|
on Artificial Intelligence |
| By: | Joacim Tåg; Fredrik Heyman; Malin Gardberg; Martin Olsson |
| Abstract: | We examine how gender-based occupational sorting before the release of ChatGPT relates to predicted exposure to generative AI and its potential implications for the gender wage gap. Using Swedish administrative data, we find that women are overrepresented in occupations predicted to be more affected by generative AI. Mechanical partial-equilibrium simulations, based on hypothesized deviations from the 2021 occupational and wage distribution and incorporating predicted AI exposure and task complementarity, show that generative AI can widen the gender wage gap through existing patterns of gender-based occupational sorting. |
| Keywords: | Generative AI, gender wage gap, technological change, occupational sorting, complementarity |
| JEL: | J16 J31 O33 J24 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:crm:wpaper:26118 |
| By: | Abdel Haq, Omar (Harvard Business School); Chandra, Amitabh (Harvard Business School and Harvard Kennedy School); Jagelka, Tomáš (University of Bonn); Luttmer, Erzo (Dartmouth College); Schwartzstein, Joshua (Harvard Business School) |
| Abstract: | Large Language Models (LLMs) are trained on a prodigious corpus of human writing and may reveal human preferences over characteristics of life courses, such as income, longevity, and working conditions. We present OpenAI's GPT-5.4 and a broadly representative sample of Americans with pairs of life stories and ask them to choose the life they would prefer for themselves. A person's choice is better predicted by the LLM's choice than by another person’s choice over the same stories, and LLM valuations of several life attributes are similar to those derived from human responses. Our results suggest that LLM responses offer a scalable and cost-effective complement to existing methods for studying human preferences. |
| Keywords: | generative AI, preference estimation methods, choice experiments, survey validation |
| JEL: | D0 H0 I0 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18634 |
| By: | Kathryn Bonney; Cory Breaux; Emin Dinlersoz; Lucia Foster; John Haltiwanger; Aditya Pande |
| Abstract: | Using novel, nationally representative data from the 2026 AI supplement to the U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS), we characterize AI diffusion across three interconnected layers: overall firm use, deployment across business functions, and worker-task use. This multi-layered approach provides a nuanced picture of business AI adoption. During the supplement reference period (Nov 2025-Jan 2026), 18% of firms used AI in a business function, rising to 32% on an employment-weighted basis; adoption is expected to reach 22% within six months. AI use is substantially higher in large firms and knowledge-intensive sectors, with use rates reaching 50%-60% (60%-70%, employment-weighted) for very large firms in the Information, Professional Services, and Finance sectors. Among adopting firms, the scope of use remains limited: 57% of users integrate AI in three or fewer business functions, most commonly Sales and Marketing (52%), Strategy and Business Development (45%), and IT (41%). In 23% (41%, employment-weighted) of firms, workers use AI in work-related tasks. Writing, document analysis, and information search are the leading Generative AI use in tasks, though 65% of firms limit use to three or fewer tasks. The evidence points to both top-down and bottom-up diffusion channels: worker task use sometimes occurs without formal firm-level adoption, and firm-level adoption sometimes occurs without worker task use. Most users (66%) rely on AI solely to augment tasks, while AI-related employment decreases are rare, occurring in only 2% of firms. Regression analysis shows a robust positive correlation between firm commercial performance and the breadth of AI integration, including functional deployment, task-level use, and operational investment. A distinct divergence emerges, however, with respect to labor outcomes. Functional breadth and operational investment are positively associated with employment decreases, whereas worker-task integration shows no significant link to headcount reduction once functional integration and operational investment are taken into account. |
| Keywords: | Artificial Intelligence, AI, Technology Diffusion, Generative AI |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:cen:wpaper:26-25 |
| By: | Richard Audoly; Miles Guerin; Giorgio Topa |
| Abstract: | As generative AI tools become more widely used, a key issue is the technology’s impact on labor demand. Where might we find evidence of that impact? In this post, we examine whether early evidence of AI’s effect on the labor market appears in firms’ job postings. We combine an occupational measure of AI exposure with detailed U.S. job-posting data from Lightcast, which aggregates listings from company career pages, national and local job boards, and job-listing aggregators. Using this data, we test whether postings for AI-exposed occupations declined disproportionately since the release of ChatGPT in late 2022. We find that, while overall hiring has slowed since then, the evidence from job postings provides little indication of a distinct AI-driven decline in labor demand. |
| Keywords: | artificial intelligence (AI); labor demand; job postings; task automation |
| JEL: | J23 O33 |
| Date: | 2026–05–14 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fednls:103247 |
| By: | Lee C. Tucker |
| Abstract: | Using detailed tabulations from matched employer-employee administrative data, I document evidence of an immediate, sizable, and persistent decrease in the level of early career (22-24 year old) hires following introduction of ChatGPT within the industry-state cells that are most exposed to AI. The decline in hires is the primary cause of large observed declines in employment over the subsequent period. Regressionadjusted employment of early career workers in the most AI-exposed quintile of industry-state cells declined by 12% over the 10 quarters following the introduction of ChatGPT, even as employment in lessexposed industries has remained stable. The rate of hiring largely recovered by early 2025, attributable to a smaller employment base. Earnings growth of early career workers in the most exposed industries slowed slightly relative to those in less exposed industries. Although the most AI-exposed quintile of detailed industries is dominated by a handful of industry sectors, I find that the association of higher AI exposure with reduced early career employment and fewer hires is observed across most sectors of the economy. Timing of effects in event studies is consistent with an immediate effect on hiring following introduction of ChatGPT. However, triple difference estimates provide some evidence of earlier trend shifts on employment, hiring, and separations around the onset of the COVID pandemic. I discuss potential explanations, including the increase in remote work and increased educational attainment among workers in AI-exposed occupations. Nonetheless, job gains to early career workers and backfill hires show evidence of discontinuous decline at the time of ChatGPT’s release in comparison to older workers in the same industries. A local projections analysis at the NAICS industry group level shows that industries with high AI exposure are not particularly sensitive to unexpected fluctuations in monetary policy on average relative to other industries in employment, hiring, or separations. A historical decomposition suggests that up to one quarter of relative early career employment declines through 2025q2 may be attributable to monetary policy shocks through 2023, but the analysis does not find evidence that these shocks can explain the rapid decline in hires at the most AI-exposed firms in comparison to others. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:cen:wpaper:26-27 |
| By: | Philip Moreira Tomei; Bouke Klein Teeselink |
| Abstract: | Which jobs can AI learn to do? We examine this for every occupation in the US economy. Existing indices measure the overlap between AI capabilities and occupational tasks rather than which tasks AI systems can learn to perform, and as a result misclassify occupations where the gap between present capability and learnability is large. Reinforcement learning in post-training, now the dominant paradigm at the frontier, is structured around task completion and maps more directly onto the task-based architecture of occupational classifications than prior approaches. Using LLM annotators guided by a rubric developed with RL experts and validated against confirmed deployment cases, we score all 17, 951 ONET tasks for training feasibility and aggregate to the occupation level, producing an RL Feasibility Index. The index diverges sharply from existing AI exposure measures for specific occupation groups: power plant operators, railroad conductors, and aircraft cargo handling supervisors score high on RL feasibility but low on general AI exposure, while creative and interpersonal roles (musicians, physicians, natural sciences managers) show the reverse. These divergences carry direct implications for policy interventions. |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2605.02598 |
| By: | Erdinc Akyildirim; Suzan Bekci; Giray Gozgor |
| Abstract: | This paper constructs an AI Attention Index from LexisNexis news coverage and embeds it within an augmented Phillips curve framework. It then examines the relationship between AI attention and unemployment in the United States using monthly data from January 2000 to December 2025. We find that greater AI attention is associated with lower unemployment. Nonlinear estimates reveal a U-shaped relationship, indicating diminishing marginal effects within the observed data range. The relationship weakens after the onset of COVID-19, with both linear and nonlinear effects reduced. These findings indicate that labour market effects of AI-related expectations are sensitive to macroeconomic regime shifts. |
| Keywords: | artificial intelligence, AI, unemployment, Phillips Curve, nonlinearity, economic policy uncertainty, COVID-19 |
| JEL: | E24 E31 O33 D84 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_12653 |
| By: | James Lennox; Janine Dixon |
| Abstract: | This paper develops a task-based computable general equilibrium model to analyse the long-run economic effects of generative AI (GenAI) on the Australian economy. Each occupation performs a continuum of tasks executed in three modes: with raw labour; with AI-augmented labour; or automated using equipment and AI services. Task-level productivities in AI-using modes are draws from correlated Frechet distributions, captur ing heterogeneous within-occupation exposure. The model covers 45 industries and 97 occupations, calibrated to occupation-level GenAI exposure scores. The reference simulation yields a 29.8% real GDP increase: roughly one third from task-level productivity gains, the rest from capital deepening and general equilibrium reallocation. Real consumption - our long-run welfare metric - rises by 16.2%, substantially less because additional investment is required to equip automated tasks. Augmentation accounts for more tasks than automation in nearly all industries and occupations. Labour-market adjustment is dominated by within-occupation change - extensive-margin task reallocation equivalent to two thirds of current work - rather than net employment shifts between occupations. Losses con centrate in clerical, administrative, and sales roles, while most blue-collar occupations gain. Real wage effects are weakly correlated with initial wages; the rising capital share of income may matter more for distribution. Sensitivity analysis shows aggregate outcomes hinge on the distribution of task-level productivity gains: fatter tails roughly double the GDP gain while preserving the adjustment pattern, whereas variation in the dependence parameter shifts the augmentation - automation balance and the incidence of adjustment. Conventional substitution elasticities matter less. |
| Keywords: | Generative artificial intelligence, Computable general equilibrium, Task-based production, Occupational reallocation, Augmentation, Automation |
| JEL: | C68 J23 J24 O33 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:cop:wpaper:g-367 |
| By: | Levy Yeyati, Eduardo |
| Abstract: | We study how the speed of Artificial Intelligence (AI) adoption affects labor market outcomes during technological transitions. In a dynamic model where displaced routine workers enter a retraining pipeline with finite capacity, faster adoption compresses the displacement window without reducing total displacement, overwhelming the pipeline and generating permanent labor force exit through worker discouragement. The central result is that, even when two economies share the same long-run automation level, adoption speed alone determines transition welfare: faster adoption produces a larger discourage stock, a steeper and more persistent decline in labor force participation, and a sustained compression of the labor share throughout the transition window. Non-routine employment and wages exhibit a crossing pattern initially higher under fast adoption, then lower so that faster adoption can simultaneously raise long-run wages for survivors while permanently reducing participation. Social welfare is strictly concave in adoption speed and maximized at an interior optimum below the market rate, because firms do not internalize the congestion externality they impose on the retraining queue, the irreversibility of permanent exit, or the wage depression borne by non-routine incumbents. The socially optimal speed and retraining capacity are complements: stronger institutions raise the optimal adoption speed. |
| Keywords: | artificial intelligence;labor market;labor force |
| JEL: | O33 J24 J64 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:idb:brikps:14580 |
| By: | Ujjwal KC (Melbourne Institute of Applied Economic and Social Research); Jan Kabatek (Melbourne Institute of Applied Economic and Social Research) |
| Abstract: | Evidence-based policy design increasingly relies on integrated administrative data, yet access to and effective use of these data remain constrained by technical and institutional barriers. This paper introduces the Breaking Down Barriers Assistant, an AI-powered analytical platform designed to democratise access to complex, linked administrative datasets for policy analysis. The system enables users to query secure, pre-aggregated Australian administrative and survey data using natural language and supports the full policy research cycle through three integrated tools: variable discovery, visual analytics, and automated reporting. Methodologically, the BDB Assistant combines Retrieval-Augmented Generation with controlled code generation and execution, ensuring that all analytical outputs are grounded in verified metadata, auditable Python code, and human-in-the-loop validation. Using benchmark comparisons with established platforms such as YouthView and the BDB Community Profiles, we demonstrate that the Assistant can accurately reproduce complex spatial and longitudinal policy analyses that traditionally require advanced technical expertise. The findings show that the system substantially reduces time-to-insight while maintaining strict standards of accuracy, transparency, and privacy. The BDB Assistant illustrates how responsibly governed generative AI can expand analytical capacity within government and the social sector, supporting more timely, rigorous, and locally targeted evidence-based policymaking. |
| Keywords: | Evidence-based policymaking, Artificial Intelligence, Generative AI, Large Language Models, AI-Assistants |
| URL: | https://d.repec.org/n?u=RePEc:iae:iaewps:wp2026n04 |
| By: | Fernando E. Alvarez; David Argente; Joyce Chow; Diana Van Patten |
| Abstract: | Data centers are the physical infrastructure behind cloud computing, artificial intelligence, and enterprise software. The rapid diffusion of artificial intelligence (AI) is intensifying demand for compute, accelerating investment in data centers, and raising concerns about the local economic and environmental footprint of these facilities. Their expansion creates a local policy tradeoff. A data center can bring capital investment, construction activity, and specialized employment, but it can also increase demand for electricity, land, and grid capacity. This paper studies these effects at the U.S. county level. We assemble a facility-level panel of global data centers with precise coordinates, scale metrics, and annualized revenue. We map facilities to U.S. counties and combine them with County Business Patterns, county-level IRS income, county-level house prices, and electricity prices. To address endogenous siting, we instrument for data center growth using two shift-share instruments, which leverage pre-existing proximity to InterTubes long-haul fiber nodes and the 1980 county share of U.S. urban college population as shares, and both Chinese and rest-of-the-world data center revenue growth as shifts. The IV estimates show positive effects on total employment, data-processing employment, construction employment, establishments, house prices, and electricity prices at different horizons after data center growth. We also find positive effects on tax returns, adjusted gross income, and wages, while annual payroll responds less robustly. The results suggest that data centers create measurable local activity, increase house prices, and affect local electricity markets through higher prices. |
| JEL: | D8 O3 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35194 |
| By: | Viet Trinh; Tan Nguyen; Minh-Huyen Phan; Quan Luu |
| Abstract: | This study presents a novel, AI-driven framework for assessing Environmental, Social, and Governance (ESG) performance in European small and medium-sized enterprises (SMEs). An initial phase established expert-validated ESG baseline scores from a subset of the Flash Eurobarometer FL549 survey data. In the second phase, a scalable AI agent system, built on the n8n automation platform, applied these baselines to perform automated ESG classification and generate contextual recommendations using large language models (LLMs). The results demonstrate the AI system's high consistency with human-derived outputs, thereby supporting more effective monitoring and intervention strategies aligned with the European Green Deal. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2605.00841 |
| By: | Jens Robert Schoendube; Barbara Schoendube-Pirchegger (Faculty of Economics and Management, Otto-von-Guericke University Magdeburg) |
| Abstract: | In this paper we model the interaction between an auditor and a client firm. The client firm’s manager can either report truthfully or commit fraud. The auditor needs to plan a two stage audit that allows to detect fraud. In the first stage an AI tool is employed that provides a signal about the quality of the client’s internal control system (ICS). Classifying the ICS as weak or strong, the signal alters the auditor’s expectations regarding the client’s fraud probability. In the second stage, the auditor decides about her audit effort conditional on the information provided by the AI. Comparing the AI setting to a benchmark setting without AI use, we find that employing the AI tool reduces the manager’s incentives to commit fraud. At the same time it reduces the equilibrium effort provided by the auditor. As a consequence, the probability that actual fraud is detected remains unchanged. We extend our model and allow the AI tool to be customized such that it can either focus on detection of the weak ICS, the strong ICS, or on both equally. We find that the AI specification that minimizes ex ante probability for fraud not necessarily coincides with the specification that minimizes auditing costs. It follows that the auditor in charge of customizing the AI cannot necessarily be expected to do so in a fraud minimizing way. |
| Keywords: | Artificial Intelligence, Auditing, Game Theory, Fraud detection |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:mag:wpaper:25004 |
| By: | Anand, Kartik; Leonello, Agnese; Panetti, Ettore; Kazinnik, Sophia |
| Abstract: | Does artificial intelligence (AI) pose a threat to financial stability? We study AI investor behavior, specifically Q-learning and large language model (LLM) investors, in a mutual fund redemption problem with economic and strategic uncertainty. Different AI architectures generate systematically different outcomes. Q-learning investors coordinate well but under default risk exhibit excessive redemption that amplifies fragility. LLM investors internalize equilibrium structure but display belief heterogeneity, weakening coordination and predictability. Our findings show that AI architecture is a first-order determinant of financial stability. JEL Classification: G01, G23, C63 |
| Keywords: | AI agents, coordination games, financial stability, large language models, Q-learning, strategic uncertainty |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:ecb:ecbwps:20263225 |
| By: | Heng-fu Zou (The World Bank, Washington, D. C., 20433, USA) |
| Abstract: | This paper reassesses F. A.Hayek's "The Use of Knowledge in Society" in light of the rise of artificial intelligence. AI substantially strengthens the case for certain forms of planning by expanding data collection, prediction, codification, monitoring, logistics optimization, and hierarchical coordination. Recent arguments by Acemoglu and by Brynjolfsson and Hitzig suggest that AI may reduce the informational advantage of local actors, enlarge the effcient scale of firms, and revive some forms of algorithmic or AI-assisted planning. The paper argues, however, that AI does not fundamentally invalidate Hayek's knowledge problem. Hayek's central claim was not merely that planners lack computing power, but that economically relevant knowledge is dispersed, tacit, contextual, incentive-laden, forward-looking, and often generated only through decentralized action. AI can process data, but data are not identical to knowledge; AI can predict, but prediction is not discovery; AI can simulate prices, but simulated prices are not market prices generated by property, exchange, competition, profit, and loss The paper further argues that comprehensive AI planning raises serious problems of truthful revelation, metric gaming, surveillance, political power, and objective-function selection. AI may improve planning within firms, platforms, and governments, but such planning works best when embedded in a competitive, price-guided, polycentric market order. The conclusion is not anti-AI but anti-monopoly over social knowledge: AI should augment decentralized discovery rather than replace it with centralized control. |
| Keywords: | Artificial intelligence; Hayek; knowledge problem; socialist calcula tion debate; central planning; market prices; tacit knowledge; dispersed knowledge; entrepreneurship; economic calculation; algorithmic governance; surveillance; firms; platforms; Coase; Mises; polycentric order. |
| JEL: | B25 B31 B53 D46 D80 D83 D85 L14 L16 L22 L23 O31 O33 P11 P16 P21 P51 P52 |
| Date: | 2025–04 |
| URL: | https://d.repec.org/n?u=RePEc:cuf:wpaper:808 |
| By: | Hung Q. Tran |
| Abstract: | Divergent regulatory regimes for data, driven by different motivations, ranging from privacy protection in the European Union to information control in China, could eventually produce distinctively different, and possibly contradictory, bodies of data. Artificial-intelligence models trained on those datasets could produce differing and possibly even conflicting outputs. To the extent that AI outputs start to shape human perception and to influence decisions, in governments and businesses, and among the public, antagonistic AI models would reinforce the mutual mistrust and hostility inherent in the current geopolitical environment, potentially making it harder to resolve conflicts. As a consequence, the fragmentation of data is becoming an important issue in the evolution of AI and its potential impact on human society. |
| Date: | 2026–02 |
| URL: | https://d.repec.org/n?u=RePEc:ocp:pbecon:pb08_26 |