|
on Artificial Intelligence |
| By: | Leonardo Bursztyn; Alex Imas; Rafael Jiménez-Durán; Aaron Leonard; Christopher Roth |
| Abstract: | Anxiety about falling behind can drive people to embrace emerging technologies with uncertain consequences. We study how social forces shape demand for AI-based learning tools early in the education pipeline. In incentivized experiments with parents—key gatekeepers for children’s AI adoption—we elicit their demand for unrestricted AI tools for teenagers’ education. Parental demand rises with the share of other teenagers using the technology, with social forces increasing willingness to pay for AI by more than 60%. Providing information about potentially adverse effects of unstructured AI use negatively shifts beliefs about the merits of AI, but does not change individual demand. Instead, this information increases parents’ preference for banning AI in schools. Follow-up experiments show that social information has little effect on beliefs about AI quality, perceived skill priorities, or support for bans, suggesting that effects operate through social pressure rather than social learning. Our evidence highlights social pressure driving individual technology adoption despite widespread support for restricting its use. |
| JEL: | D83 D91 I20 O33 |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:34488 |
| By: | Priyanka Verma; Balagopal Unnikrishnan |
| Abstract: | Fair resource division algorithms, like those implemented in Spliddit platform, have traditionally been considered difficult for the end users to manipulate due to its complexities. This paper demonstrates how Large Language Models (LLMs) can dismantle these protective barriers by democratizing access to strategic expertise. Through empirical analysis of rent division scenarios on Spliddit algorithms, we show that users can obtain actionable manipulation strategies via simple conversational queries to AI assistants. We present four distinct manipulation scenarios: exclusionary collusion where majorities exploit minorities, defensive counterstrategies that backfire, benevolent subsidization of specific participants, and cost minimization coalitions. Our experiments reveal that LLMs can explain algorithmic mechanics, identify profitable deviations, and generate specific numerical inputs for coordinated preference misreporting--capabilities previously requiring deep technical knowledge. These findings extend algorithmic collective action theory from classification contexts to resource allocation scenarios, where coordinated preference manipulation replaces feature manipulation. The implications reach beyond rent division to any domain using algorithmic fairness mechanisms for resource division. While AI-enabled manipulation poses risks to system integrity, it also creates opportunities for preferential treatment of equity deserving groups. We argue that effective responses must combine algorithmic robustness, participatory design, and equitable access to AI capabilities, acknowledging that strategic sophistication is no longer a scarce resource. |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2511.14722 |
| By: | Brodeur, Abel (University of Ottawa); Cook, Nikolai (Wilfrid Laurier University); Valenta, David (University of Ottawa) |
| Abstract: | Artificial intelligence (AI) technologies are increasingly viewed as both a potential driver of environmental sustainability and a contributor to global energy demand. Yet little is known about how the public interprets these dual narratives. We conducted a pre-registered online experiment (N = 2142) on a representative sample of the United States to examine how framing the environmental impacts of AI—as positive or negative—affects beliefs, policy preferences, and behavioral intentions. Positive messaging led to greater optimism about AI’s environmental impact, lower support for regulation, increased support for government subsidies of AI-enabled technology adoption, and increased consumer preferences for AI-enabled appliances. Negative messaging increased support for regulation and decreased support for government subsidies. Consistent with previous evidence, the messenger (scientist vs journalist) had minimal impact. Our findings highlight the power of environmental framing in shaping public narratives around AI, with implications for science communication, sustainability governance, and technology acceptance. |
| Keywords: | online experiment, energy use, Artificial Intelligence, energy conservation, behavior |
| JEL: | O3 Q4 Q5 |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18263 |
| By: | Luca Corazzini; Elisa Deriu; Marco Guerzoni |
| Abstract: | Large language models (LLMs) increasingly mediate economic and organisational processes, from automated customer support and recruitment to investment advice and policy analysis. These systems are often assumed to embody rational decision making free from human error; yet they are trained on human language corpora that may embed cognitive and social biases. This study investigates whether advanced LLMs behave as rational agents or whether they reproduce human behavioural tendencies when faced with classic decision problems. Using two canonical experiments in behavioural economics, the ultimatum game and a gambling game, we elicit decisions from two state of the art models, Google Gemma7B and Qwen, under neutral and gender conditioned prompts. We estimate parameters of inequity aversion and loss-aversion and compare them with human benchmarks. The models display attenuated but persistent deviations from rationality, including moderate fairness concerns, mild loss aversion, and subtle gender conditioned differences. |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2511.12319 |
| By: | Foucart, Renaud; Zeng, Fanqi; Wang, Shidong |
| Abstract: | In the age of Artificial Intelligence (AI), we have access to high-quality advice from intelligent machines, creating tensions between efficiency and autonomy. While it is often individually beneficial to follow AI recommendations, it can lead to dangerous herding behavior when AI provides incorrect information, a phenomenon observed in corporate scandals such as biased hiring algorithms, financial automation, or the monitoring of employees. In this paper, we show how a small group of stubborn workers following their own, imperfect but independent, information may help mitigate the risks of AI advice in their organization. However, stubbornness is costly---such individuals persist only if subsidized in the recruitment and promotion system despite lower average performance, or if they have a strong intrinsic preference for autonomy. |
| Date: | 2025–11–11 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:nfgy3_v1 |
| By: | Gonzalez Vazquez Ignacio (European Commission - JRC); Fernandez Macias Enrique (European Commission - JRC); Wright Sally; Villani Davide (European Commission - JRC) |
| Abstract: | This report presents new evidence on the platformisation of work in the European Union, examining the prevalence and potential impacts of digital tools, digital monitoring and algorithmic management. The report is based on data from the new AIM-WORK survey, conducted in 2024-2025 and representative of the working age population in all 27 EU Member States. The data reveals that over 90% of EU workers use digital devices, with the use of AI tools at work, particularly AI chatbots powered by Large Language Models, rising rapidly: on average, a third of EU workers report using AI for work-related purposes. Digital monitoring is common, particularly for working hours and entry or exit. Algorithmic management is less prevalent but also quite significant, taking diverse forms, including automated task allocation and performance evaluation. We identify two distinct types of platformisation, typical respectively of industrial and office workplaces. Our evidence indicates that some types of platformisation have no significant implications for working conditions. However, the full platformisation of work, which includes simultaneously all the forms of digital monitoring and algorithmic management that we identify on the basis of the data, is associated with generally worse working conditions. This applies also to the forms of platformisation more prevalent in manual work settings. |
| Date: | 2025–10 |
| URL: | https://d.repec.org/n?u=RePEc:ipt:iptwpa:jrc143072 |
| By: | Hellsten, Mark (University of Tübingen); Khanna, Shantanu (Northeastern University); Lodefalk, Magnus (Örebro University); Yakymovych, Yaroslav (Uppsala University) |
| Abstract: | Artificial intelligence (AI) is expected to reshape labor markets, yet causal evidence remains scarce. We exploit a novel Swedish subsidy program that encouraged small and mid-sized firms to adopt AI. Using a synthetic difference-in-differences design comparing awarded and non-awarded firms, we find that AI subsidies led to a sustained increase in job postings over five years, but with no statistically detectable change in employment. This pattern reflects hiring signals concentrated in AI occupations and white-collar roles. Our findings align with task-based models of automation, in which AI adoption reconfigures work and spurs demand for new skills, but hiring frictions and the need for complementary investments delay workforce expansion. |
| Keywords: | hiring, labor markets, Artificial Intelligence, task content, technological change |
| JEL: | J23 J24 O33 |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18267 |
| By: | Storm, Eduard (Institute for Advanced Studies (IHS) and RWI – Leibniz Institute for Economic Research); Gonschor, Myrielle (Kienbaum Consultants); Schmidt, Marc Justin (TU Dortmund, RTG 2484) |
| Abstract: | We study how artificial intelligence (AI) affects workers’ earnings and employment stability, combining German job vacancy data with administrative records from 2017–2023. Identification comes from changes in workers’ exposure to local AI skill demand over time, instrumented with national demand trends. We find no meaningful displacement or productivity effects on average, but notable skill heterogeneity: expert workers with deep domain knowledge gain while non-experts often lose, with returns shaped by occupational task structures. We also document AI-driven reinstatement effects toward analytic and interactive tasks that raise earnings. Overall, our results imply distributional concerns but also job-augmenting potential of early AI technologies. |
| Keywords: | AI, Online Job Vacancies, Skill Demand, Worker-level Analysis, Employment, Earnings, Expertise |
| JEL: | D22 J23 J24 J31 O33 |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:ihs:ihswps:number61 |
| By: | Voraprapa Nakavachara |
| Abstract: | This paper investigates the relationship between AI use and worker well-being outcomes such as mental health, job enjoyment, and physical health and safety, using microdata from the OECD AI Surveys across seven countries. The results reveal that AI users are significantly more likely to report improvements across all three outcomes, with effects ranging from 8.9% to 21.3%. However, these benefits vary by generation and gender. Generation Y (1981-1996) shows the strongest gains across all dimensions, while Generation X (1965-1980) reports moderate improvements in mental health and job enjoyment. In contrast, Generation Z (1997-2012) benefits only in job enjoyment. As digital natives already familiar with technology, Gen Z workers may not receive additional gains in mental or physical health from AI, though they still experience increased enjoyment from using it. Baby Boomers (born before 1965) experience limited benefits, as they may not find these tools as engaging or useful. Women report stronger mental health gains, whereas men report greater improvements in physical health. These findings suggest that AI's workplace impact is uneven and shaped by demographic factors, career stage, and the nature of workers' roles. |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2511.11021 |
| By: | Soogand Alavi; Salar Nozari; Andrea Luangrath |
| Abstract: | Recent advances in large language models (LLMs) have dramatically improved performance on a wide range of tasks, driving rapid enterprise adoption. Yet, the cost of adopting these AI services is understudied. Unlike traditional software licensing in which costs are predictable before usage, commercial LLM services charge per token of input text in addition to generated output tokens. Crucially, while firms can control the input, they have limited control over output tokens, which are effectively set by generation dynamics outside of business control. This research shows that subtle shifts in linguistic style can systematically alter the number of output tokens without impacting response quality. Using an experiment with OpenAI's API, this study reveals that non-polite prompts significantly increase output tokens leading to higher enterprise costs and additional revenue for OpenAI. Politeness is merely one instance of a broader phenomenon in which linguistic structure can drive unpredictable cost variation. For enterprises integrating LLM into applications, this unpredictability complicates budgeting and undermines transparency in business-to-business contexts. By demonstrating how end-user behavior links to enterprise costs through output token counts, this work highlights the opacity of current pricing models and calls for new approaches to ensure predictable and transparent adoption of LLM services. |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2511.11761 |
| By: | Marius Müller; Carsten Lausberg |
| Abstract: | Artificial Intelligence (AI) is becoming an integral part of everyday processes in the real estate industry. In real estate valuation, AI has long been used for automated valuations, but so far rarely for manual valuations. One reason for the reluctant adoption is the complex and know-ledge-intensive process that requires the careful evaluation of numerous factors. This paper investigates for condominiums whether AI can improve manual real estate valuations by redu-cing time and enhancing accuracy. To address this question, we first provide a comprehensive review of the current literature on AI applications in real estate valuation and discuss the po-tential advantages and drawbacks of integrating AI into valuation practices. Then we present the results of an experiment in which the 28 participants were asked to appraise an apartment using either an AI-supported tool or a conventional Excel-based tool. Performance indicators show that the AI tool significantly reduces time and inter-valuer variability, while valuation ac-curacy is largely unaffected. The insights gained from this analysis contribute to understanding the practical applicability of AI in real estate valuation and highlight opportunities for further research and industry adoption. |
| Keywords: | Appraisal; Artificial Intelligence; Real Estate Valuation; valuation accuracy |
| JEL: | R3 |
| Date: | 2025–01–01 |
| URL: | https://d.repec.org/n?u=RePEc:arz:wpaper:eres2025_216 |
| By: | Arenal, Alberto; Aguado, Juan Miguel; Armuña, Cristina; Ramos, Sergio; Feijóo, Claudio |
| Abstract: | Artificial Intelligence (AI) plays a pivotal role throughout every stage of the music industry value chain. In fact, AI has been integral to the value proposition of music streaming platforms since their inception. This research article investigates how AI—and in particular, Generative AI (GenAI)—affects the creation and consumption stages, the two most socially significant components of the music streaming value chain. It explores users' perceptions of the impact of AI/GenAI on their music-listening experience and examines how artists and performers view the role and influence of AI on their position and opportunities within the streaming model. Drawing on the findings from two separate focus groups with users and artists / performers from different geographies and professional development, this study complements industry debates—often dominated by technology companies and record and publishing firms—by providing valuable insights into the perceptions at both ends of the music industry value chain regarding the impact of these technologies on music creation, dissemination, and consumption. As key findings, while users exhibited a nuanced response to AI-generated music, both existing literature and insights from artists and performers suggest that AI may further amplify the endemic dysfunctions of music streaming platforms, arising governance issues and ethical concerns, particularly regarding to transparency. In addition, both groups highlighted a significant paradox: while AI has the potential to democratise music creation by lowering barriers to entry, it also poses a threat to the existing ecosystem of music professionals, which have relevant implications in terms of the role of culture in societies, policy and practice. |
| Keywords: | Artificial Intelligence, Generative Artificial Intelligence, digital music value chain, music performers, streaming platforms, consumption of music |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:itse25:331251 |
| By: | Hildebrandt, Mireille |
| Abstract: | In this contribution, based on my Keynote at the Opening Night of CPDP2025, I briefly discuss how law institutes economic markets, how national law depends on international law and where geo-economics matters. I will then investigate how AI infrastructure’s need for massive amounts of electricity, water and rare earth minerals is transforming the playing field of international law, noting that the consolidation of Big Tech corporations has already impacted the political economy of law. Finally, I will discuss the ideology of artificial general intelligence (AGI), and propose that the European approach to legislating, governing and judging artificial intelligence (AI) as ‘a normal technology’ is a viable though contested alternative that must be consolidated, defended and further developed to survive geopolitical power grabs by the US, Russia and China. |
| Date: | 2025–11–08 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:n95vd_v1 |
| By: | Purificato Erasmo (European Commission - JRC); Bili Danai (European Commission - JRC); Jungnickel Robert (European Commission - JRC); Ruiz Serra Victoria (European Commission - JRC); Fabiani Josefina (European Commission - JRC); Abendroth Dias Kulani (European Commission - JRC); Fernandez Llorca David (European Commission - JRC); Gomez Emilia (European Commission - JRC) |
| Abstract: | Artificial Intelligence (AI) is fundamentally transforming the scientific process across all stages, from hypothesis generation and experimental design to data analysis, peer review and dissemination of results. In many research fields, such as the examined protein structure prediction, materials discovery and computational humanities, AI accelerates discovery, fosters interdisciplinary collaboration and enhances reproducibility, while improving access to advanced analytical and computational capabilities. These developments align with the European Union (EU)’s vision to make AI tools and infrastructure more accessible, strengthening research in areas of strategic importance such as climate change, health, and clean technologies. However, this progress introduces new challenges, including concerns about algorithmic bias, the proliferation of hallucinations and fabricated data, and the potential erosion of critical thinking skills. AI Adoption remains uneven across scientific domains, and addressing these risks requires robust governance, transparency and alignment with open-science principles. This report, accompanying the publication of the European Strategy for AI in Science, provide scientific evidence to support policymakers in maximising AI’s benefits for EU’s research excellence, innovation and competitiveness, while ensuring its deployment in science remains ethical, inclusive and aligned with European values. |
| Date: | 2025–10 |
| URL: | https://d.repec.org/n?u=RePEc:ipt:iptwpa:jrc143482 |
| By: | Iñaki Aldasoro; Ajit Desai |
| Abstract: | Using prompt-based experiments with ChatGPT’s reasoning model, we evaluate whether a generative artificial intelligence (AI) agent can perform high-level intraday liquidity management in a wholesale payment system. We simulate payment scenarios with liquidity shocks and competing priorities to test the agent’s ability to maintain precautionary liquidity buffers, dynamically prioritize payments under tight constraints, and optimize the trade-off between settlement speed and liquidity usage. Our results show that even without domain-specific training, the AI agent closely replicates key prudential cash-management practices, issuing calibrated recommendations that preserve liquidity while minimizing delays. These findings suggest that routine cash-management tasks could be automated using general-purpose large language models, potentially reducing operational costs and improving intraday liquidity efficiency. We conclude with a discussion of the regulatory and policy safeguards that central banks and supervisors may need to consider in an era of AI-driven payment operations. |
| Keywords: | Digital currencies and fintech; Financial institutions; Financial services; Financial system regulation and policies; Payment clearing and settlement systems |
| JEL: | A12 C7 D83 E42 E58 |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:bca:bocawp:25-35 |
| By: | Julian Lütticke; Lukas Lautenschlaeger; Wolfgang Schäfers |
| Abstract: | The role of investor sentiment in real estate investment trust (REIT) markets is well-documented. However, traditional sentiment indicators often fail to fully capture real-time market dynamics. This study explores the potential of GenAI-extracted sentiment in forecasting U.S. REIT returns by leveraging large language models (LLMs) to analyze textual data from news media sources. The hypothesis underpinning this study is that LLMs can process textual data in a manner analogous to that of humans. The novel sentiment score is integrated into a machine learning model to predict REIT returns. The analysis differentiates between overall index returns and sector-specific REIT performance, thereby providing a more granular view of sentiment-driven market behavior. In addition to traditional statistical metrics the model performance is assessed by evaluating an active trading strategy based on sentiment signals. This strategy is benchmarked against a buy-and-hold approach to determine whether sentiment-based predictions can systematically outperform the market. The findings contribute to the growing field of AI-driven financial forecasting and offer valuable insights for investors and policymakers in the indirect real estate sector. |
| Keywords: | Generative AI; Large Language Model; News Sentiment; REIT |
| JEL: | R3 |
| Date: | 2025–01–01 |
| URL: | https://d.repec.org/n?u=RePEc:arz:wpaper:eres2025_242 |
| By: | Daniel Moulton; Larry Santucci; Robyn Smith |
| Abstract: | This paper examines methodological approaches for extracting structured data from large-scale historical document archives, comparing “hyperspecialized” versus “adaptive modular” strategies. Using 56 years of Philadelphia property deeds as a case study, we show the benefits of the adaptive modular approach leveraging optical character recognition (OCR), full-text search, and frontier large language models (LLMs) to identify deeds containing specific restrictive use language— achieving 98% precision and 90–98% recall. Our adaptive modular methodology enables analysis of historically important economic phenomena including re strictive property covenants, their precise geographic locations, and the localized neighborhood effects of these restrictions. This approach should be easily adapt able to other research involving deeds and similar document. |
| Keywords: | large language models (LLMs); artificial intelligence (AI); machine learning (ML); restrictive covenants; deeds; property; real estate; housing; John Coltrane; digitization |
| JEL: | C81 N32 R31 R38 |
| Date: | 2025–10–25 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedpdp:102114 |
| By: | Fang, Qixiang; Garcia-Bernardo, Javier; van Kesteren, Erik-Jan |
| Abstract: | Large language models (LLMs) have become an essential tool for social scientists and humanities (SSH) researchers who work with textual data. One particularly valuable use case is automating text annotation, traditionally a time-consuming step in preparing data for empirical analysis. Yet, many SSH researchers face two challenges: getting started with LLMs, and understanding how to evaluate and correct for their limitations. The rapid pace of model development can make LLMs appear inaccessible or intimidating, while even experienced users may overlook how annotation errors can bias results from downstream analyses (e.g., regression estimates, $p$-values), even when accuracy appears high. This tutorial provides a step-by-step, hands-on guide to using LLMs for text annotation in SSH research for both Python and R users. We cover (1) how to choose and access LLM APIs, (2) how to design and run annotation tasks programmatically, (3) how to evaluate annotation quality and iterate on prompts, (4) how to integrate annotations into statistical workflows while accounting for uncertainty, and (5) how to manage cost, efficiency, and reproducibility. Throughout, we provide concrete examples, code snippets, and best-practice checklists to help researchers confidently and transparently incorporate LLM-based annotation into their workflows. |
| Date: | 2025–11–13 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:v4eq6_v1 |
| By: | White, Jordan |
| Abstract: | A range of systems mapping approaches are widely used to support the analysis and design of public policy, but can be time and resource intensive to implement. Generative AI tools may be able to streamline the use of systems mapping by helping researchers to quickly synthesise existing data on policy systems, freeing resources to foster greater stakeholder participation and use of maps. To explore and test the potential of these tools to help with systems mapping exercises, we examine the performance of seven proprietary vision language models (VLMs) with a key task in potential workflows - extraction of relevant information from images of system maps already created. VLMs present value as they allow for the synthesis of both textual and image data simultaneously. We test on images of three types of system map diagrams: Causal Loop Diagrams, Fuzzy Cognitive Maps and the Theory of Change maps, and test three different formats for structuring data: DOT, JSON and Markdown table. We find that models summarise factors in maps better than connections, with some models extracting factor labels perfectly for certain images and formats. Models appear to perform better with diagrams that have bolder graphics and when there is greater internal consistency between separate node and edge lists. We also find that models appear to omit correct information more than they include false information, although falsehoods are still common. Our formal approach to testing introduces an empirical framework that will allow researchers to conduct similar research in the future, to maintain pace as the application and capabilities of language models continue to evolve. |
| Keywords: | Systems mapping, generative AI, vision language models, policy research, causal graphs |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:amz:wpaper:2025-26 |