<?xml version="1.0" encoding="utf-8"?>
<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:rss="http://purl.org/rss/1.0/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/"><rss:channel rdf:about="http://lists.repec.org/mailman/listinfo/nep-ain">
<rss:title>Artificial Intelligence</rss:title>
<rss:link>http://lists.repec.org/mailman/listinfo/nep-ain</rss:link>
<rss:description>Artificial Intelligence</rss:description>
<dc:date>2026-11-09</dc:date>
<rss:items><rdf:Seq><rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:sza:wpaper:wpapers399&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:arx:papers:2610.32355&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:arx:papers:2610.17674&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:osk:wpaper:2613&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:arx:papers:2610.33447&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:arx:papers:2610.18435&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:arx:papers:2610.14984&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:rco:dpaper:589&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:arx:papers:2610.14036&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:zbw:glodps:1861&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:ces:ceswps:_13105&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:ces:ceswps:_13098&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:arx:papers:2610.31693&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:arx:papers:2611.01226&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:zbw:glodps:1856&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:arx:papers:2610.14850&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:cge:wacage:828&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:ces:ceswps:_13109&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:idb:brikps:14824&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:zbw:esrepo:344265&amp;r=&amp;r=ain"/>
<rdf:li rdf:resource="https://d.repec.org/n?u=RePEc:ces:ceswps:_13112&amp;r=&amp;r=ain"/>
</rdf:Seq></rss:items>
</rss:channel>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:sza:wpaper:wpapers399&amp;r=&amp;r=ain">
<rss:title>When Everyone Writes Well: The Economics of Writing with AI</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:sza:wpaper:wpapers399&amp;r=&amp;r=ain</rss:link>
<rss:description>Writing produces two things: the manuscript published now and the capability its author will hold next year. Artificial intelligence can raise the first while lowering the second, and which happens depends on the task the author delegates. The relevant comparison is between workflows rather than between using and refusing AI. This article separates writing into its tasks and applies a standard tool of economics to each, synthesising existing models into five predictions. Cheaper drafting lets an author test more versions of an argument, and the largest gains should go to researchers writing in a second language, whose language costs are unrelated to their evidence. Manuscript studies, including new evidence from economics working papers, fit this pattern, although they measure prose and output rather than argument quality. Polishing prose with a model while delegating the first formulation of a problem can leave a better text and a weaker author, and it makes the generated draft an anchor and a default. Cheap fluency makes polished prose a weaker guide to the underlying research. The evidence is uneven, and I note where it is thin. As models improve, the bottleneck moves, and so does the division of labour between author and machine.</rss:description>
<dc:creator>Johan Fourie</dc:creator>
<dc:subject>artificial intelligence, scholarly writing, human capital, language barriers, anchoring, judgement</dc:subject>
<dc:date>2026</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:arx:papers:2610.32355&amp;r=&amp;r=ain">
<rss:title>AI-Augmented Research: Free to Write, Costly to Read</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:arx:papers:2610.32355&amp;r=&amp;r=ain</rss:link>
<rss:description>We study a model in which artificial intelligence lowers the cost of writing faster than the cost of reading. Authors differ in reputation, and time-constrained readers, such as referees, choose between ignoring, verifying, and trusting papers. As writing costs fall, the band of verified papers shrinks: outsiders are no longer read, and established authors are no longer checked. In the ``costless writing limit, '' the share of verified papers vanishes, and a paper's impact depends on its author's reputation alone. Readers can escape the flood by asking AI to summarize the literature. These summaries aggregate papers reliably only if their errors are sufficiently independent. Once the share of papers written with the same AI exceeds a threshold, summaries inherit the AI's errors, and AI review filters out the original research that could correct them.</rss:description>
<dc:creator>Wolfgang Kuhle</dc:creator>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:arx:papers:2610.17674&amp;r=&amp;r=ain">
<rss:title>AI Information Synthesis and Content Polarization</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:arx:papers:2610.17674&amp;r=&amp;r=ain</rss:link>
<rss:description>Artificial intelligence (AI) can synthesize dispersed information into balanced and informative content, raising the possibility that AI may help reduce polarization. We show that this promise can be undermined by the strategic response of content producers. We develop a game-theoretic model in which consumers acquire information to make downstream decisions, content producers allocate limited informational capacity between broadly valuable vertical information and horizontally differentiated information, and AI provides an alternative source of synthesized information. The effects of AI unfold in three phases as its reliability increases. When reliability is low, AI has no effect on the market. At intermediate levels, AI disciplines prices without attracting consumers or changing content. When AI becomes sufficiently reliable, consumers begin to substitute toward it, and content producers respond by shifting informational capacity away from the vertical dimension and toward horizontally differentiated content, thereby increasing content polarization. The mechanism operates through audience composition: AI attracts consumers who are less well matched to individual producers, leaving producers with stronger incentives to cater to their remaining audiences. The polarization effect intensifies with AI reliability but weakens with greater competition among content producers. Our findings highlight a potential wedge between AI's direct and equilibrium effects: a technology that provides consumers with more balanced information can indirectly make the broader content ecosystem more polarized.</rss:description>
<dc:creator>Chang Liu</dc:creator>
<dc:creator>Yuting Zhu</dc:creator>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:osk:wpaper:2613&amp;r=&amp;r=ain">
<rss:title>More Knowledge, Less Difference: Generative AI and the Distinctiveness of Subsidiary Knowledge</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:osk:wpaper:2613&amp;r=&amp;r=ain</rss:link>
<rss:description>Multinational enterprises (MNEs) are increasingly providing the same generative AI tools to employees across their subsidiaries. Research on AI in international business has emphasized how AI transforms the way MNEs learn and decide, while research on AI and homogenization has shown that AI-assisted outputs tend to become more alike. Neither has examined how individual employeesâ€™ use of AI in subsidiaries affects the knowledge that reaches headquarters. I argue that generative AI creates a paradox for MNEs: by lowering writing and language barriers, it increases the amount of knowledge subsidiaries send to headquarters, while potentially reducing how distinctive that knowledge is. I identify four points at which AI enters the production of subsidiary knowledge (gathering, interpreting, writing up, and receiving) and develop propositions on how AI use at these points affects distinctiveness, which employees are most affected, and how subsidiariesâ€™ role in the MNE may change over time. I also propose a competing outcome: AI can increase distinctiveness when it enables employees previously limited by the corporate language to share their own observations. The decisive question is whether AI replaces the first-hand work through which local knowledge is produced or carries knowledge employees already hold. The paper contributes to research on reverse knowledge transfer, AI in international business, and AIdriven homogenization.</rss:description>
<dc:creator>Jiawei Song</dc:creator>
<dc:subject>generative AI; reverse knowledge transfer; MNE subsidiaries; homogenization; knowledge distinctiveness</dc:subject>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:arx:papers:2610.33447&amp;r=&amp;r=ain">
<rss:title>Efficiency vs Capability in the Intelligence Explosion</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:arx:papers:2610.33447&amp;r=&amp;r=ain</rss:link>
<rss:description>AI progress can feed back into AI research in two ways. Capability improvements allow AIs to perform a larger share of AI research tasks, while efficiency improvements raise AI research output per unit of compute on tasks already automated. In a task automation model of AI research, I show that a software-driven intelligence explosion - a self-sustaining acceleration in AI progress with fixed compute and human labor - requires both margins to improve sufficiently fast, and neither margin can substitute for the other. To see whether these conditions hold, I calibrate both margins with data on recent AI progress. I estimate that capability progress is likely fast enough to support an SIE, while efficiency progress is on the knife-edge. In Monte Carlo simulations, an SIE occurs in 49% of draws, indicating that an SIE is plausible but uncertain based on historical trends in AI progress. The single largest source of uncertainty is the rate of efficiency progress, whose uncertainty is responsible for 45% of the uncertainty about whether an SIE occurs, more than all other parameters combined. Thus, better data on efficiency progress in frontier models is essential for understanding the likelihood of an SIE.</rss:description>
<dc:creator>Karthik Tadepalli</dc:creator>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:arx:papers:2610.18435&amp;r=&amp;r=ain">
<rss:title>How Well Do LLMs Perform in Peer Review? Evidence from Finance and Accounting</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:arx:papers:2610.18435&amp;r=&amp;r=ain</rss:link>
<rss:description>Referees increasingly use large language models (LLMs), while finance and accounting journals disagree about whether to allow it. We ask which part of a referee's job the models can do. GPT-4o, whose training data end before the 2024 conference season, referees the papers presented at the 2024 AFA and NBER finance meetings, and we compare its verdicts with publication in the top-three finance journals over the next two years. Papers the model scores one standard deviation higher are 44\% more likely to be published, net of author reputation. The result is not driven by look-ahead bias, as the model does not recall the outcomes. The information comes from its judgements of correctness and exposition. Its judgements of originality and significance carry none, and a referee guide that stresses importance does not help. The split between correctness and contribution holds for accounting papers and for a newer model, GPT-5. LLMs can check whether research is right, but not whether it matters. The technical check can go to the model, while the judgement of contribution stays with people.</rss:description>
<dc:creator>Hai Lu</dc:creator>
<dc:creator>Yichen Luo</dc:creator>
<dc:creator>Luping Yu</dc:creator>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:arx:papers:2610.14984&amp;r=&amp;r=ain">
<rss:title>Too Late to Prune: Generative AI, the Screen, and Team Creative Decisions</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:arx:papers:2610.14984&amp;r=&amp;r=ain</rss:link>
<rss:description>Generative AI makes options cheap and shifts the bottleneck of creative work from generating options to screening them. We argue that AI's value depends on whether its options arrive before or after the team's screen. In our model, early options compete for a place in the portfolio; late options arrive after that competition and are kept unscreened. In a preregistered experiment, we randomly assigned 148 three-person teams to five schedules of access to the same language model during a product-design task with timed search and selection stages. Moving the same 15 minutes of access from search to selection lowered expert-rated creativity by 0.62 standard deviations, although teams used the model at similar rates. Platform logs locate the difference at the screen. Early access enlarged the pool of ideas fivefold, and teams pruned it. Late-access teams reached the screen with the same small pool as teams without AI and then searched again, devoting 29% of their selection-stage queries to generating new options, compared with 11% for teams with access throughout. Late access added breadth, not quality: these teams submitted 73% more proposals than did teams without AI. Language-model judges detected the benefit of access but not the cost of late timing.</rss:description>
<dc:creator>Difang Huang</dc:creator>
<dc:creator>Shumin Yu</dc:creator>
<dc:creator>Fadong Chen</dc:creator>
<dc:creator>Weihua Zhou</dc:creator>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:rco:dpaper:589&amp;r=&amp;r=ain">
<rss:title>The Good, the Bad, and the Bayesian: Self-Confidence and Algorithmic Advice</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:rco:dpaper:589&amp;r=&amp;r=ain</rss:link>
<rss:description>Understanding how decision makers incorporate external information is increasingly important as algorithmic predictions become a common source of advice. The literature on advice-taking suggests that people discount others' advice, especially when they feel competent. We experimentally study how self-confidence affects advice utilization and how algorithmic advice shapes confidence in final judgments. In our online experiment, participants make age predictions and assess the prediction accuracy before and after receiving algorithmic advice. We exogenously vary self-confidence through (perceived) task difficulty and vary the (observable) quality of the algorithmic advisor. While participants generally put more weight on high-quality ("good") advice, we find, contrary to the existing advice-taking literature, that they are less influenced by advice when they perceive the task as difficult. By contrasting age and accuracy predictions, we show that advice can influence beliefs without substantially changing point predictions: people who feel proficient at a task barely adjust their predictions toward high-quality advice but become more confident in them. When people feel less proficient, they are more reluctant to adjust their predictions toward low-quality ("bad") advice. Our design shows that prediction and confidence updating respond differently to perceived task difficulty and advice quality, implying that measures based solely on changes in point predictions can misstate the influence of advice. Comparing observed updating with a Bayesian benchmark further reveals systematic underutilization of advice, driven in particular by insufficient confidence updating following high-quality advice in tasks perceived as easy.</rss:description>
<dc:creator>Rongrong Fu</dc:creator>
<dc:creator>Fabian Gerstmeier</dc:creator>
<dc:creator>Michel Tolksdorf</dc:creator>
<dc:subject>algorithmic advice; algorithm appreciation; confidence; online experiment;</dc:subject>
<dc:date>2026-10-22</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:arx:papers:2610.14036&amp;r=&amp;r=ain">
<rss:title>AI Advice Reinforces Traditional Gender Roles</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:arx:papers:2610.14036&amp;r=&amp;r=ain</rss:link>
<rss:description>Professionals are increasingly turning to artificial intelligence (AI) for career advice. We tested whether large language models (LLMs) provide different advice to female and male job candidates in the context of career negotiations. Across two large-scale audit experiments, including four leading models, the LLMs not only recommended lower salary targets for female than male personas, but also constructed distinct negotiation agendas, prioritizing flexibility, caregiving, and well-being more highly for female candidates, and financial rewards, mobility, and advancement more highly for male candidates. These differences closely tracked independent human ratings of employment terms associated with primary caregiver and breadwinner roles, respectively. Our findings identify a subtle pathway through which personalized AI advice may reinforce gender inequality at work: steering otherwise equivalent users toward traditional gender roles.</rss:description>
<dc:creator>Nadja Born</dc:creator>
<dc:creator>Jared R. Curhan</dc:creator>
<dc:creator>Hannah Riley Bowles</dc:creator>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:zbw:glodps:1861&amp;r=&amp;r=ain">
<rss:title>From cheap answers to lasting learning: AI tutoring and educational inequality</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:zbw:glodps:1861&amp;r=&amp;r=ain</rss:link>
<rss:description>Nearly half of 15-year-old students in OECD countries, and more in some poorer countries, use AI weekly for learning. Generative AI makes schoolwork help cheap for connected students, but cheaper help need not yield lasting learning. We link the economics of supplementary education with learning science to examine how effort, entry costs, public provision and assessment shape AI tutoring's benefits. Randomized trials find gains and losses varying by setting, use and outcome, while lasting, equitable gains remain uncertain. We distinguish program effects from generative AI's contribution, compare full costs with alternatives, and propose decision rules and a trial design.</rss:description>
<dc:creator>Dang, Hai-Anh</dc:creator>
<dc:creator>Behrman, Jere</dc:creator>
<dc:subject>artificial intelligence, tutoring, learning, educational inequality, public investment, randomized trials</dc:subject>
<dc:date>2026</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:ces:ceswps:_13105&amp;r=&amp;r=ain">
<rss:title>How is AI Affecting College Major and Course Choices? Evidence from Finland</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:ces:ceswps:_13105&amp;r=&amp;r=ain</rss:link>
<rss:description>Have concerns about the potential job displacement effects of artificial intelligence affected the education choices of students? We analyze administrative data on the university applications and course choices of Finnish students. We find, after the launch of ChatGPT 3.5 in November 2022, a persistent drop in applications to fields of study classified as more exposed to AI under several different measures. The drop is particularly large in computer science and other ICT fields. Examining the elective course choices of non-ICT students, we find declining enrollment in programming and algorithms courses, particularly among students who had not taken mathematics as part of their high school exit exam. At the same time, non-ICT students enrolled more in AI/ML courses, and this increase was relatively larger among men.</rss:description>
<dc:creator>Ciprian Domnisoru</dc:creator>
<dc:creator>Tuomas P. Heikkinen</dc:creator>
<dc:subject>artificial intelligence, college major choice, higher education, unemployment</dc:subject>
<dc:date>2026</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:ces:ceswps:_13098&amp;r=&amp;r=ain">
<rss:title>Integration and Competition in Generative AI: Evidence from Google’s AI Mode Rollout</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:ces:ceswps:_13098&amp;r=&amp;r=ain</rss:link>
<rss:description>Integrating generative AI into digital platforms may reshape competition among AI assistants. We study whether Google’s introduction of AI Mode within Search reduces traffic to competing services. We exploit the August 2025 expansion of Google’s AI Mode, which reached 25 sample countries, using website traffic estimates for ten AI assistants and a difference-in-differences design with 30 countries where AI Mode was unavailable. Making AI Mode available within Google Search reduced ChatGPT website visits by 5.9 percent over the first four weeks, implying 3.5 million fewer daily visits across treated countries. The decline is concentrated in direct visits, while organic-search referrals change little. Daily unique visitors decline by a similar proportion, with little change in visits per daily visitor or average visit duration, suggesting the reduction primarily reflects lower daily reach. Point estimates remain negative throughout the 40-day follow-up, after which comparison countries also gain access to AI Mode. We find no statistically significant average effect on Gemini or the eight smaller assistants combined over the first four weeks. These findings suggest integrating conversational AI into established search interfaces can reduce traffic to standalone competitors, highlighting distribution channels’ importance for competition among AI assistants.</rss:description>
<dc:creator>Michelangelo Rossi</dc:creator>
<dc:creator>Adèle Diehl</dc:creator>
<dc:creator>Louis-Daniel Pape</dc:creator>
<dc:subject>generative AI, search engines, platform integration, platform competition</dc:subject>
<dc:date>2026</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:arx:papers:2610.31693&amp;r=&amp;r=ain">
<rss:title>When AI Gets Cheaper, Do Prices Follow? Information Delays and Algorithmic Pricing</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:arx:papers:2610.31693&amp;r=&amp;r=ain</rss:link>
<rss:description>We examine whether lower costs of running AI models lead to lower prices, and whether delays in observing competitors' prices and limits on price changes affect how much consumers benefit. We study a simulated market with two differentiated sellers, logit demand, and tabular Q learning. A factorial experiment varies the age of rival price information, how often sellers can revise prices, whether production costs fall, and differences in firm quality and costs. We evaluate policies trained on 60 paired market seeds in each of 16 conditions, for 960 training runs, and measure expected consumer surplus after learning ends. In the main setting with identical firms and falling costs, the estimated interaction between information delay and repricing frequency on consumer surplus is -0.00025 normalized units, with a 95% paired seed bootstrap interval of [-0.0402, 0.0393]. The interval includes zero, as do intervals for tests of whether the interaction changes with the cost regime or differences between firms. With the shortest information delay and the ability to revise prices every period, average price reductions equal 55.6% and 44.0% of the imposed cost reduction for the two firms. Mean consumer surplus under the learned policies is below the corresponding static grid Nash benchmark in all 16 conditions. The findings show that simulated cost savings need not be fully passed on to buyers. They do not establish collusion or describe pricing by actual AI providers.</rss:description>
<dc:creator>Krish Tadigotla</dc:creator>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:arx:papers:2611.01226&amp;r=&amp;r=ain">
<rss:title>Learning about Markets from Models: Evidence from an AI Hiring Platform</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:arx:papers:2611.01226&amp;r=&amp;r=ain</rss:link>
<rss:description>Machine learning models are increasingly used to not only inform, but make decisions. Sometimes platforms add value by pooling data across contexts to improve models' decision making---but when the optimal decision differs by context, pooling may not increase accuracy. We argue that a platform storing starkly distinct models across contexts nevertheless captures a model comparison externality. Together, these models reflect how decision makers respond differently to the same data in the current equilibrium. Policymakers can use this differentiation to prioritize interventions; firms can learn how they are already differentiated from competitors to focus future investments. Our evidence comes from an AI hiring platform used by large employers of low-skill workers. The platform uses a standardized psychometric assessment to create firm-specific models estimating each application's retention likelihood at that firm. Each firm's model is highly predictive for itself but no better than chance at other firms. We use these models to create a firm-application matrix: each firm scores each application, not only the ones it receives. An application's improved retention likelihood if submitted to a different firm is highly heterogeneous based on traits. Further, these gains are due to two distinct mechanisms. Finally, firms value traits differently even in this low skill market: they disagree about whether important traits improve or worsen retention at the margin. These insights are only possible when a platform can access both the full application pool and each firm-specific model. We explain the value of these insights to firms, policymakers, and platforms.</rss:description>
<dc:creator>Hossein Alidaee</dc:creator>
<dc:creator>Tarun Khanna</dc:creator>
<dc:creator>Rembrand Koning</dc:creator>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:zbw:glodps:1856&amp;r=&amp;r=ain">
<rss:title>Artificial Intelligence and Support for Labor Market Policies: Evidence from a Survey Experiment</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:zbw:glodps:1856&amp;r=&amp;r=ain</rss:link>
<rss:description>We study whether alternative narratives about Artificial Intelligence (AI) shape workers' support for strengthening labor market policies and whether these responses differ by workplace AI use. We conduct a four-arm survey experiment on a representative sample of 2, 105 Italian employees. Respondents are randomly assigned to a control group or to one of three information treatments framing AI as a technology that may substitute for human labor, complement workers' tasks, or increase the need for reskilling. We then elicit support for five labor market policy instruments, including both passive and active measures. Average treatment effects on policy support are small, but mask substantial heterogeneity by workplace AI use. Among non-users, the substitution narrative increases support for publicly funded training and hiring subsidies, while complementarity increases support for training. Among workplace AI users, both substitution and complementarity reduce support for unemployment insurance and short-time work schemes. Reskilling produces no statistically significant effects in either group. We interpret these heterogeneous effects using evidence that, among users, the substitution narrative increases the probability of expecting household financial losses from public intervention addressing AI's labor market effects, without a detectable change in perceived personal job-loss risk. Overall, workers' responses to the same information differ by workplace AI experience, both in the magnitude and direction of changes in support for specific labor market policies. The findings suggest that support for labor market protection may evolve as AI exposure and perceived vulnerability become more salient.</rss:description>
<dc:creator>Angei, Fabio</dc:creator>
<dc:creator>Balia, Silvia</dc:creator>
<dc:creator>Brau, Rinaldo</dc:creator>
<dc:creator>Sulis, Giovanni</dc:creator>
<dc:subject>artificial intelligence, labor market policy, workers' perceptions, policy support, survey experiment</dc:subject>
<dc:date>2026</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:arx:papers:2610.14850&amp;r=&amp;r=ain">
<rss:title>Displacement Is Endogenous: AI Workers in a Search-and-Matching Labor Market</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:arx:papers:2610.14850&amp;r=&amp;r=ain</rss:link>
<rss:description>We build a labor market in which AI systems enter as workers: they apply for jobs, pass screenings, get hired, and get dismissed, alongside humans. Firms post vacancies, workers search with friction, and product demand slopes downward. Across 30 seeded runs per configuration, AI entry at 70% of the human wage raises unemployment by 4.1 percentage points (95% CI 3.5 to 4.6) and leaves 12.2 fewer humans employed (CI 10.5 to 13.9). Layoffs dominate when AI is much cheaper than humans: the layoff margin accounts for 7.2 of the lost jobs and reduced hiring for 5.0, though most dismissed workers find work again. A wider price gap reduces displacement, because cheap AI makes firms expand: at price parity, with zero dismissals, the unemployment effect is largest at 9.2 points (CI 8.6 to 9.9), all of it through hiring. That parity result survives cutting AI applications from five per period to one. Demand elasticity decides how many workers AI displaces. Screening changes who gets hired and leaves the displacement total unchanged.</rss:description>
<dc:creator>Ibrahim Denis Fofanah</dc:creator>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:cge:wacage:828&amp;r=&amp;r=ain">
<rss:title>Who Pays for AI?</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:cge:wacage:828&amp;r=&amp;r=ain</rss:link>
<rss:description>We document who pays for AI tools using a panel of millions of open banking and cardholder accounts from 2022 to mid-2026, covering billions of transactions. We find that subscriptions to products like Claude and ChatGPT have roughly doubled every 12 months since their introduction, reaching one in twenty UK adults by mid-2026. Adoption is five times higher for under 30s than over 60s, 27 percent higher for men, and grows 1.8 times larger from the bottom to the top quintile of earnings. While ChatGPT held 74 percent of direct subscription spending in late 2025, its share fell 21 percentage points by mid 2026 as Claude surged from 7 to 29 percent of direct spending, driven partly by growth in users who work in technical and professional roles and pay for the most expensive plans. High-frequency banking data are well suited to study the rapidly evolving market for paid consumer AI products.</rss:description>
<dc:creator>Datta, Nikhil</dc:creator>
<dc:creator>Lambert, Peter John</dc:creator>
<dc:creator>Schneider, Patrick</dc:creator>
<dc:subject>artificial intelligence; technology adoption; subscriptions; transaction data; occupational exposure; product competition JEL Classification: O33, D12, L86, J24, C81, E21</dc:subject>
<dc:date>2026</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:ces:ceswps:_13109&amp;r=&amp;r=ain">
<rss:title>Drawing the AI Boundary in Production Networks</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:ces:ceswps:_13109&amp;r=&amp;r=ain</rss:link>
<rss:description>A technology can appear less connected to the economy when its definition expands to include more connected industries. We separate constituent connectivity from exposure crossing the group's boundary and show how internalizing links can reverse rankings. Using 2017 U.S. input–output accounts, we compare artificial intelligence (AI) with matched industry groups. Narrow and broad AI rank above the 96th percentile of internal shares among similarly sized groups, but fall to the 24th and 55th percentiles after also matching sector composition and connectivity. Sector composition accounts for most of the decline, distinguishing AI's industry mix from unusually strong connections among its members.</rss:description>
<dc:creator>Luisa Corrado</dc:creator>
<dc:creator>Christos A. Makridis</dc:creator>
<dc:subject>artificial intelligence, production networks, aggregation, input--output accounts</dc:subject>
<dc:date>2026</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:idb:brikps:14824&amp;r=&amp;r=ain">
<rss:title>Regulating AI Displacement: Public Demand, Institutional Credibility, and the Politics of Employment Protection in Latin America</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:idb:brikps:14824&amp;r=&amp;r=ain</rss:link>
<rss:description>How citizens want artificial intelligence (AI) regulated will shape which AI governance instruments are politically feasible. Yet we know little about public preferences over AI policy in middle-income democracies, where state capacity is contested. This paper provides the first multi-country experimental evidence on preferences over AI labordisplacement policy instruments in middle-income democracies. Using a randomized framing experiment in a sur-vey of Argentina, Brazil, Colombia, and Mexico, we examine how citizens map beliefs about AI-driven job disruption into demand for three governance responses: worker-retention reg-ulation, tax-financed redistribution to displaced workers, or non-intervention. Nearly three-quarters of respondents prefer employment-preserving regulation, about one in twenty favors tax-financed redistribution, and one-fifth prefer nonintervention. A three-arm framing taxonomy spanning alarmist, delayed-threat, and task reallocation narratives produces no average shift in this distribution, and the heterogeneous effects we detect at the margin run from non-intervention toward regulation, not toward redistribution. The pattern is consistent with preferences anchored in priors about the state's redistributive capacity rather than in correctable beliefs about AI, with direct implications for AI governance design in low-credibility institutional environments. A within-survey wording experiment on employment protection regulation elicits a substantial shift in preferences among the same respondents, indicating that the null on the AI framings is not an artifact of respondent inattention.</rss:description>
<dc:creator>Rosino, Maria Victoria</dc:creator>
<dc:creator>Scartascini, Carlos</dc:creator>
<dc:subject>regulation;employment protection;Trust;Survey experiment;AI governance;Labor Market Regulation;Public Opinion;Framing Experiment</dc:subject>
<dc:date>2026-10</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:zbw:esrepo:344265&amp;r=&amp;r=ain">
<rss:title>AI Labels and Offering Size: A Hedonic Analysis of the SEC Form D Census, 2021–2026</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:zbw:esrepo:344265&amp;r=&amp;r=ain</rss:link>
<rss:description>Between 2021 and 2026 the artificial intelligence investment cycle reshaped venture capital, but whether AI labels command higher start-up offering sizes remains contested. Using the SEC Form D regulatory record across nine non-consecutive quarters (2021Q2-2026Q1; 130, 057 deduplicated filings and 71, 702 declared offerings above USD 10, 000), we estimate hedonic offering-size regressions. The corrected specification does not confirm a positive post-2022 AI premium and does not rule out small-to-moderate premiums: the AI × post-2022 interaction is -0.0029 (robust standard error 0.1740; 95% confidence interval -0.344 to +0.338; p = 0.9867), and none of five specifications is positive and significant. The apparent divergence is a right-tail phenomenon: in 2026 the mean AI offering is 2.31× the non-AI mean (bootstrap 95% confidence interval 0.25-7.94, including 1), while an exploratory rank-shift test is significant (p = 0.018, sensitive to classifier breadth); the median AI offering is smaller (0.70×) and the mean ratio falls to 0.93× under 1/99 winsorization. Overall top-1% capital concentration fell from 92.88% in 2021 to 59.11% in 2026 even as the largest AI rounds grew. We read the cycle as capital intensity plus extreme right-tail concentration in a few frontier financings, not a broad-based AI premium.</rss:description>
<dc:creator>Mandal, Karishma</dc:creator>
<dc:subject>Venture Capital, Start up financing, Artificial Intelligence</dc:subject>
<dc:date>2026</dc:date>
</rss:item>
<rss:item rdf:about="https://d.repec.org/n?u=RePEc:ces:ceswps:_13112&amp;r=&amp;r=ain">
<rss:title>Artificial Intelligence Innovation and Stock Market Returns: Some International Evidence</rss:title>
<rss:link>https://d.repec.org/n?u=RePEc:ces:ceswps:_13112&amp;r=&amp;r=ain</rss:link>
<rss:description>This paper examines the relationship between artificial intelligence (AI) innovation and stock market returns across 61 countries over the period 2010–2022. Using AI-related patent families as a proxy for AI innovation, it investigates whether AI patenting activity is associated with subsequent national equity market performance. The empirical analysis employs a panel regression framework with country fixed effects and explores geographical heterogeneity across three regional subsamples, the EU and the UK, Asia, and the Rest of the World, as well as the G7 economies. The results indicate a positive relationship between lagged AI patenting activity and annual stock market returns, although the strength and statistical significance of this association vary across country groups. The most robust evidence is found for the EU and UK and the G7 economies, where the estimated coefficients are positive and statistically significant. By contrast, the association is weaker in Asia and statistically insignificant in the Rest of the World. These findings contribute to the literature on technological innovation and financial markets by providing cross-country evidence on the relationship between AI innovation and equity market performance. They also highlight the geographical heterogeneity of this relationship, suggesting that differences in financial market structures and technological development may play a role in shaping how AI innovation is reflected in stock market returns.</rss:description>
<dc:creator>Guglielmo Maria Caporale</dc:creator>
<dc:creator>Ida Colella</dc:creator>
<dc:creator>Fabio Spagnolo</dc:creator>
<dc:creator>Nicola Spagnolo</dc:creator>
<dc:subject>AI, stock market returns, panel data, fixed effects</dc:subject>
<dc:date>2026</dc:date>
</rss:item>
</rdf:RDF>
