|
on Artificial Intelligence |
| By: | Fulvio Castellacci; Tommaso Ciarli; Yuan Gao; Marianna Marino; Giacomo Marzi; Massimo Riccaboni; Maria Savona; Simone Vannuccini |
| Abstract: | This paper examines the tension between the benefits of generative artificial intelligence (AI) for scientific research and the unresolved governance questions that accompany its rapid adoption. Drawing on an academic roundtable held at the AI for Science and Innovation Workshop (Scuola IMT Alti Studi Lucca, April 2026) and on a fast-expanding empirical literature, it maps the disagreement within the research community across four stages of the research process: funding, research tasks, publication and peer review, and use and uptake. The empirical case for AI's productivity, augmentation, and democratization effects has strengthened. The picture changes once productivity is disaggregated: AI-assisted work shows measurable gains in publication volume and citation share, while the evidence on novelty, disruption, and breakthrough output remains ambiguous or negative. We argue that the divergence between private and social returns arises through three analytically distinct mechanisms, namely information asymmetry, negative externalities on a shared knowledge base, and depletion of research capacity, and that each calls for a different governance instrument. We propose Responsible Research with AI (RRAI), an extension of the Responsible Research and Innovation tradition organized around four principles that operate at different levels of the research system: disclosure, differentiation, narrative, and proportionality. RRAI builds on existing institutional scaffolding, including the EU AI Act, UNESCO, and the OECD, and aims to preserve AI's productivity gains while addressing systemic risks that individual researchers can neither observe nor manage on their own. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.24879 |
| By: | Lall, Ranjit (University of Oxford) |
| Abstract: | While advances in artificial intelligence (AI) are feared to bring about widespread job losses, workers highly exposed to the technology tend to anticipate productivity-driven improvements in their earnings and employment prospects. I argue that this paradox has important political economy implications: if labor complementarities are expected to outweigh substitution effects—raising income without commensurately intensifying employment risks—exposure to AI should weaken rather than strengthen support for redistributive policies and their political advocates. I test this argument using a combination of observational and original experimental data from Western Europe, finding that occupation-level AI exposure is negatively associated with support for redistribution, the left, and the (increasingly pro-welfare) populist right but positively associated with support for the mainstream right. The results enhance our understanding of the political economy of digitalization, suggesting a discrepancy between the perceived distributional consequences of AI and earlier automation technologies that have primarily displaced labor. |
| Date: | 2026–07–21 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:vmgdj_v1 |
| By: | Yue Ma; Tianli Feng; Robert W. Fairlie; Chengfang Liu; Prashant Loyalka; Scott Rozelle; Xinwu Zhang |
| Abstract: | The emergence of artificial intelligence (AI) has heightened interest in personalizing computer assisted learning (CAL) programs to tailor their instruction to individual students. Despite the proliferation of AI-driven CAL programs, evidence for their effectiveness remains limited. We present findings from a large-scale field experiment in rural China examining whether an AI-driven personalized CAL program improves student achievement. We randomly assign 8, 647 students from 315 primary school classes to one of three treatment arms: (i) AI-CAL, (ii) non-personalized CAL (active control), and (iii) non-CAL educational activities (pure control). Results indicate that AI-CAL does not significantly improve achievement, with estimates precise enough to rule out non-trivial positive effects. This finding holds across the difficulty of assessment items and across the baseline achievement of students. The finding that R-CAL modestly benefits students in the middle ability tercile while AI-CAL shows no impact on students in any ability tercile suggests caution when projecting the promise of scaling adaptive AI-driven educational technologies in under-resourced settings. |
| Keywords: | AI, education technology, computer assisted learning, adaptive learning, ICT, RCT, rural China |
| JEL: | I21 O15 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_12837 |
| By: | James M. Zumel Dumlao; Meng Wang; Zhonghan Xie; Junyao Hu; Ivan Bar; George Chaney III; Henry Gold; Misha Teplitskiy |
| Abstract: | The spread of generative AI (GenAI) in higher education has raised concerns that students offload cognitive effort to AI, earning high grades without learning. If this "GenAI substitution hypothesis" is true, grades should rise disproportionately in GenAI-susceptible courses--those relying more on assessments like take-home problem sets and essays rather than in-class exams. Substitution could also affect student satisfaction, measured here as self-reported understanding and interest in the subject, which prior research links to assessments. We test the substitution hypothesis using syllabus and administrative data from a large U.S. university (2015-2025; 156, 135 students; 87, 936 course offerings). We measure courses' GenAI susceptibility using a human-validated LLM pipeline to extract assessment types from syllabi, and use a differences-in-differences design comparing outcomes across courses before and after ChatGPT's release, while modeling COVID-19 pandemic effects as either persistent or transient. We find no significant differential effect of GenAI availability on grades overall or among previously lower-performing students. Effects on self-reported understanding are likewise insignificant; effects on interest are significant only assuming transient pandemic effects. Our findings temper concerns that GenAI inflates grades and reduces students' satisfaction. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.21534 |
| By: | Yuyu Chen; Hongbin Li; Lingsheng Meng; Xinyao Qiu; Qingxu Yang |
| Abstract: | Generative AI is fast becoming the first place people turn for expert advice. The advice it provides can be directional rather than neutral, shaped in part by the choices of its designers and regulators. When clients consult AI before meeting an expert, they carry this directional advice into a relationship that once rested on the expert's judgment alone. We study its consequences in healthcare through a large-scale preregistered field experiment at a Chinese hospital, where we randomize patients' access to an AI chatbot before their outpatient visit. Examination of the conversation logs shows that the chatbot routinely cautions against the use of medications, especially Traditional Chinese Medicine and antibiotics, while issuing clean recommendations for diagnostic testing, consistent with the liability-driven guardrails encoded in AI training. This directionality propagates into clinical practice. Prescription rates decline among treated patients while diagnostic testing increases, and these effects are more pronounced among physicians who are receptive to patient input and those with more intensive prescribing styles. Beyond shifting healthcare utilization, survey results show that AI access reduces patient compliance and satisfaction, shifting the balance of authority between patients and physicians. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.08706 |
| By: | Friberg, Richard |
| Abstract: | This paper provides an overview of the theoretical and empirical literatures on the effect of algorithm use on prices, focusing on the type of algorithms relevant for gasoline markets. Against this background we examine pricing and algorithm use on the Swedish gasoline market 2021-2023, relying on detailed information of what algorithms that are used by what station at what time. Only a handful of stations are using AI. Pricing at these AI stations change markedly when AI is adopted, resulting in many more price changes. On average margins are somewhat lower with AI but AI stations charge relatively higher prices during the afternoon peak in demand. In contrast use of rule-based pricing algorithms is pervasive and three out of the four major chains use rule-based algorithms from external algorithm providers. Examining of duopoly markets (stations with only one competitor within a 10-minute drive) suggest that algorithms are faster to respond to price decreases than manual pricing. |
| Keywords: | Algorithmic pricing |
| JEL: | D22 D43 L13 L71 |
| Date: | 2025–01 |
| URL: | https://d.repec.org/n?u=RePEc:cpr:ceprdp:19830 |
| By: | Nicolas Eschenbaum; Janusz M. Meylahn |
| Abstract: | Detecting algorithmic collusion is challenging because regulators often have limited access to firms' algorithms, training data, and market information. We study an intermediate-information regime in which an auditor can query firms' frozen pricing policies and construct the induced strategy graph. Using a complete characterization of Nash equilibria in a repeated pricing game, we identify graph-theoretic features of strategy graphs that are associated with collusive reward-and-punishment schemes, including maximum betweenness, attractor in-degree, and average path length. We then test these metrics on policies learned by decentralized Q-learning and the Q-learning algorithm of Calvano et al. (2020). We find that especially the maximum betweenness and attractor in-degree are strongly correlated with the standard profit-based Collusion Index. Importantly, the proposed metrics rely only on the unlabeled topology of strategy graphs and require neither price histories, demand estimates, nor competitive and monopoly benchmarks. Our results suggest that the structure of frozen pricing policies contains robust signals of collusion among reinforcement learning algorithms and provides a promising basis for auditing algorithmic pricing systems under limited information. |
| Date: | 2026–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2608.07098 |
| By: | Xing, Xianghui; Dai, Hongwei; Zhou, Yiwei; Lan, Zhou; Wu, Xinyue; Jiang, Jie; Cao, Ling |
| Abstract: | Artificial intelligence technologies are rapidly penetrating and reshaping the labor market. As a result, perceived AI substitution risk among workers has become an important issue affecting employment stability and social well-being. Using data from the Chinese Social Survey 2023, this study empirically examines the determinants of workers' perceived AI substitution risk and explores group heterogeneity. The results show that the perceived unemployment risk, number of children, and unemployment insurance are significantly positively associated with perceived AI substitution risk. In contrast, gender (Female = 0, Male = 1), age, perceived life difficulties, overall job satisfaction, job skill level, perceived socioeconomic status, and ethnicity (non-Han ethnicity = 0, Han ethnicity = 1) have a significant negative impact. The mediation analysis indicated that job satisfaction and perceived socioeconomic status reduced perceived AI substitution risk by lowering individuals' subjective evaluations of perceived unemployment risk. The heterogeneity analysis further shows that the perceived unemployment risk functions as a common core pressure source across groups. Its effect strength is constrained by the degree of technological penetration and occupational stability. Meanwhile, the influence of individual characteristics and resource endowments follows a threat appraisal-stress response pathway, with significant differences across geographical regions, urban-rural attributes, and work patterns. This study provides empirical evidence from China and offers a new analytical perspective for countries seeking to address employment anxiety triggered by technological substitution, optimize education and skills training systems, and maintain stability in the global labor market. |
| Keywords: | artificial intelligence substitution;perceived artificial intelligence substitution risk;determinants |
| JEL: | R14 J01 |
| Date: | 2026–07–16 |
| URL: | https://d.repec.org/n?u=RePEc:ehl:lserod:140342 |
| By: | Nikolova, Milena (University of Groningen); Milanova, Vilian (University of Groningen); Wang, Feicheng (University of Groningen) |
| Abstract: | This paper provides the first causal evidence that merely attributing identical creative work to AI rather than to a human affects how much meaning people derive from a task and how much effort they are willing to contribute. We conducted a preregistered survey experiment in nationally representative samples from the United States (N = 1, 511) and the Netherlands (N = 2, 117). Participants evaluated identical public health campaign slogans that were randomly attributed either to an AI system or to a human professional, allowing us to isolate the causal effect of AI attribution while holding the creative output constant. AI attribution reduced perceived task meaning modestly and made participants 13% less likely to contribute a slogan of their own, indicating lower voluntary effort. These findings suggest that AI can influence work not only by changing productivity but also by altering the perceived value of human contribution itself. |
| Keywords: | Artificial intelligence (AI), meaning, effort, survey experiment |
| JEL: | C91 J01 I30 O33 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18784 |
| By: | Nikolova, Milena (University of Groningen) |
| Abstract: | This paper provides the first causal evidence on how Artificial Intelligence (AI)-based workplace safety systems shape perceived job quality. I conducted a preregistered vignette experiment with a nationally representative sample of 2, 172 Dutch adults who evaluated otherwise identical workplaces introducing one of three safety systems: human supervisors, AI-only monitoring, or hybrid AI-human supervision. Compared with human supervision, both AI-only and hybrid systems reduced perceived job satisfaction, work meaningfulness, and perceived social value of the job. Contrary to expectations, combining AI with human supervisors did not mitigate these negative effects. Respondents also viewed AI-based systems as less respectful of workers' privacy and dignity, despite viewing them as effective as human supervisors. Perceived fair wages changed little across conditions. These findings suggest that AI can influence work not only by improving safety but also by reducing important non-pecuniary dimensions of job quality, highlighting that the welfare consequences of workplace AI extend beyond productivity and accident prevention. |
| Keywords: | Artificial Intelligence (AI), safety systems, survey experiment, work meaningfulness, job quality |
| JEL: | I39 J01 J28 O33 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18782 |
| By: | Jennifer L. Steele; Isabella Cruz |
| Abstract: | How should people choose careers when artificial intelligence (AI) is rapidly transforming the nature of work? We first compare six recent projections of occupational exposure to task automation with AI, examining their methods and assumptions. We then propose a new empirical model of occupational AI exposure based on 2025 query data from Anthropic and OpenAI. We find marked heterogeneity in model predictions, though models published since 2020 show positive relationships among AI exposure, salaries, and occupational complexity. To reduce uncertainty due to heterogeneous assumptions about task automation potential, we average the projections from five models, including our own. Using these averages, we report on likely tradeoffs between salaries and AI exposure across interest categories, O*NET Job Zones, and job fields. Jobs in healthcare practice show the strongest balance of higher pay with lower AI exposure. Among jobs making high use of Anthropic's Claude, those that use it as a complement rather than a substitute for human work are modestly higher-paying, though whether this pattern holds will depend on usage norms adopted in each field. |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2607.15506 |
| By: | Danny, Lauren |
| Abstract: | Generative artificial intelligence has penetrated professional and academic environments to alter standards of intellectual output and productivity. Relying on semi-structured interviews with 45 participants across technology, legal, and academic sectors, this study examines how individuals utilize AI tools to construct professional identity. We propose the notion of algorithmic prestige: the strategic appropriation of machine-generated outputs to signal superior cognitive bandwidth and technical sophistication within competitive institutional hierarchies. We argue that while AI is ostensibly adopted for efficiency, it functions primarily as a mechanism for reinforcing existing power structures and performative competence. Our findings indicate that participants navigate AI use through a complex tension between productivity and authenticity. Transparency in tool usage often triggers professional penalties. Reliance on automated systems creates new dependencies. The pursuit of efficiency frequently devolves into intensive digital labor. Algorithmic prestige serves as a sociotechnical veneer that masks underlying anxieties regarding the obsolescence of human-centric expertise, suggesting that future policy must address the ethical imperatives of algorithmic transparency in high-stakes environments. |
| Date: | 2026–07–21 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:mdz83_v1 |
| By: | Danny, Lauren |
| Abstract: | Generative artificial intelligence integration into professional workflows alter the landscape of cognitive labor and knowledge production. Utilizing a qualitative study of 45 knowledge workers across technology and creative sectors, this research investigates the shifting boundaries of professional autonomy. We propose the notion of Prompt Dependency, defined as the systemic erosion of self-directed analytical capacity resulting from the habitual delegation of complex problem-solving to large language models. Users they perceive these tools as efficiency enhancers, but our analysis reveals an inversion of agency where the mastery of algorithmic interaction begins to supersede the mastery of the domain itself. Findings indicate that participants experience reduced critical synthesis and increased reliance on automated heuristics. Further observations it suggest the atrophy of foundational domain skills and a normalization of algorithmic output as a baseline for professional truth. By mapping these dimensions, we demonstrate that the efficiency gains of AI are paradoxically tethered to a long-term degradation of individual expertise. These results suggest that organizational policies must prioritize cognitive resilience or human-in-the-loop verification to mitigate the risks of structural deskilling in the age of automated cognition. |
| Date: | 2026–07–21 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:rmvqu_v1 |
| By: | Phurichai Rungcharoenkitkul |
| Abstract: | The AI build-out ranks among the largest technology-driven investment booms in US history. Its scale, reliance on debt and circular equity ties raise questions about the boom's sustainability and financial stability. We study a dynamic contest in which firms competing for a few dominant positions over-commit resources. The over-investment leaves the sector exposed to revenue disappointment that could turn boom into bust. The larger the boom, the deeper the eventual bust. The race to commit early through debt and circular financing also makes a bust more likely. Calibrated to balance sheet and deal data, the model points to over-investment of around 1.5 times the efficient level, rising to around three times where demand is less elastic. A network analysis shows that stress in one firm could cascade to others through chains of financial exposures. |
| Keywords: | artificial intelligence, investment, contest theory, circular financing, boom-bust cycle, financial fragility, network |
| JEL: | G01 G32 L13 O33 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:bis:biswps:1367 |
| By: | Ms. Natasha X Che; Weining Xin; Taichi Yoshida |
| Abstract: | Using a small open economy overlapping generations model, this paper examines how AI can drive economic divergence across and within Asian economies. While AI adoption may promise sizable productivity gains, it could create temporary but potentially longlasting divergence across countries. Structurally-prepared advanced economies tend to adopt AI earlier and see immediate growth gains while emerging markets and developing economies (EMDEs) face delayed adoption and initial growth headwinds from rising costs of capital. Structural reforms that boost productivity and strengthen human capital not only accelerate adoption in EMDEs but also amplify the growth gains. Within countries, AI adoption could widen inequality along multiple dimensions: across skill groups, as high-skilled workers benefit disproportionately from complementarity with the more capital-intensive technology, and across generation, as the shift of national income toward capital favors asset-rich older households relative to younger workers who rely primarily on labor income. Redistributive policies can help mitigate these distributional pressures, though they entail equity efficiency trade-offs that vary with country-specific fiscal and demographic conditions. |
| Keywords: | Artificial Intelligence; Automation; Technological Change; Economic Divergence; Structural Reform; Inequality; Redistribution; Asia; Overlapping Generations Model; IMF working papers; equity-efficiency trade-off; growth gain; productivity gain; capital share; Capital productivity; Income; Income inequality; Global; Asia and Pacific |
| Date: | 2026–08–07 |
| URL: | https://d.repec.org/n?u=RePEc:imf:imfwpa:2026/166 |
| By: | Yao, Becatien; Shanoyan, Aleksan |
| Abstract: | In many U.S. rural counties, agriculture remains a central economic activity, with jobs, incomes, and local public revenues closely tied to agri-food production. As generative artificial intelligence (AI) transforms workplaces across the economy, understanding how exposure varies across agri-food labor markets is an important first step toward evaluating potential implications for rural communities. Existing exposure measures often rely on detailed occupation data that are not consistently available at the county level, particularly in rural areas. This paper develops an occupation-based framework for measuring AI exposure using broad occupation groups reported in the American Community Survey and applies it to all U.S. counties. Exposure scores decline with rurality and are generally lower in farming, mining, and manufacturing-dependent counties. Comparison with an established task-based measure yields a state-level correlation of 0.93, indicating strong consistency between the two approaches. Using Quarterly Workforce Indicators data, the paper also examines employment trends across counties with different exposure levels. In highly exposed urban counties, employment growth among younger workers weakens relative to older workers after 2022. Similar patterns are not evident in the broader sample of rural counties and are less pronounced in farming-dependent counties. The results suggest that occupational composition remains an important source of variation in AI exposure and that the early labor market changes associated with generative AI may differ across local labor markets. |
| Keywords: | Agribusiness |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ags:aaea26:404319 |
| By: | Kellner, Domenic; Lang, Jan Hannes; Rusnák, Marek; Nagy, Lukas Joseph |
| Abstract: | Financial stability risks consist of two distinct components: vulnerabilities and possible trigger events. While there has been considerable progress regarding the measurement of vulnerabilities, the assessment of possible trigger events remains largely qualitative. To fill this gap, we employ Large Language Models to extract information about the Severity and Probability Of potential Trigger events (SPOT) from a large dataset of financial news articles over the period2005 – 2026. The SPOT indicator increases ahead of major historical trigger events, correctly identifies trigger sources, and helps to improve forward looking model estimates of downside risks to the economy. The results indicate that the use of AI-based signal extraction from text can be a promising avenue to improve the monitoring of financial stability risks. JEL Classification: C55, C88, E32, E44, G01 |
| Keywords: | artificial intelligence, crisis indicators, financial stability, growth-at-risk |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:ecb:ecbwps:20263262 |