|
on Artificial Intelligence |
| By: | Liu, Jing (University of Maryland); Sweet, Tracy (University of Maryland, College Park); Chen, Marty (University of Maryland, College Park); Engelberg, Jonathan (University of Maryland, College Park); Masters, Megan (University of Maryland, College Park); Clark, Martyn (University of Maryland, College Park); Persaud, Axel (University of Maryland, College Park); Lancaster, Alia (University of Maryland, College Park); Hollingsworth, Jeffrey (University of Maryland, College Park); Rice, Jennifer (University of Maryland, College Park) |
| Abstract: | We evaluate access to a course-integrated generative AI (GenAI) tutor in a randomized controlled trial involving 2, 379 undergraduate students and 30 instructors across multiple disciplines in a large U.S. public university. Among sections of the same course, tutor access reduced final grades by 0.37 standard deviations and learning management system participation by 0.90 standard deviations. Participation declines were similarly large in the full sample, and estimated academic losses were greater among first-generation students. These findings suggest that introducing an AI tutor may displace engagement with existing learning activities without improving academic performance, with potentially differential impact across student groups. |
| Keywords: | generative AI, higher education, AI tutoring, randomized controlled trial, academic performance, student engagement |
| JEL: | I23 I21 C93 O33 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18989 |
| By: | Adi Tarabeih |
| Abstract: | Prospect theory splits risky choice into three parts: loss aversion, measured by the relative weight of losses to gains, $\lambda$ (e.g., $\lambda = 2$ implies that losses weigh twice as much as gains), diminishing sensitivity to the size of outcomes (i.e., the curvature and linearity of the value function), and the nonlinear weighting of probabilities. I examined whether Large Language Models (LLMs) display the same properties as people by meta-analyzing 102 fitted estimates from 11 studies covering 32 language models. The three components emerge differently. Loss aversion is modest: the pooled $\lambda$ is 1.26 (95% CI 1.06 to 1.46), far below the strongest human meta-analytic estimate ($\lambda = 1.96$) and compatible with more conservative ones (1.07 and 1.31). Matched by design, LLMs are compatible with people when gains and losses are symmetric ($\lambda = 1.15$ against 1.07), but well below them when losses are smaller than gains (1.46 against 2.3), where human loss aversion is strongest. Value functions are close to linear rather than concave under the Tversky-Kahneman method (1.01, against a human value of 0.67), but more concave than in people under the Tanaka method (0.52, against 0.67), so diminishing sensitivity depends on how it is measured. Probability weighting shows a human-like inverse S shape ($\gamma = 0.81$ under Tversky-Kahneman and 0.84 under the Tanaka method), though on few studies. In sum, LLMs show little loss aversion and little diminishing sensitivity, but retain human-like probability distortion, yet these generalizations shift when looking at specific model families and fitting methods. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.33103 |
| By: | Yuhang Ji; Fasheng Xu; Tianjun Feng |
| Abstract: | Large language model (LLM) platforms introducing advertising must decide how much capability ad revenue should fund and what consumers should pay to avoid ads. We study a monopolistic platform that jointly chooses costly response quality, advertising intensity, and subscription prices when advertising reduces user value but better service raises advertising returns. Three findings follow. First, when partly ad-supported and ad-free plans coexist, advertising can finance useful service for consumers who receive none under pure subscription. Lower effective quality in middle plans weakens the alternatives available to premium users, allowing higher prices for unchanged ad-free service. Some fully ad-supported users gain utility, while partly ad-supported and ad-free users lose: broader access can coexist with utility polarization. Second, an intermediate advertising cap can maximize consumer surplus by preserving some revenue that finances service while limiting degradation. Third, cheaper compute can raise quality and reduce ad intensity, yet sufficiently large cost reductions can increase aggregate advertising revenue above its initial level because better service raises the value of the remaining advertised interactions. An extension shows how advertising-funded upgrades can bring plans to the same maximum capability and advertising exposure, eliminating the basis for different prices. These results make menu design central to evaluating LLM advertising: the consequences extend beyond users who see ads to the alternatives and prices faced by those who do not. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.32945 |
| By: | Mark L. Egan; Dominic Russel; Claire Shi; Vincent Yao |
| Abstract: | How competitive will the market for generative AI be as it matures? Using novel transaction level credit and debit card data covering 174 million consumers and 114 billion transactions, we study the adoption of paid services and the demand-side foundations of competition in this market. As of April 2026, approximately 2.9% of U.S. consumers pay for at least one major generative AI service, with adoption highest among younger and higher-income consumers. Nearly all of these paid users subscribe to basic plans that cost $30 or less per month. Three patterns suggest limited differentiation and low switching costs. First, although multihoming is common, it has declined as providers’ capabilities have expanded. Second, annual customer retention rates range from 20% to 45% across major AI providers, lower than those for many other subscription products. Third, switching rises around major model releases. Motivated by these findings, we develop and estimate a structural model of AI service demand and supply that reveals strong competitive pressure. Consumers are highly price sensitive, have much weaker preferences for particular providers than for AI overall, and view AI services as increasingly close substitutes. Our results suggest that, absent a substantial increase in AI model quality differentiation, growing substitutability among services will constrain providers’ pricing power and allow consumers to capture a large share of the surplus generated by these services. |
| JEL: | D12 L13 O33 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35953 |
| By: | Eduardo Levy Yeyati; Dar\'io Judzik; Juan Ignacio Barbiero |
| Abstract: | Generative AI changes the scarcity of different forms of human capital. It makes codified knowledge cheaply available at the point of production, while judgment acquired through practice remains tied more closely to the worker. We ask what this does to two familiar prices in the wage structure: the return to schooling and the return to experience. In a short-run framework, machine-supplied codified knowledge substitutes directly for the codified component of worker human capital, lowering the schooling gradient. If codified and practice-based knowledge are complements, the return to experience rises where learning through practice is fastest; technology-specific experience can also become obsolete, rotating the experience profile rather than shifting it uniformly. Using US Current Population Survey earnings data from 2018 through June 2026, we compare occupations in the top and bottom quartiles of the pre-ChatGPT Felten-Raj-Seamans exposure index. After the public release of ChatGPT, the return to a year of schooling falls by 0.55 log points in exposed occupations. The return to a year of experience rises by 0.26 log points at three years of potential experience, is unchanged at fifteen, and falls by 0.12 log points at twenty-five; under the baseline quadratic, the effect crosses zero at 18.1 years. These breaks appear with the release, disappear under a pandemic placebo, survive a common censoring rule across the full sample, and remain when each occupation is allowed its own pre-shock schooling and experience gradients. The pattern is a repricing of human capital inside exposed occupations: codifiable knowledge becomes less scarce, while the value of experience depends increasingly on whether it transfers to the new technology. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.36008 |
| By: | Vahagn Jerbashian; Montserrat Vilalta-Bufi |
| Abstract: | We provide evidence from a 2025 survey of individuals in Spain on workplace use of artificial intelligence (AI) and other modern digital technologies. AI use is already economically meaningful but far from pervasive, with about 16% of workers reporting AI use. Its adoption is highly uneven across workers and jobs, with strong gradients by education, age, earnings, and firm size and scope, as well as substantial within-occupation heterogeneity. AI use is strongly correlated with the use of other modern technologies, which accounts for a larger share of variation than any worker or job characteristic considered separately, including occupational exposure measures. The purposes of AI use also vary markedly across demographic groups and jobs. Users report predominantly positive productivity effects in abstract cognitive tasks, together with limited but systematic time reallocation across tasks. |
| Keywords: | AI at work, complementary technologies, task reallocation, productivity |
| JEL: | J24 O33 L23 C83 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_13094 |
| By: | Joshua S. Gans; Peter Landry |
| Abstract: | Firms ask workers to train AI systems that will replace them. We model this with a worker who can shirk, cooperate with training, or obstruct training while maintaining productivity. Because training ends the job but only cooperation completes training, dismissal becomes a weaker penalty for shirking. The firm must raise the worker’s rent, and the worker gains from their replacement. AI can create jobs paying more than they produce and eliminate jobs it never automates. Firms paying only wages may slow training. A credible completion bonus equal to the worker’s rent without AI removes both the gain and the slowdown. |
| JEL: | D86 J24 J41 M52 O33 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35950 |
| By: | Marco Francesconi; Xiaoyu Xia |
| Abstract: | Artificial intelligence (AI) may transform labor markets not only by changing workers' tasks but also by reorganizing production. We study radiology, where AI has spread rapidly without increasing interpretations per radiologist. In our model, AI facilitates outsourcing by letting hospitals verify the quality of remotely supplied interpretations. Using US Medicare data for 2013-2024, we document declining local radiology activity in smaller markets despite stable imaging demand. Interpretations shift toward larger markets, and radiologists at documented AI-adopting platforms end up serving more markets. AI can thus reshape firm boundaries and the geographic organization of professional work without replacing workers' tasks. |
| Keywords: | artificial intelligence, labor demand, technology adoption, task substitution, firm boundaries, outsourcing, radiology, health care |
| JEL: | J23 L22 L24 O33 I11 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_13086 |
| By: | Manish Jha; Jialin Qian; Michael Weber; Baozhong Yang |
| Abstract: | Using generative AI, we extract managers’ economic expectations from over 120, 000 corporate conference call transcripts. The resulting AI Economy Score predicts GDP growth, production, employment, and wages five or more quarters ahead, beyond survey forecasts and other existing measures. Industry- and firm-level scores capture sector- and firm-specific activity, and a composite of firm, industry, and macroeconomic expectations further improves forecasting power and horizon. Predictability concentrates among historically more accurate, less biased managers, consistent with an information rather than a noise channel. Managerial expectations thus offer incremental insights into economic activity, with implications for macro- and microeconomic decision-making. |
| JEL: | C81 D2 E2 E3 E44 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35949 |
| By: | Fernández, Gastón P. (Luxembourg Institute of Socio-Economic Research (LISER)); Rammer, Christian (Department Economics of Innovation and Industrial Dynamics ZEW - Leibniz Centre for European Economic Research, Mannheim, Germany); Czarnitzki, Dirk (Department of Management, Strategy and Innovation, KU Leuven, Belgium) |
| Abstract: | This paper provides an overview of evidence on firms’ adoption of Artificial Intelligence (AI). We revisit the main conceptual drivers of AI uptake discussed in the literature and summarize global trends in AI diffusion. We then examine the firm characteristics associated with adoption—including the distinction between in-house development and external acquisition—and highlight how patterns differ between general AI and generative AI technologies (GenAI). Next, we review the emerging literature on the productivity effects of AI, distinguishing non-experimental firm-level studies from recent experimental and quasi-experimental evidence on task-level impacts. As a case study, we draw on recent firm-level data from the German Innovation Survey to investigate how adoption decisions correlate with firm size, age, R&D intensity, human capital, location, and strategic orientation, and how these patterns vary across stages and types of AI adoption. |
| Keywords: | Artificial Intelligence, Determinants, Impacts, Literature Survey |
| JEL: | O14 O31 O32 O33 L25 M15 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18991 |
| By: | Alessandro Villa |
| Abstract: | An AI productivity boom can warrant lower nominal rates even when most gains are anticipated. I derive analytical representations of optimal rates under commitment and discretion in an economy with sticky prices and wages and hand-to-mouth households. Holding the realized productivity path fixed, I vary expectations. Anticipated gains raise the natural rate, but below-target inflation supports real wages and keeps nominal rates lower. Surprises leave the natural rate unchanged, calling for cuts. After surprises, expected redistribution toward hand-to-mouth households calls for deeper rate cuts to sustain savers’ spending. Raising rates requires more anticipation when wages are stickier relative to prices. |
| Keywords: | Optimal monetary policy; Sticky-prices; Hand to mouth |
| JEL: | D84 E25 E31 E52 O33 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedhwp:103965 |
| By: | Jin Huang; Diego Ferreras Garrucho; Yutong Xie; Walter M. Yuan; Qiaozhu Mei; Chen Lian; Jonathon Hazell |
| Abstract: | Large language models (LLMs) have the potential to meet a key goal in economics: a quantitative model of household decision making, across a variety of settings. Yet existing evaluations cover few surveys and outcomes, and do not study how households adjust to changing economic conditions. We introduce a new evaluation, HouseholdBench, which unites 6 U.S. household surveys and 32 prediction tasks spanning numeric, categorical and probabilistic outcomes, related to consumption, income, labor, expectations, and housing. Using past behavior, demographics and macroeconomic conditions, the tasks test whether LLMs predict behavior, including how households adjust to changes in various policies. We evaluate 13 proprietary and open-weight LLMs against a no-change baseline and a gradient boosted tree model. Most LLMs outperform the no-change baseline, including for policy response tasks—with the best model lowering error for numeric outcomes by 12.2%. Across most tasks, gradient-boosted trees rank first; leading proprietary LLMs approach their performance, but open-weight models lag. LLMs exhibit systematic over- and underprediction across different tasks. We identify methods that enable a 4 billion parameter open-weight model to match proprietary models’ performance: fine-tuning and aggregating 16 predictions per observation. Improvements generalize to policy-response tasks, which are excluded from fine-tuning. We release our datasets, code, and leaderboard on our website. |
| JEL: | C45 C53 D12 E21 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35927 |
| By: | Bhagwat, Vineet; Cookson, J. Anthony; Dim, Chukwuma Chijioke; Niessner, Marina |
| Abstract: | Can large language models (LLMs) reveal how investors disagree? We endow an LLM with 216 representative investor personas, elicit buy/hold/sell responses to 1.25 million S&P 500 news headlines, and measure disagreement as within-headline dispersion of responses. Disagreement is highest for socially oriented news and lowest for fundamentals, and predicts elevated same- and next-day abnormal trading volume, even after the LLM's training cutoff. Disagreement rooted in income, education, and age is most strongly related to trading, whereas political disagreement, though largest, is least related. Persona responses reflect non-pecuniary rationales, align with human-survey benchmarks, and are stable across prompt variations. |
| Keywords: | Investor Disagreement, LLMs, Beliefs, Sentiment, Narratives, Trading |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:cbscwp:344253 |
| By: | Buchetti, Bruno; Lin, Luca X.; Miquel-Flores, Ixart; Eufinger, Christian |
| Abstract: | General-purpose technologies create large opportunities but make their distribution across firms difficult to assess in real time. We use banks as interpreters of the rise of generative AI by studying lender-specific probability-of-default (PD) estimates for private firms in four large euro-area economies. As public attention to AI rises after ChatGPT, banks lower assessed default risk for borrowers in highly AI-exposed industries by about 5%. The reassessment is weaker for labor-intensive firms and stronger among lenders with pre-existing industry expertise. Yet the favorable industry-level effect is accompanied by a systematic reordering within product markets: banks reassess firms with stronger pre-ChatGPT market positions less favorably relative to their competitors, compressing the risk advantage associated with market leadership. Credit allocation moves in the same direction, with market leaders experiencing relatively lower drawn and committed credit, more relationship exits, and fewer new lenders. The reordering also predicts subsequent bad-credit events beyond banks’ consensus PDs. Internal bank assessments therefore provide a forward-looking measure of how generative AI is expected to reshape competitive risk. JEL Classification: D82, D83, G21, G32, O33 |
| Keywords: | artificial intelligence, bank lending, credit risk, generative AI, industry specialization, probability of default, relationship lending, technological change |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:ecb:ecbwps:20263302 |
| By: | Giorgos Iacovides; Oliver Hannibal; Wuyang Zhou; Danilo Mandic |
| Abstract: | Recent advances in Generative AI have substantially improved financial sentiment analysis through post-trained financial large language models (LLMs). However, existing approaches remain confined to predicting the sentiment of each article in isolation and operate as black boxes, which limits contextual understanding and hinders deployment in high-stakes financial decision-making. To this end, we introduce FinLLM-KG, a model-agnostic framework that combines causal LLMs with a financial knowledge graph, in order to inject temporally grounded, company-level knowledge from SEC 10-K filings at inference time, without updating model weights. Within FinLLM-KG, context is supplied either as a natural-language company profile or as a small set of relevance-ranked factual triples. This enables sentiment predictions to account for broader firm-level information rather than the article alone, at almost zero computational cost and for any causal LLM. In addition, attributing each article-level sentiment score to individual sentences and to the retrieved knowledge-graph facts makes the evidence behind the prediction inspectable, thereby enhancing explainability and supporting human-in-the-loop review. Simulations demonstrate that FinLLM-KG consistently improves portfolio performance over the state-of-the-art baseline. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.36083 |
| By: | Javier Ojea Ferreiro; Andrea Ugolini; Juan C. Reboredo |
| Abstract: | Artificial intelligence (AI) has a double-edged relationship with financial stability: it may improve efficiency and risk management, while also creating new vulnerabilities and amplification channels. This paper studies whether market-perceived exposure to AI is reflected in the tails of financial-sector returns, and whether this relationship depends on macroeconomic conditions. We develop a framework that decomposes financial-sector tail outcomes across contemporaneous AI and macroeconomic scenarios, and identifies whether the AI-related episodes are primarily associated with extreme gains or losses. Using daily equity returns for financial sectors in Canada, the United States, Europe, the United Kingdom, and China, we document substantial heterogeneity. Financial-sector-specific scenarios dominate tail outcomes across markets, while the contribution of AI-related scenarios to return asymmetry differs substantially across countries. AI-sector scenarios contribute to downside asymmetry in Europe, China, and especially the United Kingdom, but make only a limited contribution in Canada and have the smallest tail weights in the United States. China is the only market exhibiting positive average skewness, driven by the financial and systemic scenarios rather than by AI-related outcomes. Overall, AI-related market movements are neither uniformly stabilizing nor destabilizing; their financial-stability interpretation depends on the market considered and on whether AI-sector tail events occur in isolation or jointly with broader macroeconomic movements. |
| Keywords: | Financial system; Financial institutions and intermediation; Financial stability and systemic risk; Models and tools; Econometric, statistical and computational methods; Structural challenges; Digitalization and productivity |
| JEL: | C C5 C58 G G0 G01 G2 G21 G3 G32 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:bca:bocawp:26-40 |