|
on Artificial Intelligence |
| By: | Bergemann, Dirk; Bonatti, Alessandro; Smolin, Alex |
| Abstract: | We develop an economic framework to analyze the optimal pricing and product design of Large Language Models (LLM). Our framework captures several key features of LLMs: variable operational costs of processing input and output tokens; the ability to customize models through fine-tuning; and high-dimensional user heterogeneity in terms of task requirements and error sensitivity. In our model, a monopolistic seller offers multiple versions of LLMs through a menu of products. The optimal pricing structure depends on whether token allocation across tasks is contractible and whether users face scale constraints. Users with similar aggregate value-scale characteristics choose similar levels of fine-tuning and token consumption. The optimal mechanism can be implemented through menus of two-part tariffs, with higher markups for more intensive users. Our results rationalize observed industry practices such as tiered pricing based on model customization and usage levels. |
| Keywords: | Large Language Models |
| JEL: | D47 D82 D83 |
| Date: | 2025–05 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20226 |
| By: | Chengcheng Wang; Zexin Ye |
| Abstract: | As firms increasingly adopt AI-powered pricing algorithms, a key and urgent policy concern is how to regulate the potential algorithmic collusion. This paper approaches the regulatory question through the lens of information design and examines how different disclosure rules, committed to by a third-party intermediary, shape learning outcomes when firms delegate pricing to Q-learning algorithms under stochastic demand. We analyze three disclosure rules: no disclosure, full disclosure, and upper censorship. Upper censorship, which truthfully reveals low-demand states while pooling high-demand ones, delivers higher profits than full disclosure, consistent with theoretical predictions. However, we uncover a profit reversal: when the discount factor is high, no disclosure yields higher profits than full disclosure, whereas when the discount factor is low, full disclosure performs better. This pattern is exactly the opposite of what classical collusion theory predicts. Overall, these findings show that Q-learning agents respond systematically to the information structure and further suggest that restricting information sharing may backfire when algorithms are sufficiently patient, highlighting the need to reassess regulatory approaches in AI-mediated markets. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.04345 |
| By: | Matthew Kovach; Daniel Martin; Gerelt Tserenjigmid |
| Abstract: | We use a controlled experiment to study how beliefs are updated after receiving qualitative information (AI recommendations) from an unknown data-generating process (DGP). Across 60, 252 pairs of prior and posterior beliefs, we document three behavioral patterns: updates close to zero when recommendations confirm extreme priors, larger updates when recommendations contradict extreme priors, and smaller updates for intermediate priors. These three behavioral patterns suggest four testable properties of belief updating, which we assess at the aggregate and individual levels. Finally, we examine how well updates are captured by three models of belief updating. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.10460 |
| By: | Decarolis, Francesco; Pellegrinetti, Tommaso; Rovigatti, Gabriele; Rovigatti, Michele; Shakhgildyan, Ksenia |
| Abstract: | This paper examines how proprietary algorithms used by dominant digital platforms create informational advantages in search auctions, reshaping market competition. Using experimental evidence and counterfactual simulations, we quantify the impact of algorithmic bidding on auction outcomes and competitive dynamics. Our findings reveal how platforms can leverage superior information to significantly improve their revenues, distorting competition and creating welfare losses for independent advertisers. We also show why platforms prefer selling a bidding algorithm service over directly selling data. These results highlight the need for greater scrutiny of algorithmic decision-making in platform markets, offering new insights for competition policy in digital economies. |
| Keywords: | Collusion |
| JEL: | C73 D82 D83 D18 D44 |
| Date: | 2025–02 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:19983 |
| By: | Tomas Havranek (Charles University, Prague & Centre for Economic Policy Research, London & Meta-Research Innovation Center at Stanford); Zuzana Irsova (Charles University, Prague & Centre for Economic Policy Research, London) |
| Abstract: | Probably not, at least for meta-analyses in economics. In a pre-registered, identity-masked, within-paper experiment, the authors of 44 meta-analyses ranked three AI reports on their own paper by usefulness for improving it: a single pass by a frontier model against two multi-agent debate tools we built and expected to win. All reports were held to a common length and template. The authors preferred the single pass, by 0.66 rank points over mad-research (95% CI 0.32 to 1.00) and 0.57 over paper-workshop (0.16 to 0.95), though paper-workshop spent roughly thirty times the tokens. Authors who recalled their journal referee report usually placed it first and never last; in a separate exercise, three AI judges almost always placed the real journal referee report last. Among the three AI reports, Gemini (the judge whose model family wrote none of the reports) would have ranked paper-workshop first in the authors’ place, reversing the single-pass preference. The reversal warns against substituting an AI judge for the author. We measure perceived usefulness for finished papers; whether AI should referee papers is a separate question. |
| Keywords: | LLM-as-a-judge, meta-science, multi-agent debate, pre-registration, test-time compute |
| JEL: | C18 C93 O33 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:fau:wpaper:wp2026_19 |
| By: | Gambacorta, Leonardo; Shreeti, Vatsala |
| Abstract: | The rapid advancement of artificial intelligence (AI) relies on a complex supply chain comprising five key layers: hardware, cloud infrastructure, training data, foundation models and AI applications. This paper examines the market structure of each layer and highlights the economic forces shaping them: rapid technological change, high fixed costs, economies of scale, network effects and, in some cases, strategic behaviour by dominant firms. We also highlight the expanding influence of big tech companies across the AI supply chain. We discuss the challenges for consumer choice, innovation, operational resilience, cyber security and financial stability. |
| Keywords: | Market structure; Competition; Artificial intelligence; Financial stability; Cyber risk; Big tech; Generative AI |
| JEL: | E31 J24 O33 O40 |
| Date: | 2025–04 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20143 |
| By: | Chiara Zisler; Uschi Backes-Gellner |
| Abstract: | Will generative AI (GenAI) displace entry-level IT workers, or change the work they do? The answer depends on whether AI substitutes for or complements the tasks that make up entry-level jobs, a distinction that demand-side employment data cannot resolve because they do not capture task content. We study the intensive margin of entry-level IT work-i.e., how AI shapes the job content conditional on employment-rather than the extensive margin of hiring or job displacement. Using survey data on entry-level IT workers with migration backgrounds, we introduce a worker-side measure of AI-attributed task change: for each task, workers report how their working time has changed and how much of this change they attribute to AI. The measure therefore isolates AI-attributed reallocations of working time across tasks. Within-worker comparisons point more strongly to task complementarity than substitution: Workers spend more time on tasks for which they report AI-related changes, and their task portfolios shift toward a more complex core. These complementarity patterns are strongly associated with AI use, but not with education background or wages, thereby suggesting that the benefits of GenAI in entry-level IT work may depend less on formal credentials than on active engagement with the technology. Our findings provide implications for firms' decisions on job design, task allocation, and AI-related training in IT jobs. |
| Keywords: | Generative AI, task recomposition, entry-level work, substitution or complementarity, job crafting, on-the-job learning |
| JEL: | J23 J24 M51 O33 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:iso:educat:0258 |
| By: | F. Cerina; S. Nobili; M. Rosso |
| Abstract: | We study how the release of ChatGPT affected posted U.S. labor demand, using 368 million Lightcast job postings (2016–2025) and a usage-anchored measure of LLM exposure. While we find a negative and causal effect on the volume of postings in AI-exposed occupations, we find no post-release effect on their seniority composition. After November 2022, posting volume in top-quartile-exposed occupations contracts by 9 log points (approximately 8.6 percent) relative to less-exposed occupations within the same metropolitan market, industry and month. The estimate survives an extensive battery of robustness checks and is concentrated in measured LLM usage rather than AI capability. In exposed occupations, junior postings fall considerably more than senior postings after the release, but the divergence predates ChatGPT - it opens in 2021–22, during the COVID recovery, and shows no break at the release. The differential that a simple pre/post comparison would attribute to ChatGPT is absorbed by predetermined remote-work exposure. Our results caution against reading exposure-based entry-level declines as evidence of AI-driven seniority-biased technological change. |
| Keywords: | Generative AI, ChatGPT, labor demand, seniority, job postings, difference-in-difference |
| JEL: | J23 J24 O33 M51 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:cns:cnscwp:202613 |
| By: | Gabriel Montes-Rojas (IIEP-UBA/CONICET); Fernando Toledo (UNLP); Juan Manuel Rodríguez Repeti (IIEP-UBA) |
| Abstract: | This paper studies what happens when AI gets cheaper, with emphasis on the labor market outcomes, whether it creates formal jobs or whether it pushes workers into informality. We argue that the answer depends on the elasticity of substitution between imported AI capital and formal labor. We build a small open economy DSGE model with a dual labor market, imported AI capital, and country risk, calibrated to an economy where informality is pervasive. The same decline in AI prices produces sharply different labor-market outcomes depending on whether AI substitutes or complements formal workers. Under substitution, cheaper AI weakens formal labor demand and increases the role of the informal sector as an employment buffer. Under complementarity, it expands formal employment and amplifies output, wages, investment, and capital accumulation. The model therefore shows that AI can become either a source of displacement pressure or a driver of formal-sector expansion, depending on how it interacts with human labor. |
| Keywords: | Artificial Intelligence, Informal Economy, Dual Labor Markets, DSGE, Latin America |
| JEL: | E26 F41 O33 J46 C68 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:aoz:wpaper:401 |
| By: | Joshua Brault; Maryam Haghighi; Jing Yang |
| Abstract: | We study the monetary policy response to AI adoption in a two-sector New Keynesian model with a task-based microfoundation, sticky prices, and downward nominal wage rigidity. We distinguish between two forms of AI-driven technological change: augmentation, which raises the productivity of labor within existing tasks, and automation, which displaces labor by reallocating tasks from workers to machines, contracting the set of tasks requiring human input. In the short run, both shocks lower labor demand on net, and with downward nominal wage rigidity, unemployment emerges unless monetary policy provides accommodation. But because monetary policy operates through aggregate demand and cannot target sectors differentially, accommodation that reduces unemployment in the AI-affected sector raises inflationary pressure in the unaffected one, opening a sectoral wedge between the policy rates required to clear the two labor markets. Since, for output-equivalent shocks, automation generates a larger decline in labor demand, the associated wedge is wider and the Phillips curve lies above and to the right of the curve for augmentation---restoring full employment comes at a greater cost of inflation. In addition to the nature of the shock, the aggregate inflationary consequences depend on the breadth of AI adoption across the economy. Under augmentation, as the AI-affected sector grows, its falling sectoral price increasingly offsets the inflation generated elsewhere by monetary accommodation---making aggregate inflation an unreliable signal of the underlying trade-off. For automation both sectoral prices rise and no such offset exists. In our framework, sector-specific AI adoption poses an unambiguous short-run labor market stabilization problem, while its implications for aggregate inflation depend on the nature of technological change, the breadth of adoption, and the response of monetary policy. |
| Keywords: | Monetary policy; Inflation dynamics and pressures; Monetary policy framework and transmission; Structural challenges; Digitalization and productivity |
| JEL: | E E2 E24 E3 E31 E32 E5 E52 J J2 J23 O O3 O33 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:bca:bocawp:26-27 |
| By: | Ferrando, Annalisa; Lamboglia, Sara; Rariga, Judit; Schmidt, Maurice |
| Abstract: | This paper explores the adoption of artificial intelligence (AI) technologies among euro area firms, using harmonised firm-level data from two dedicated modules of the Survey on the Access to Finance of Enterprises (SAFE) conducted in June and December 2025. Based on responses from around 6, 000 firms across 12 euro area countries, the study examines AI adoption rates, drivers, barriers and economic implications. The findings suggest that AI diffusion among euro area firms is progressing rapidly but unevenly, with significant variation across countries and firm characteristics. Approximately 70% of firms report some level of AI use, but only 7% classify their adoption as significant. Adoption is highest in the Netherlands, Finland and Austria, and lowest in Italy and Ireland. Larger and younger firms, particularly in technology-intensive sectors, are leading adopters. Firms identify expected improvements in business processes as the main driver of adoption, while key barriers include skill shortages, data privacy concerns and system incompatibilities. Current AI use and investment are primarily financed through internal funds, complemented by grants and subsidised bank loans. AI adoption is positively associated with firm productivity, turnover growth, fixed investment and own selling price expectations, particularly among intensive users. Survey data show no evidence yet of aggregate labour shedding; instead, AI adoption is positively associated with employment growth. However, firms’ inflation expectations appear largely unaffected by current AI use. JEL Classification: C93, D22, E31, L25, O33 |
| Keywords: | artificial intelligence, firm-level survey data, inflation expectations, productivity |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:ecb:ecbops:2026395 |
| By: | Rachel Yuting Fan; Ha Minh Nguyen |
| Abstract: | How large is the labor cost saved by AI, and how is it distributed across occupations? Using five waves of the Anthropic Economic Index (January 2025 to February 2026), we construct two novel measures from observed AI usage across countries over the world. The AI concentration index (ACI) shows that in developing economies, virtually all AI usage-based value is generated in a small professional enclave (ACI near 1.0); high-income economies average 0.4 to 0.5, with concentration declining in many countries. The labor cost equivalent (LCE) values the time currently saved by AI at $2.7 trillion annually (3.4% of GDP), an indicative measure of the labor cost of that time. Income and regulatory readiness predict concentration; lacking an official English language slows broadening, a barrier for developing countries. |
| Keywords: | AI adoption; AI concentration index; occupational composition; technology diffusion; developing countries |
| Date: | 2026–07–10 |
| URL: | https://d.repec.org/n?u=RePEc:imf:imfwpa:2026/147 |
| By: | D'Alessandro, Francesco; Santarelli, Enrico; Vivarelli, Marco |
| Abstract: | This study examines how regional technological relatedness and local AI knowledge influence regional innovative activity, as measured by patenting activity. Using a novel three-way longitudinal dataset (670 four-digit CPC classes × 302 NUTS-2 regions × nine four-year periods, 1986-2021) and leveraging a deep learning-based identification of AI patents, we show that two broad mechanisms operate in parallel. First, in accordance with the extant literature, technologies that are cognitively close to a region's existing patent portfolio enjoy higher patenting activity, confirming that relatedness remains a strong and persistent predictor of innovative output. Second, local AI endowments are positively associated with patenting across technological fields, even after conditioning on relatedness, indicating that AI plays an enabling and cross-cutting role in a given regional innovation system. Moreover, the interaction between relatedness and AI turns out to be negative and statistically significant, implying that AI attenuates the extent to which local innovative efforts depend on the technology's proximity to the regional portfolio. In sum, AI appears to enhance overall local innovative activity while reducing its reliance on pre-existing regional knowledge structures. |
| Keywords: | Artificial intelligence, AI, technological change, regional innovation, relatedness |
| JEL: | O31 R11 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:glodps:1792 |
| By: | Magnus Lundgren; Jonas Tallberg |
| Abstract: | Artificial intelligence (AI) is rapidly transforming economies, societies, and polities, raising fundamental questions about how it should be regulated. Policymakers face choices over whether to prioritize innovation or safety, rely on public oversight or private self-regulation, and govern nationally or internationally. Yet little is known about how citizens evaluate these competing priorities. Here we report a conjoint survey experiment conducted in seven countries with diverse political and economic profiles. We find that citizens strongly support regulating AI and generally prioritize safety over innovation, public governance over private self-regulation, and international over national approaches. The preference for safety is strongest among those who perceive AI as risky, unpredictable, and personally consequential. These findings reveal a systematic misalignment between dominant regulatory approaches and citizen preferences. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.14585 |
| By: | Berg, K.; Danyu-Zhang, J.; Gaviano, L. G.; Yannelis, C. |
| Abstract: | For most households, human capital is the largest asset they own, and rapid advances in artificial intelligence (AI) may change its value. This paper studies whether workers whose occupations are more exposed to AI use financial and labor markets to hedge this risk, by investing in firms that gain from the new technology. We develop a portfolio-choice model with nontradable human capital in which AI-related equity pays off in states where exposed workers’ labor income falls through technological unemployment. The model predicts that more exposed workers should hold more equity, especially when human capital is large relative to financial wealth. We test these predictions using linked Norwegian administrative data on workers’ occupations, employers, income, wealth, and equity holdings. Workers in more AI-exposed occupations are more likely to participate in equity markets and, conditional on participation, hold more equity, especially from firms located in countries with firms more exposed to the AI boom. The exposure–equity relationship is stronger for younger workers, consistent with life-cycle hedging. Following the release of ChatGPT, workers with greater AI exposure also become more likely to move into lower-exposure industries and senior management roles. Our results highlight a channel through which financial markets may partially insure workers against technological unemployment. |
| Keywords: | Artificial Intelligence, Portfolio Allocation, Income Risk, Stock Market Participation |
| JEL: | G11 G51 J32 |
| Date: | 2026–07–27 |
| URL: | https://d.repec.org/n?u=RePEc:cam:camdae:2660 |
| By: | Gambacorta, Leonardo; Sabatini, Fabiana; Schiaffi, Stefano |
| Abstract: | We study the interaction between banks’ adoption of artificial intelligence (AI) in credit scoring and relationship lending. Using a unique dataset on Italian banks’ investments in AI for the purpose of integrating their credit scoring techniques, matched with credit register data from one year before and one year after the outbreak of the Covid-19 crisis, we find that AI investments help banks mitigate the typical countercyclical effects of relationship lending on firms’ credit supply, as well as on their investment and employment decisions. |
| Keywords: | Artificial intelligence; Machine learning; Credit supply; Relationship lending |
| JEL: | G01 G21 E50 |
| Date: | 2025–03 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:20010 |
| By: | Barbara Livorova; Josef Sveda |
| Abstract: | We show that AI exposure in the Czech labour market is sizeable, but highly uneven. Using occupational AI exposure mapped to worker-level microdata, we find that around one quarter of tasks associated with Czech employees' occupations can be supported, performed, or transformed by AI. Exposure is concentrated in larger urban areas, information-intensive sectors, and higher-wage jobs. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:cnb:rbrief:2026/02 |
| By: | Rashidghalam, Masoomeh; Heshmati-Kim, Jieun; Heshmati, Almas |
| Abstract: | AI is becoming a driving force in Vietnam's economic transformation. The country is moving beyond a growth model based on low-cost labour and export-led industries. Adoptions of the widespread generative AI and AI-powered assistants have accelerated the transformation. AI by reshaping the nature and future of work, it redefines the rules and productivity. This research overviews the recent research investigating how AI is transforming work, demand for digital skills, and productivity gains in AI-adopting Vietnamese industries. Optimal blend of AI and human collaboration influence positively its productivity impacts. Application of AI enable use of its potentials, but it has also significance challenges and risks of skill gaps, training costs, trust, job quality and distribution of its effects. Focus on adaptability, lifelong learning, and integration of AI ensures a positive future of work. This study identifies factors determining adoption of AI and heterogeneity in its productivity and future of work impacts. |
| Keywords: | AI application, Nature of work, Future of work, Economic transformation, Skill requirements, AI productivity impacts, Vietnam |
| JEL: | D24 E24 F63 J24 L52 O33 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:glodps:1796 |