Despite the interest in this area, we have limited ability to study the effects of AI on occupations, firms, industries, and geographies because of limited availability of data that measures exposure to AI. We describe how our measures can be useful to scholars and policy‐makers interested in identifying the effect of AI on markets.

Most empirical policy work focuses on causal inference. Solving these “prediction policy problems” requires more than simple regression techniques, since these are tuned to generating unbiased estimates of coefficients rather than minimizing prediction error.

Abstract Can machine learning improve human decision making? Even accounting for these concerns, our results suggest potentially large welfare gains: one policy simulation shows crime reductions up to 24.

This paper examines the potential impact of artificial intelligence (A. One theme that emerges is based on Baumol’s “cost disease” insight: growth may be constrained not by what we are good at but rather by what is essential and yet hard to improve.

Using worker-level tasks rather than treating whole occupations as automatable, the authors estimate that about 9 percent of jobs across 21 OECD countries are automatable. The result is far below occupation-based estimates and varies with workplace organization and worker education.

Artificial intelligence may greatly increase the efficiency of the existing economy. We distinguish between automation-oriented applications such as robotics and the potential for recent developments in “deep learning” to serve as a general-purpose method of invention, finding strong evidence of a “shift” in the importance of application-oriented learning research since 2009.

The authors explain why rapid advances in AI may coexist with weak measured productivity. Historical experience suggests that complementary innovation, business-process redesign, and organizational learning create long implementation lags before aggregate gains appear.

I provide general instructions and demonstrate specific examples of how to take advantage of each of these, classifying the LLM capabilities from experimental to highly useful. Moreover, these gains will grow as the performance of AI systems continues to improve.

The paper links patent text to occupational task descriptions to measure technology exposure. Unlike software and industrial robots, AI is directed toward high-skilled tasks; extrapolating historical substitution patterns implies lower 90:10 wage inequality but little change at the very top.

Abstract In 1950, Alan Turing proposed a test of whether a machine was intelligent: could a machine imitate a human so well that its answers to questions were indistinguishable from a human's? Ever since, creating intelligence that matches human intelligence has implicitly or explicitly been the goal of thousands of researchers, engineers, and entrepreneurs.

AI methods have diffused rapidly across scientific fields and are associated with higher-impact but less recombinatorially novel research. The authors frame AI as a general method of invention that is beginning to reshape discovery and the organization of science.

Third, we provide several simple economic models to describe how policy can counter these effects, even in the case of a “singularity” where machines come to dominate human labor. Fourth, we describe the two main channels through which technological progress may lead to technological unemployment – via efficiency wage effects and as a transitional phenomenon.

Progress in artificial intelligence and related forms of automation technologies threatens to reverse the gains that developing countries and emerging markets have experienced from integrating into the world economy over the past half century, aggravating poverty and inequality. We analyze the economic forces behind these developments and describe economic policies that would mitigate the adverse effects on developing and emerging economies while…

End-of-life health care spending In the United States, one-quarter of Medicare spending occurs in the last 12 months of life, which is commonly seen as evidence of waste. used predictive modeling to reassess this interpretation.

It starts from a task-based model of AI’s effects, working through automation and task complementarities. So long as AI’s microeconomic effects are driven by cost savings/productivity improvements at the task level, its macroeconomic consequences will be given by a version of Hulten’s theorem: GDP and aggregate productivity gains can be estimated by what fraction of tasks are impacted and average task-level cost savings.

We find that fewer than 6% of firms used any of the AI‐related technologies we measure, though most very large firms reported at least some AI use. AI use in production, while varying considerably by industry, was found in every sector of the economy and clustered with emerging technologies, such as cloud computing and robotics.

We construct the first measure of firms’ workforce exposures to Generative AI and show that an “Artificial-Minus-Human” (AMH) portfolio earned 5% in the two weeks following the release of ChatGPT. The labor-exposure effect is more pronounced for firms with greater data assets and is distinct from the effect of firms’ product exposures to AI.

While the utopian vision of the current Information Age was that computerization would flatten economic hierarchies by democratizing information, the opposite has occurred. Information, it turns out, is merely an input into a more consequential economic function, decision-making, which is the province of elite experts.

Generative artificial intelligence (AI) is a potentially important new technology, but its impact on the economy depends on the speed and intensity of adoption. This suggests that substantial productivity gains from generative AI are possible.

Full automation using Artificial Intelligence (AI) predictions may not be optimal if humans have information not available to the AI (contextual information). Results show that providing (i) AI predictions does not improve performance on average, whereas (ii) contextual information does.

Abstract We use machine learning as a tool to study decision making, focusing specifically on how physicians diagnose heart attack. We provide suggestive evidence on the psychology underlying these errors.

AI can raise output while producing sharply different distributional outcomes depending on whether it substitutes for workers or complements them. The authors emphasize policy choices that broaden ownership, guide innovation, and share the gains.

We study the adoption of ChatGPT, the icon of Generative AI, using a large-scale survey linked to comprehensive register data in Denmark. Surveying 18,000 workers from 11 exposed occupations, we document that ChatGPT is widespread, especially among younger and less-experienced workers.

In this paper, we estimate that wider adoption of AI could lead to savings of 5 to 10 percent in US healthcare spending—roughly $200 billion to $360 billion annually in 2019 dollars. These estimates are based on specific AI-enabled use cases that employ today’s technologies, are attainable within the next five years, and would not sacrifice quality or access.

To interpret these patterns, we develop a model that separates direct substitution from indirect reallocative effects of labor-saving technologies. Using an instrument based on historical university hiring networks, we find causal evidence consistent with these predictions.

We study the impact of AI on labor markets using establishment-level data on vacancies with detailed occupation and skill information comprising the near-universe of online vacancies in the US from 2010 onwards. We find no discernible relationship between AI exposure and employment or wage growth at the occupation or industry level, however, implying that AI is currently substituting for humans in a subset of tasks but it is not yet having detectable…

We summarize existing empirical findings regarding the adoption of robotics and AI and its effects on aggregated labor and productivity, and argue for more systematic collection of the use of these technologies at the firm level. Further, firm-level data would also allow for studies of effects on firms of different sizes, the role of market structure in technology adoption, the impact on entrepreneurs and innovators, and the effect on regional…

This paper describes the adoption of automation technologies by US firms across all economic sectors by leveraging a new module introduced in the 2019 Annual Business Survey, conducted by the US Census Bureau in partnership with the National Center for Science and Engineering Statistics (NCSES). Adopters report that these technologies raised skill requirements and led to greater demand for skilled labor but brought limited or ambiguous effects to their…

We find that reliance on AI, a prediction tool, increases decision variation, which, in turn, raises challenges if decisions across the organization interact. Consequently, we show that there are important cases where AI adoption will be enhanced when it can be adopted beyond tasks but as part of a designed organizational system.

Abstract This article studies the effects of the introduction of artificial intelligence (AI) into teams in a laboratory experiment. We demonstrate that even in a task where AI outperforms humans, automation decreases overall team performance and increases coordination failures.

We show that they autonomously sustain collusive supra-competitive profits without agreement, communication, or intent. We demonstrate that two separate mechanisms are underlying this collusion and characterize when each one arises.

firms' workforce composition and organization associated with the use of AI technologies. Furthermore, AI investments are associated with a flattening of the firms' hierarchical structure, with significant increases in the share of workers at the junior level and decreases in shares of workers in middle-management and senior roles.

We study the early labor market impacts of AI chatbots by linking large-scale adoption surveys to administrative labor market records in Denmark. Yet these currents have not broken the surface: using difference-indifferences, we estimate precise null effects on earnings and recorded hours at both the worker and workplace levels, ruling out effects larger than 2% two years after the launch of ChatGPT.

Despite ethical and historical arguments for removing race from clinical algorithms, the consequences of removal remain unclear. More broadly, this study shows that race adjustments may be beneficial when the data quality of key predictors in clinical algorithms differs by race group.

We present evidence from a field experiment across 66 firms and 7,137 knowledge workers. Workers were randomly selected to access a generative AI tool integrated into applications they already used at work for email, meetings, and writing.

We study the adoption of ChatGPT, the icon of Generative AI, using a large-scale survey experiment linked to comprehensive register data in Denmark. Surveying 100,000 workers from 11 exposed occupations, we document ChatGPT is pervasive: half of workers have used it, with younger, less experienced, higher-achieving, and especially male workers leading the curve.

First, when a social planner builds the algorithm herself, her equity preference has no effect on the training procedure. Under such disclosure, the use of algorithms strictly reduces the extent of discrimination relative to a world in which humans make all the decisions.

We find, perhaps surprisingly, that the pace of change has slowed over time. This comparative decline is not because the job market is stable today but rather because past changes were so profound.

The paper analyzes markets in which platforms compete for scarce consumer attention and monetize it through advertisers. It highlights how platform incentives, congestion, and user switching shape prices, content, and welfare.

Behavioral economics suggests one reason these algorithms so often fail: choices can systematically deviate from preferences. For example, research shows that prejudice can arise not just from preferences and beliefs, but also from the context in which people choose.

Abstract This article studies the effects of automation in a task-based economy in which some jobs pay workers rents—wages above workers' outside options. We show that automation targets high-rent tasks, dissipating rents, amplifying wage losses, and reducing within-group wage dispersion in exposed groups.

Drawing insights from the field of innovation economics, we discuss the likely competitive environment shaping generative AI advances. We suggest the likely paths through which incumbent firms may restrict entry, confining newcomers to subordinate roles and stifling broad sectoral innovation.

This review treats privacy as an economic problem shaped by incomplete information, behavioral biases, and externalities. It shows why individual disclosure choices can diverge from socially desirable outcomes and surveys the trade-offs facing firms and regulators.

We survey nearly 6,000 senior business executives at US, UK, German, and Australian firms to develop new evidence on AI adoption and its effects on jobs, productivity, and output. Specifically, we ask executives about AI usage, its effects at their own firms over the past three years and, looking ahead, what they anticipate over the next three years.

I first introduce the task model and explain why this framework offers a compelling way to think about recent labor market trends and the effects of automation technologies. This substitution reduces costs, creating a positive productivity effect, but also reduces employment opportunities for workers displaced from automated tasks, creating a negative displacement effect.

We also provide a new way of measuring AI knowledge spillovers across firms and find large spillovers. Finally, our work suggests numerous ways in which LLMs such as ChatGPT can be used in other applications.

Production is a sequence of steps that can be executed (1) manually, (2) augmented with AI, or (3) fully automated within contiguous AI-executed steps called “chains. Empirical evidence supports the model’s key predictions that (1) AI-executed steps co-occur in chains, (2) dispersion of AI-exposed steps lowers AI execution at the job level, and (3) adjacency to AI-executed steps increases the likelihood that a step is AI-executed.

We find that automation is accelerating and supplanting a broader set of low-wage routine jobs since the 2008–2009 financial crisis. However, interpersonal job growth does not appear to be enough, as it was prior to the financial crisis, to fully offset the negative effects of automation on low-wage routine jobs.

This paper surveys the relevant existing literature that can help researchers and policy makers understand the drivers of competition in markets that constitute the provision of artificial intelligence products. The focus is on three broad markets: training data, input data, and AI predictions.

Does generative artificial intelligence (AI) reinforce or reduce productivity differences across workers? Existing evidence largely studies AI within firms and occupations, where organizational selection compresses educational heterogeneity, leaving unclear whether AI narrows productivity gaps across individuals with different levels of education.

This study investigates the effects of artificial intelligence (AI) adoption in organizations. Fourth, the optimal AI adoption increases average wages and reduces intra-team wage inequality.

Artificial intelligence (AI) is transforming production across all sectors of the economy, with the potential to both complement and substitute for traditional labor inputs. Dozens of recent academic studies demonstrate that AI can contribute to the healthcare value chain, by improving both diagnostic accuracy and treatment recommendations.

We use novel data from a survey of nearly 750 corporate executives to study the effects of artificial intelligence (AI) on productivity and the workforce. Labor productivity gains are positive, vary across sectors, and are expected to strengthen in 2026, with the largest effects concentrated in high-skill services and finance.

When the rival uses a learning rule to set prices, we show via simulations that outcomes rapidly converge to a coercive outcome. Finally, we demonstrate the implications of our framework for platform design.

Economists have often viewed the adoption of artificial intelligence (AI) as a standard process innovation where we expect that efficiency will drive adoption in competitive markets. It is shown that, in a competitive market, this increases the short-run elasticity of supply and may or may not increase average equilibrium prices.

This paper reviews firm-level data on artificial intelligence (AI) and the emerging evidence on AI’s economic effects. It synthesizes evidence on AI’s effects on firm growth, valuation, productivity, risk, labor, competition, financial markets, and applications.

Evidence suggests both top-down and bottom-up diffusion: worker use can occur without firm adoption, and vice versa. Regression results show a positive relationship between firm performance and AI integration breadth.

This paper examines the evolving structure and competition dynamics of the rapidly growing market for foundation models, with a focus on large language models (LLMs). We describe the technological characteristics that shape the AI industry and have given rise to fierce competition among the leading players.

We calibrate the boom size to match the observed increase in investment projected through 2027, implying that a boom raises AI-sector productivity by a factor of roughly 2. We then calibrate a two-year window of a 50% annual probability of an increase of the same magnitude, generating a range of scenarios consistent with the wide variety of industry forecasts, along with an elevated permanent probability tied to the valuation of the aggregate market.

Using new data from the Gallup Workforce Panel, we show that the apparent partisan divide in workplace AI adoption is largely an artifact of educational and occupational sorting rather than ideological differences in technology adoption. 1% versus 25% in Q1:2026—and exhibit deeper task-level integration across a broader range of work activities, this raw gap shrinks to statistical insignificance once we control for educational attainment, and reverses…

We use recent advances in natural language processing and large language models to construct novel measures of technology exposure for workers that span almost two centuries. Combining our measures with Census data on occupation employment, we show that technological progress over the 20th century has led to economically meaningful shifts in labor demand across occupations: it has consistently increased demand for occupations with higher education…

Adoption reduces unit costs, displaces some types of workers, and depresses wages for those workers via diminishing returns elsewhere, while leaking AI fees abroad. We identify conditions under which market power in AI leads to a “double harm” for displaced workers, who may experience real wages cuts when AI becomes available at low prices, and then experience further harm from increases in AI prices.

We measure how workers use genAI for their jobs in a nationally representative survey linking genAI adoption to detailed occupations and tasks. Our data provide the first task-level genAI adoption indexes, which we show can inform analyses of genAI’s labor market impact.

We document and explain the gap between measures of AI exposure and measures of AI adoption in the workplace. Using the representative German DiWaBe employee survey linked to worker and establishment information, we compare worker-reported AI use to prominent exposure measures and find that the relationship is weak.

We develop a dynamic task-based model to quantify the general-equilibrium effects of task-specific technical change. We develop a computationally efficient procedure to estimate the model using panel data and a new database of task-level skill requirements.

Using a new firm-level measure of AI investment based on AI-skilled employment—spanning machine learning through generative and agentic AI—we show that AI investments are associated with productivity growth in recent years, but not over the previous decade. Overall, our findings suggest that AI investment generates productivity growth by creating organization capital.

Do generative AI models, particularly large language models (LLMs), exhibit systematic behavioral biases in economic and financial decisions? Prompting LLMs to make rational decisions reduces biases.

The paper develops a framework for evaluating credential-coded algorithmic screens under existing civil rights law. AI-powered hiring tools trained on historical data often encode and automate bachelor's degree requirements as a proxy for worker skill, producing what this paper terms algorithmic credentialism.

This study provides pre-registered, experimental evidence on the use of non-generative artificial intelligence (AI) chatbots to support students in large-enrollment undergraduate courses. We find the chatbot messaging increased students’ final grades and engagement with academic supports, such as tutoring.

In two focal applications, we show that this standard alignment practice can backfire. We show that while the adjustments engineers use correctly incentivize choosing, they can simultaneously reduce the incentives to learn.

The paper studies how the release of generative AI changed work on online labor platforms. It finds early evidence of reduced demand and earnings in highly exposed freelance occupations, alongside shifts in the types of skills clients request.

, we show that the distributions of both wages and productivity have spread out over time, as the right tail lengthens for both. The most likely international factor explaining these wage increases is the skill-biased technological change of the digital revolution.

Specifically, we show that algorithmic discrimination exists when measurement errors exist in either the outcome or the predictors, and there is endogenous selection for participation in the observed data. We show that although equalized odds constraints can be employed as bias-mitigating strategies, such constraints may increase algorithmic discrimination when there is measurement error in the dependent variable.

In a field experiment with consultants, access to GPT-4 improved speed and quality for tasks inside the model's capabilities. For tasks outside that frontier, reliance on AI made participants less likely to reach the correct answer.

The projected effects span all wage levels, with higher-income jobs potentially facing greater exposure to LLM capabilities and LLM-powered software. Our analysis suggests that, with access to an LLM, about 15% of all worker tasks in the US could be completed significantly faster at the same level of quality.

Generative AI tools hold promise to increase human productivity. Observed heterogenous effects show promise for AI pair programmers to help people transition into software development careers.

The paper models how highly capable AI could automate research and other growth-producing tasks. It identifies the assumptions under which AI leads to very rapid growth and the bottlenecks that could keep the transition more gradual.

The analysis asks how the tax system should respond when automation displaces routine labor. It finds a case for temporarily taxing robots or automation rents during the transition, while relying more on broader redistribution in the long run.

The paper studies how automation changes the distribution of income and wealth through wages, capital returns, and endogenous investment. It shows that the transition can generate substantial inequality even when automation raises aggregate productivity.

General-purpose technologies can initially depress measured productivity because firms must make large, unmeasured investments in software, workflows, skills, and organizational change. Productivity accelerates only after this complementary capital is accumulated.

The paper studies neural-network methods for estimating individual-level heterogeneity in economic relationships. It develops conditions for estimation and inference when flexible deep-learning models are used to recover heterogeneous effects.

The paper develops an economic framework for balancing the social value of statistical data against disclosure risk. It argues that privacy protection and data usefulness must be evaluated jointly rather than as separate technical goals.

The model shows that automation can raise aggregate output while reducing wages and increasing inequality during the transition. Whether workers ultimately benefit depends on productivity gains, capital accumulation, and redistributive policy.