|
on Artificial Intelligence |
| By: | Mallory Avery (Department of Economics, Monash University); Edwin Ip (Department of Economics, University of Exeter); Andreas Leibbrandt (Department of Economics, Monash University); Joseph Vecci (Department of Economics, University of Gothenburg) |
| Abstract: | Recent technological advancements are reshaping pathways to employment by automating the interview process. Asynchronous interviews, in which job applicants submit answers to interview questions via an online platform without interacting with an interviewer, are replacing more traditional face-to-face job interviews. At the same time, AI algorithms are now widely used to assess these interview answers. In this paper, we use a field experiment to comprehensively study how these new technologies affect applicants and employers in the recruitment process. Over 3, 000 job applicants are randomized into asynchronous audio or video interviews, live online interviews, and a control group. Their job interviews are then assessed by both professional recruiters and a commercial AI recruitment tool used by most Fortune 100 companies. We find that asynchronous interviews cause an over 50% decrease in application continuation, including among the most qualified applicants, and that this decline is largest for women. A complementary vignette experiment provides evidence that this deterrence is driven by perceptions about the competitiveness and fairness of the recruitment process. In terms of assessments, we find that the AI evaluation tool scores women and underrepresented racial minorities higher than human evaluators, while the opposite is true for men, Whites and Asians. We track our applicants' subsequent labor market outcomes and find that the AI assessment tool predicts subsequent employment success substantially better than human recruiters, suggesting that AI captures soft skills and potential that humans overlook. In addition, we provide evidence that, unlike AI, human recruiters' assessments suffer from multiple cognitive biases. Our findings provide some of the first key evidence on how recent technological advances are transforming the hiring process. |
| Keywords: | technological change, artificial intelligence, gender, field experiment |
| JEL: | C93 J23 J71 J78 |
| Date: | 2026–03–25 |
| URL: | https://d.repec.org/n?u=RePEc:exe:wpaper:2602 |
| By: | Alexander Erlei; Tahir Abbas; Kilian Bizer; Ujwal Gadiraju |
| Abstract: | Privacy concerns significantly impact AI adoption, yet little is known about how information environments shape user responses to data leak threats. We conducted a 2 x 3 between-subjects experiment (N=610) examining how risk versus ambiguity about privacy leaks affects the adoption of AI personalization. Participants chose between standard and AI-personalized product baskets, with personalization requiring data sharing that could leak to pricing algorithms. Under risk (30% leak probability), we found no difference in AI adoption between privacy-threatening and neutral conditions (ca. 50% adoption). Under ambiguity (10-50% range), privacy threats significantly reduced adoption compared to neutral conditions. This effect holds for sensitive demographic data as well as anonymized preference data. Users systematically over-bid for privacy disclosure labels, suggesting strong demand for transparency institutions. Notably, privacy leak threats did not affect subsequent bargaining behavior with algorithms. Our findings indicate that ambiguity over data leaks, rather than only privacy preferences per se, drives avoidance behavior among users towards personalized AI. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.08848 |
| By: | Chowdhury Mohammad Sakib Anwar; Konstantinos Georgalos |
| Abstract: | This paper investigates how natural language communication with an AI agent affects human cooperative behaviour in indefinitely repeated Prisoner's Dilemma games. We conduct a laboratory experiment (n = 126) with two between-subjects treatments varying whether human participants chat with an AI chatbot (GPT-5.2) before every round or only before the first round of each supergame, and benchmark against human-human data from Dvorak and Fehrler (2024) (n = 108). We find four main results. First, cooperation against the AI is high and initially comparable to human-human levels, but unlike in the human-human setting, where cooperation converges to near-complete levels, cooperation against the AI plateaus and never reaches full cooperation. Second, repeated communication, which substantially increases cooperation in human-human interactions, has no detectable effect in the human-AI setting. Third, strategy estimation reveals that human-AI subjects favour Grim Trigger under pre-play communication and remain dispersed under repeated communication, whereas human-human subjects converge to Tit-for-Tat and unconditional cooperation respectively. Fourth, human-AI conversations contain more explicit strategy commitments but fewer emotional and social messages. These results suggest that humans cooperate with AI at high rates but do not develop the trust observed in human-human interactions. Cooperation in the human-AI setting is sustained through conditional rules rather than through the social bonds and mutual understanding that characterise human-human cooperation. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.15852 |
| By: | Lingxiao Huang; Wenyang Xiao; Nisheeth K. Vishnoi |
| Abstract: | As AI systems enter institutional workflows, workers must decide whether to delegate task execution to AI and how much effort to invest in verifying AI outputs, while institutions evaluate workers using outcome-based standards that may misalign with workers' private costs. We model delegation and verification as the solution to a rational worker's optimization problem, and define worker quality by evaluating an institution-centered utility (distinct from the worker's objective) at the resulting optimal action. We formally characterize optimal worker workflows and show that AI induces *phase transitions*, where arbitrarily small differences in verification ability lead to sharply different behaviors. As a result, AI can amplify workers with strong verification reliability while degrading institutional worker quality for others who rationally over-delegate and reduce oversight, even when baseline task success improves and no behavioral biases are present. These results identify a structural mechanism by which AI reshapes institutional worker quality and amplifies quality disparities between workers with different verification reliability. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.02961 |
| By: | Lingxiao Huang (Nanjing University); Nisheeth K. Vishnoi (Yale University) |
| Abstract: | As AI systems shift from tools to collaborators, a central question is how the skills of humans relying on them change over time. We study this question mathematically by modeling the joint evolution of human skill and AI delegation as a coupled dynamical system. In our model, delegation adapts to relative performance, while skill improves through use and decays under non-use; crucially, both updates arise from optimizing a single performance metric measuring expected task error. Despite this local alignment, adaptive AI use fundamentally alters the global stability structure of human skill acquisition. Beyond the high-skill equilibrium of human-only learning, the system admits a stable low-skill equilibrium corresponding to persistent reliance, separated by a sharp basin boundary that makes early decisions effectively irreversible under the induced dynamics. We further show that AI assistance can strictly improve short-run performance while inducing persistent long-run performance loss relative to the no-AI baseline, driven by a negative feedback between delegation and practice. We characterize how AI quality deforms the basin boundary and show that these effects are robust to noise and asymmetric trust updates. Our results identify stability, not incentives or misalignment, as the central mechanism by which AI assistance can undermine long-run human performance and skill. |
| Date: | 2026–03–01 |
| URL: | https://d.repec.org/n?u=RePEc:cwl:cwldpp:2499 |
| By: | Salomé Baslandze; Zachary Edwards; John Graham; Ty McClure; Brent H. Meyer; Michael Sparks; Sonya R. Waddell; Daniel Weitz |
| Abstract: | We use novel data from a survey of nearly 750 corporate executives to study the effects of artificial intelligence (AI) on productivity and the workforce. We document substantial heterogeneity in AI adoption across firms, with more than half having already invested, though many smaller firms are only beginning to do so. Labor productivity gains are positive, vary across sectors, and are expected to strengthen in 2026, with the largest effects concentrated in high-skill services and finance. These gains are not primarily driven by firms' capital deepening but instead reflect increases in revenue-based total factor productivity, closely associated with innovation-and demand-oriented channels. We document a productivity paradox, in which perceived productivity gains are larger than measured productivity gains, likely reflecting a delay in revenue realizations. In labor markets, we find little evidence of near-term aggregate employment declines due to AI, though larger companies anticipate AI-driven workforce reductions, while smaller firms expect modest gains. We also find evidence of compositional reallocation of labor both within and across firms, with routine clerical roles declining and a relative demand for skilled technical roles increasing. We develop an index that ranks job functions most negatively affected by AI. |
| JEL: | D22 D24 G0 J01 J24 M15 O33 |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:34984 |
| By: | Michael Blank; Gregor Schubert; Miao Ben Zhang |
| Abstract: | This paper studies the impact of generative AI on U.S. households' task allocation at home, using detailed Internet browsing data from a large sample of home devices between 2021 and 2024. Leveraging pre-ChatGPT browsing patterns, we measure households' exposure to ChatGPT and use it as an instrument for ChatGPT adoption during the post-release period. Our IV estimates show that adopting generative AI substantially increases leisure browsing on home devices while leaving time spent on productive digital tasks unchanged. To examine mechanisms, we infer the purpose of households' ChatGPT use from surrounding internet activity and find that households primarily employ it for productive non-market tasks. Together, these results suggest that generative AI frees up leisure time by raising the efficiency of productive digital activities. Interpreting these findings through a standard time-allocation model implies economically large productivity gains from generative AI at home. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.03144 |
| By: | Dubey, Rohan (National Institute of Public Finance and Policy); Chakraborty, Lekha (National Institute of Public Finance and Policy) |
| Abstract: | The rapid diffusion of artificial intelligence (AI) has generated widespread expectations of substantial productivity gains, yet empirical evidence on its macroeconomic effects remains limited. This paper provides across-country empirical assessment of the relationship between AI adoption and labour productivity using a newly constructed panel dataset covering G20 over the period 2012–2023. We develop two composite indices of AI adoption that capture both relative cross-country positioning and within-country evolution overtime, drawing on indicators of investment, innovation, computational capacity, and scientific output. Employing panel regressions with country and time fixed effects and a rich set of macroeconomic controls, we find evidence of a statistically significant short-run effect of AI diffusion on aggregate labour productivity. These results are robust across alternative index constructions and model specifications. We then extend our analysis to human development indicators and find that AI diffusion is positively associated with UNDP the Human Development Index (HDI). At the sametime, the magnitude and dynamics of the estimated effects suggest that productivity gains from AI are likely to materialize gradually and depend on complementary investments and structural conditions. Beyond the regression results, the indices developed in this paper provide a transparent framework for tracking AI diffusion and identifying areas of AI preparedness and technological lag, offering useful insights for future research and policy design. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:npf:wpaper:26/445 |
| By: | Jose Maria Barrero; Nicholas Bloom; Philip Bunn; Steven J. Davis; Kevin Foster; Aaron Jalca; Brent Meyer; Paul Mizen; Michael Navarrete; Pawel Smietanka; Gregory Thwaites; Ben Wang; Ivan Yotzov |
| Abstract: | We present the first representative international data on firm-level AI use. We survey almost 6, 000 CFOs, CEOs, and executives from stratified firm samples across the US, UK, Germany, and Australia. We find four key facts. First, around 70 percent of firms actively use AI, particularly younger, more productive firms. Second, while over two-thirds of top executives regularly use AI, their average use is only 1.5 hours a week, with one quarter reporting no AI use. Third, firms report little impact of AI over the last three years, with more than 80 percent of firms reporting no impact on either employment or productivity. Fourth, firms predict sizable impacts over the next three years, forecasting AI will boost productivity by 1.4 percent, increase output by 0.8 percent, and cut employment by 0.7 percent. We also survey individual employees who predict a 0.5 percent increase in employment in the next three years as a result of AI. This contrast implies a sizable gap in expectations, with senior executives predicting reductions in employment from AI and employees predicting net job creation. |
| Keywords: | artificial intelligence; productivity; employment |
| JEL: | E0 |
| Date: | 2026–03–24 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedawp:102928 |
| By: | Drydakis, Nick |
| Abstract: | Artificial intelligence (AI) is increasingly recognised as a key driver of business innovation, yet its adoption among small and medium-sized enterprises (SMEs) varies considerably. This study examines whether AI Capital, defined as AI-related knowledge, skills and capabilities, is associated with business innovation among SMEs in England. Using a two-wave longitudinal panel dataset comprising 504 observations from SMEs collected in 2024 and 2025, the study develops and validates a 45-item AI Capital of Business scale. Business innovation is measured across five dimensions: product and service innovation, process innovation, technology adoption, market and customer engagement, and organisational culture and strategy. Regression models, including pooled OLS, Random Effects, and Fixed Effects specifications, are employed. The findings reveal a robust positive association between AI Capital and business innovation across all model specifications. This association holds across all business innovation dimensions and remains consistent for SMEs with differing levels of financial performance, size, and operational maturity. Each component of AI Capital independently exhibits a positive association with business innovation outcomes. The results highlight the central role of AI Capital in enabling SMEs to translate AI adoption into tangible business innovation. From a policy perspective, the findings indicate the value of targeted interventions that prioritise AI upskilling, organisational capability development, and accessible support mechanisms to promote inclusive and sustainable AI-driven business innovation among SMEs. |
| Keywords: | Artificial Intelligence, Artificial Intelligence Capital, Business Innovation, Innovation, SMEs |
| JEL: | O31 O33 O32 L26 L25 M15 D83 J24 O14 O39 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:glodps:1723 |
| By: | Giovanni Guidetti; Riccardo Leoncini; Mariele Macaluso |
| Abstract: | This paper studies patenting trends in artificial intelligence (AI) and robotics from 1980 to 2019. We introduce a novel distinction between traditional robotics and robotics embedding AI functionalities. Using patent data and a time-series econometric approach, we examine whether these domains share common long-run dynamics and how their trajectories differ across major innovation systems. Three main findings emerge. First, patenting activity in core AI, traditional robots, and AI-enhanced robots follows distinct trajectories, with AI-enhanced robotics accelerating sharply from the early 2010s. Second, structural breaks occur predominantly after 2010, indicating an acceleration in the technological dynamics associated with AI diffusion. Third, long-run relationships between AI and robotics vary systematically across countries: China exhibits strong integration between core AI and AI-enhanced robots, alongside a substantial contribution from universities and the public sector, whereas the United States displays a more market-oriented patenting structure and weaker integration between AI and robots. Europe, Japan, and South Korea show intermediate patterns. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.05034 |
| By: | Sihan Qian; Amit Mehra; Dengpan Liu |
| Abstract: | The rise of foundation models has driven the emergence of AI supply chains, where upstream foundation model providers offer fine-tuning and inference services to downstream firms developing domain-specific applications. Downstream firms pay providers to use their computing infrastructure to fine-tune models with proprietary data, creating a co-creation dynamic that enhances model quality. Amid concerns that foundation model providers and downstream firms may capture excessive consumer surplus, along with increasing regulatory measures, this study employs a game-theoretic model involving a provider and two competing downstream firms to analyze how policy interventions affect consumer surplus in the AI supply chain. Our analysis shows that policies promoting price competition in downstream markets (i.e., pro-price-competitive policies) boost consumer surplus only when compute or data preprocessing costs are high, while compute subsidies are effective only when these costs are low, suggesting these policies complement each other. In contrast, policies promoting quality competition in downstream markets (i.e., pro-quality-competitive policies) always improve consumer surplus. We also find that under pro-price-competitive policies or compute subsidies, both the provider and downstream firms can achieve higher profits along with greater consumer surplus, creating a win-win-win outcome. However, pro-quality-competitive policies increase the provider's profits while reducing those of downstream firms. Finally, as compute costs decline, pro-price-competitive policies may lose their effectiveness, whereas compute subsidies may shift from ineffective to effective. These findings offer insights for policymakers seeking to foster AI supply chains that are economically efficient and socially beneficial. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.12630 |
| By: | Daniel Parshall; Andrea Lopez-Luzuriaga |
| Abstract: | Existing frameworks for measuring AI's labor market exposure decompose imperfectly across distinct dimensions: whether AI can perform a task, whether deployment is physically feasible, and whether institutions permit it. We propose CDR, a three-axis ordinal taxonomy that separates these dimensions into Cognitive complexity (C0–C4), Deployment difficulty (D0–D4), and Regulatory restrictions (R0–R4), extending Autor's (2003) routine/non-routine x cognitive/manual framework into a finer-grained classification space suitable for measuring AI exposure. Applying CDR to the full O*NET task universe (23, 850 task-activity pairs across 923 occupations, classified via multi-model LLM consensus: Claude Sonnet 4.6, GPT-5-mini, Gemini 3 Flash, validated against flagship models), we find that 40.2% of U.S. economy-weighted labor time falls in tasks that are within current AI cognitive reach (C |
| Keywords: | AI; Task Exposure; Labor Time Allocation. |
| JEL: | O33 C49 J21 J22 |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:gwc:wpaper:2026-005 |
| By: | Hanyong Cho; Geumil Bae; Jang Ho Kim |
| Abstract: | This paper investigates how large language models (LLMs) form and express investor risk profiles, a critical component of retail investment advising. We examine three LLMs (GPT, Gemini, and Llama) and assess their responses to a standardized risk questionnaire under varying prompts. In particular, we establish each model's default investment profile by analyzing repeated responses per model. We observe that LLMs are generally longterm investors but exhibit different tendencies in risk tolerance: Gemini has a moderate risk level with highly consistent responses, Llama skews more conservative, and GPT appears moderately aggressive with the greatest variation in answers. Moreover, we find that assigning specific personas such as age, wealth, and investment experience leads each LLM to adjust its risk profile, although the extent of these adjustments differs across the models. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.09303 |
| By: | Hanyong Cho; Jang Ho Kim |
| Abstract: | This study introduces a benchmark framework for evaluating the financial decision-making capabilities of large language models (LLMs) through portfolio optimization problems with mathematically explicit solutions. Unlike existing financial benchmarks that emphasize language-processing tasks, the proposed framework directly tests optimization-based reasoning in investment contexts. A large set of multiple-choice questions is generated by varying objectives, candidate assets, and investment constraints, with each problem designed to include a unique correct solution and systematically constructed alternatives. Experimental results comparing GPT-4, Gemini 1.5 Pro, and Llama 3.1-70B reveal distinct performance patterns: GPT achieves the highest accuracy in risk-based objectives and remains stable under constraints, Gemini performs well in return-based tasks but struggles under other conditions, and Llama records the lowest overall performance. These findings highlight both the potential and current limitations of LLMs in applying quantitative reasoning to finance, while providing a scalable foundation for developing LLM-based services in portfolio management. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.09301 |
| By: | Pei-Jun Liao; Hung-Shin Lee; Yao-Fei Cheng; Li-Wei Chen; Hung-yi Lee; Hsin-Min Wang |
| Abstract: | Predicting stock prices presents challenges in financial forecasting. While traditional approaches such as ARIMA and RNNs are prevalent, recent developments in Large Language Models (LLMs) offer alternative methodologies. This paper introduces an approach that integrates LLMs with daily financial news for stock price prediction. To address the challenge of processing news data and identifying relevant content, we utilize stock name embeddings within attention mechanisms. Specifically, we encode news articles using a pre-trained LLM and implement three attention-based pooling techniques -- self-attentive, cross-attentive, and position-aware self-attentive pooling -- to filter news based on stock relevance. The filtered news embeddings, combined with historical stock prices, serve as inputs to the prediction model. Unlike prior studies that focus on individual stocks, our method trains a single generalized model applicable across multiple stocks. Experimental results demonstrate a 7.11% reduction in Mean Absolute Error (MAE) compared to the baseline, indicating the utility of stock name embeddings for news filtering and price forecasting within a generalized framework. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.19286 |
| By: | Yogesh Agrawal (University of Central Florida); Aniruddha Dutta (University of Central Florida); Md Mahadi Hasan (University of Central Florida); Santu Karmaker (University of Central Florida); Aritra Dutta (University of Central Florida) |
| Abstract: | Real-world financial decision-making is a challenging problem that requires reasoning over heterogeneous signals, including company fundamentals derived from regulatory filings and trading signals computed from price dynamics. Recently, with the advancement of Large Language Models (LLMs), financial analysts have begun to use them for financial decision-making tasks. However, existing financial question answering benchmarks for testing these models primarily focus on company balance sheet data and rarely evaluate reasoning over how company stocks trade in the market or their interactions with fundamentals. To take advantage of the strengths of both approaches, we introduce FinTradeBench, a benchmark for evaluating financial reasoning that integrates company fundamentals and trading signals. FinTradeBench contains 1, 400 questions grounded in NASDAQ-100 companies over a ten-year historical window. The benchmark is organized into three reasoning categories: fundamentals-focused, trading-signal-focused, and hybrid questions requiring cross-signal reasoning. To ensure reliability at scale, we adopt a calibration-then-scaling framework that combines expert seed questions, multi-model response generation, intra-model self-filtering, numerical auditing, and human-LLM judge alignment. We evaluate 14 LLMs under zero-shot prompting and retrieval-augmented settings and witness a clear performance gap. Retrieval substantially improves reasoning over textual fundamentals, but provides limited benefit for trading-signal reasoning. These findings highlight fundamental challenges in the numerical and time-series reasoning for current LLMs and motivate future research in financial intelligence. |
| Date: | 2026–03 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2603.19225 |
| By: | Servaas Storm (Delft University of Technology) |
| Abstract: | The AI industry is betting that 'scaling', i.e., adding more and more data, GPUs, compute infrastructure and dollars, will lead to machine superintelligence or Artificial General Intelligence (AGI) - which in turn will lead to exponential growth of output, productivity and profits for the industry and the larger American economy. Focusing on AGI and generic LLMs, the point of this paper is plain: AI's 'scaling' strategy must fail and the AI data-center investment bubble will pop. The paper identifies four bottlenecks: (1) the planned $5 trillion investment in data center infrastructure (during 2026-2030) is not going to pay off; AI revenues will not increase enough and AI inference cost continue to rise faster than revenues; (2) AI firms will have to resort to hyper-scale borrowing from banks and investment-grade bond markets to fund their capex; this hyperscale borrowing will create a ticking time bomb on the balance sheets of AI firms, because the core capital expenditure on specialized GPUs and server risks becoming economically obsolete within two or three years; (3) it will be impossible to build the projected data center infrastructure fast enough, because upstream suppliers - producing everything from copper wire to turbines to transformers and switchgear - will run into labor shortages, long waiting times for power grid connections, material bottlenecks and regulatory blowback; and (4) the strategic bet of frontier AI firms that AGI can be achieved by building ever more data centers and using ever more chips is already going bad; AI products will continue to be untrustworthy for high-stake usage. As a result, the magical projections of exponential growth, which defy economic and financial logic and fatally ignore unforgiving real-world constraints will turn out to be wrong. The fact that the AI industry is the main source of growth in an otherwise sclerotic U.S. economy and is driven by a concentrated set of hyper-scalers engaging in 'circular' financial transactions based on aggressively optimistic long-term cash flow-generating potential should be a very serious cause for concern. |
| Keywords: | Artificial intelligence; AGI; AI bubble; LLMs; circular financing; revenues; price-earnings ratio; leverage; scaling; inference cost; hallucinations; Chinese competition; AI-generated work-slop; misinformation; FOMO; price war. |
| JEL: | E24 F52 G01 O30 O33 |
| Date: | 2024–12–01 |
| URL: | https://d.repec.org/n?u=RePEc:thk:wpaper:inetwp244 |