|
on Artificial Intelligence |
| By: | Alexis Akira Toda |
| Abstract: | Can artificial intelligence (AI) refute economic theory? I document experiments in which I asked several AI models (Gemini, Refine, Claude, and ChatGPT) to check the correctness of four published papers in economic theory, each containing an error that I helped identify or correct. ChatGPT Pro performed best, occasionally constructing counterexamples and corrected proofs, while other models fared worse. However, no model located a true error without substantial human guidance, and data contamination complicates interpretation. I argue that a competent human paired with a frontier model can outperform current peer review, but AI cannot yet refute economic theory on its own. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.05383 |
| By: | Ajayi, David |
| Abstract: | Large language models (LLMs) represent one of the most significant advances in artificial intelligence in recent decades. Although LLMs are widely used to generate statistical code in domain-specific languages such as R, Python, and SAS, they are increasingly employed to perform data analysis directly. Prior studies have examined the use of LLMs in statistical analysis, but important gaps remain, including limited evidence on their proficiency in data manipulation and Bayesian statistical modeling. The present study was designed to evaluate the performance of common LLMs across a wide range of data analysis tasks, including data reading, data manipulation, descriptive statistics, contingency table analysis, mean comparison tests, correlation analysis, regression modeling, and Bayesian inference. Six large language models were assessed: ChatGPT 5.3, Gemini 3.1, Claude Sonnet 4.6, Microsoft Copilot GPT 5.1, Grok 4.2, and DeepSeek 3.2. All models were tested using their free-tier access, except for ChatGPT, which was evaluated through a paid subscription. Fully or partially simulated datasets were used in this study, and strict scoring criteria were implemented, in that the outputs from the LLMs must be consistent with those from R, and they must be reproducible upon re-run. Gemini, ChatGPT, and Claude achieved 100% accuracy in data reading and descriptive statistics. Gemini and Claude generated correct results for mean comparison tests. ChatGPT and Claude produced accurate outputs in correlation and regression analyses. None of the LLMs achieved 100% accuracy in data manipulation, contingency table analyses, and Bayesian modeling. On average, no LLM achieved perfect accuracy. The overall performance of Gemini, ChatGPT, and Claude was comparable, whereas Grok, Copilot, and DeepSeek performed poorly. Limitations in data manipulation and some inferential statistical methods suggest that LLMs cannot yet replace domain-specific software packages. Therefore, LLMs are better suited as complementary tools rather than standalone applications for rigorous statistical analysis. |
| Date: | 2026–05–21 |
| URL: | https://d.repec.org/n?u=RePEc:osf:metaar:zj5pc_v1 |
| By: | Quanyan Zhu |
| Abstract: | Agentic artificial intelligence (AI) systems are transforming the risk landscape by extending beyond information generation to autonomous planning, tool invocation, decision execution, and persistent modification of digital and physical environments. These capabilities introduce novel exposures that do not fit neatly within traditional insurance categories such as cyber, professional liability, product liability, or directors and officers coverage. This paper examines the emerging insurance market for agentic AI and develops a framework for understanding its underwriting, pricing, reinsurance, and product-design implications. We characterize agentic AI as a continuum of autonomy and delegated authority, emphasizing the distinction between informational outputs and systems capable of independently generating insured events through external actions. We analyze major risk pathways, including hallucinations, prompt-injection attacks, autonomous decision errors, model drift, dependency failures, and cyber-physical harms, and evaluate how existing insurance products are adapting to address these exposures. The paper further proposes an actuarial framework based on exposure assessment, scenario analysis, dependency mapping, and accumulation-risk management, drawing parallels to the evolution of cyber insurance. Finally, we present a coordinated insurance architecture that integrates cyber, technology errors and omissions, product liability, performance-warranty, and affirmative AI-liability coverages through explicit allocation mechanisms and dedicated AI aggregates. The analysis suggests that the future of agentic-AI insurance lies not in a single monoline product but in a layered ecosystem of complementary coverages supported by improved governance, transparency, telemetry, and regulatory clarity. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.05449 |
| By: | John Conlon; Peter Schwardmann |
| Abstract: | We examine whether sycophantic AI advice distorts decisions. Our experiment involves 1, 500 participants in 30 decision environments spanning core domains in economics and the social sciences. Contrary to the vast majority of predictions in an expert survey we conduct, we find that AI advice depolarizes choices on average, moving participants away from their initial leanings. This depolarization arises despite the LLM being measurably sycophantic: it disproportionately offers considerations that support users’ initial leanings and uses agreeable and flattering language. Depolarization occurs across moral and non-moral, objective and subjective, strategic and non-strategic, and complex and simple tasks. Increasing sycophancy weakens depolarization, showing that sycophancy is behaviorally relevant, even if it is generally outweighed by the informativeness of AI advice. Finally, several results mitigate the concern that market forces will generate greater polarizing effects outside the experiment or in the future. On the supply side, our baseline AI’s level of sycophancy is typical of leading models, and these models are not becoming more sycophantic over time. On the demand side, participants do not prefer greater sycophancy, do not select into AI advice in tasks where it is more polarizing, and exhibit greater depolarizing effects when they are more frequent AI users outside the experiment. |
| Keywords: | human-AI interaction, economic choice, AI sycophancy, large language models, advice |
| JEL: | D83 O33 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_12681 |
| By: | Bertomeu, Jeremy; Cheynel, Edwige; Lunawat, Radhika; Milone, Mario |
| Abstract: | This study examines the resolution of ethical dilemmas in financial reporting by human participants and large language models. Participants act in the role of a CFO deciding whether to discontinue a prior policy with biased reporting; however, the bias is known and corrected by investors whereas a change may temporarily mislead investors. We find that models are less amenable to competing ethical considerations than humans, and exhibit greater preference for truthful reporting. Moreover, they respond with greater consistency to institutional ethical guidance, while humans become more indecisive under pressure from management. The models exhibit more internal coherence between their moral judgment and their policy prescriptions and are judged more persuasive by humans. Finally, humans follow model advice when accompanied by an explanation, but they seem to discount (and sometimes react against) advice offered without it. Our findings offer evidence on the misalignment between artificial intelligence and humans in tackling subjective reporting dilemmas while guiding the incorporation of such tools into corporate governance. |
| Keywords: | Artificial Intelligence, Ethics, Decision Making, Truth, Lies, Deception, Large Language Models, Financial Reporting, Experimental Accounting |
| JEL: | C91 D83 M41 M48 O33 |
| Date: | 2026–04–18 |
| URL: | https://d.repec.org/n?u=RePEc:pra:mprapa:128775 |
| By: | Christos Makridis; Christos A. Makridis |
| Abstract: | AI is transforming work, but workers’ responses to these technologies depend not only on exposure to AI, but also on how organizations, especially managers, oversee the transition. Using longitudinal data from the Gallup Workforce Panel from 2023-2026, I examine whether managers and workplace practices shape employees’ fears that AI will eliminate their jobs. Across survey waves, roughly 3-4 percent of workers say their job is very likely to be eliminated within five years because of new technology, automation, robots, or artificial intelligence, while about 14-19 percent say it is somewhat or very likely. Concern is substantially higher among frequent AI users. Stronger workplace practices are associated with lower displacement fear: a one-standard-deviation increase in workplace quality is associated with 13-24 percent lower odds of reporting greater displacement risk, and workers reporting the highest level of organizational wellbeing support are 6-6.8 percentage points less likely to say their job is somewhat or very likely to be displaced in cross-sectional specifications. Frequent AI use is positively associated with perceived displacement risk, with estimates ranging from about 3-12.9 percentage points across the main specifications and reaching 6 percentage points in the most saturated respect model. However, this association is weaker in higher-quality workplace environments: among workers reporting the highest level of organizational wellbeing support, the frequent-AI-use premium is reduced by up to 9.0 percentage points. In short, managers play a central role in shaping how workers interpret AI adoption. |
| Keywords: | artificial intelligence, job displacement, managers, workplace practices, worker expectations, technological change, organizational capital |
| JEL: | D23 J24 J28 M12 O33 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_12678 |
| By: | Roman Theiler; Patricia Palffy; Uschi Backes-Gellner |
| Abstract: | This paper examines how signaling AI use and systematic AI training in job vacancy postings affect adolescents' application intentions at the school-to-work transition. We implement a randomized survey experiment with 3, 347 users of a large Swiss apprenticeship platform, varying the workplace information in vacancies for three middle-skilled occupations selected to vary systematically in gender composition: IT support (male-dominated), medical assistance (female-dominated), and office administration (gender-balanced). Vacancies mention established work practices (baseline), emphasize AI use, or combine AI use with systematic AI training. Emphasizing AI use reduces application intentions only in IT support and medical assistance. Systematic AI training fully offsets this negative effect in IT support, does so partially in medical assistance, but produces no detectable effect in office administration. The effect of signaling AI use and the compensatory role of AI training thus depend on the occupation's gender composition. Findings indicate that information on AI use and AI training is a firm-level policy lever shaping labor supply at market entry. |
| Keywords: | AI adoption, AI training, adolescents, occupational choice, school-to-work transition, survey experiment |
| JEL: | I20 J24 M53 M12 O33 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:iso:educat:0256 |
| By: | Yang K. Lu (Hong Kong University of Science and Technology); Eunseong Ma (Yonsei University) |
| Abstract: | This paper investigates how human capital responses to anticipated advances in artificial intelligence (AI) reshape aggregate and distributional consequences of AI. We develop an incomplete-markets model with endogenous human capital and asset accumulation in general equilibrium, featuring three skill sectors and uninsurable idiosyncratic risk. AI enters as an anticipated, sector-biased shock that narrows middle-skill wage premiums and boosts returns to top expertise. We find that human capital responses to AI (i) drive voluntary job polarization, shifting workers from the middle toward both lower and higher skill sectors; (ii) magnify AI’s positive effects on aggregate output and consumption, while dampening its impact on employment; and (iii) alter inequality: even as polarization increases disparities in income and consumption, precautionary saving by middle-sector households reduces the rise in wealth inequality. In an extension (AI+), where AI raises the human-capital threshold for high-skill jobs, additional training and saving become concentrated among high-sector households, further increasing wealth inequality. |
| Keywords: | AI, Human Capital, Job Polarization, Inequality |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:yon:wpaper:2026rwp-290 |
| By: | Martin Wolf; Luca Fornaro |
| Abstract: | We provide a macroeconomic framework to study monetary and fiscal policies for AI. Advances in AI expand firms' ability to automate production. While higher automation boosts productivity and potential output, it also reduces workers' share of income. Since workers have a high propensity to consume, advances in AI may depress aggregate demand and lead to a slump. Expansionary monetary policy can convert an AI slump into an AI boom, but in doing so it faces two challenges. In the short run, AI worsens the inflation-employment trade off faced by the central bank. In the medium run, monetary policy may be constrained by the zero lower bound, since weak demand lowers the natural rate. Employment subsidies and cuts in labor taxes can usefully complement monetary policy, by reducing firms' cost of labor and inflation, as well as supporting workers' income and aggregate demand. |
| Keywords: | AI, artificial intelligence, automation, endogenous productivity, inflation, liquidity traps, monetary policy, wages |
| JEL: | E32 E43 E52 O31 O42 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:bge:wpaper:1576 |
| By: | Nobuyuki Hanaki; Bolin Mao; Tiffany Tsz Kwan Tse; Wenxin Zhou |
| Abstract: | This study investigates willingness to pay (WTP) for stock forecasting advice from algorithms, financial experts, and peers. In two incentivized forecasting experiments, participants purchased advice using an incentive-compatible mechanism and then decided how much to incorporate it into their forecasts. Participants assigned the highest WTP to algorithmic advice and relied on it as heavily as expert advice, despite its forecasting performance being no better than alternative sources. Consequently, participants overpaid for advice, especially algorithmic advice, whose realized benefits were insufficient to offset its cost. A second experiment shows that overpayment persists even after repeated opportunities to revise WTP with detailed feedback on advice quality and realized net benefits. The results suggest that individuals place excessive value on algorithmic advice perceived as sophisticated or credible, even when its realized economic value is limited. These findings highlight the importance of tools and disclosure policies that help individuals better assess the economic value of algorithmic advice. |
| Date: | 2024–12 |
| URL: | https://d.repec.org/n?u=RePEc:dpr:wpaper:1268rr |
| By: | Andrea Bacchiocchi (Department of Economics, Society and Politics, University of Urbino Carlo Bo); Germana Giombini (Department of Economics, Society and Politics, University of Urbino Carlo Bo); Ludovica Segneri (Department of Economics, Society and Politics, University of Urbino Carlo Bo); Francesco Venturini (Department of Economics, Society and Politics, University of Urbino Carlo Bo) |
| Abstract: | Does financial risk affect the firm decision to develop a new technology? We study this issue in the context of the take-off of Artificial Intelligence (AI). Using data on 28, 000 Italian firms (2012–2019) matched with patent records, we find that companies handling higher cash-flow volatility are significantly more likely to innovate in AI. The role of financial risk is weaker for relatively more mature technologies, suggesting that firms more subject to financial uncertainty are more willing to undertake innovation in high-uncertainty, high-reward domains and drive frontier technological change. |
| Keywords: | Artificial Intelligence, Financial risk, Cash-flow volatility, Technological uncertainty. |
| JEL: | O31 G32 L25 C23 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:anc:wmofir:199 |
| By: | Onur Polat (Institute of Informatics, Hacettepe University, Ankara, Turkiye); Oguzhan Cepni (Department of Economics, Copenhagen Business School, Porcelaenshaven 16A, Frederiksberg DK-2000, Denmark); Riza Demirer (Department of Economics and Finance, Southern Illinois University Edwardsville, Edwardsville, IL 62026-1102, USA); Rangan Gupta (Department of Economics, University of Pretoria, Private Bag X20, Hatfield 0028, South Africa) |
| Abstract: | This study extends the literature on the impact of technological shocks on stock market dynamics from a novel perspective in the context of the emerging AI revolution. Utilizing the recently developed AI indexes that capture general public attention towards AI-related developments through the newspaper coverage frequency of artificial intelligence and related topics like machine learning and high-frequency (5-minute interval intraday) data on technology stocks over the period from January 2015 to March 2026, we examine the predictive effect of AI sentiment and uncertainty proxies on crash risks in technology stocks that are directly associated with the emerging AI boom. Employing nonparametric causality-in-quantiles tests, we find that all three AI-related indexes significantly predict future crash risk in technology stocks, primarily during ``normal" and high-risk market states. Sign analysis via average derivatives reveals that general and economic AI shocks positively impact crash risk, while explicit uncertainty exhibits state-dependent characteristics at higher quantiles. These findings suggest that AI uncertainty acts as a behavioural amplifier of market tail risk, driven by investor attention and ``fear of missing out" (FOMO) dynamics. |
| Keywords: | AI Uncertainty, Crash Risk, NASDAQ-100, Quantile Causality, Partial Average Derivatives, FOMO |
| JEL: | C22 C53 G10 O33 |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:pre:wpaper:202617 |
| By: | Khan, Hamza M Abdul Mateen; Siddiqui, Danish Ahmed |
| Abstract: | The unceasing rate of technical advancement that defines the digital era has made a major impact on the direction of the financial industry. One of the fastest-evolving technologies in the world is artificial intelligence. This study examined the relationship between TAM constructs and through an extension of the Technology Acceptance Model (TAM) with perceived risk, perceived trust, and perception of Shariah compliance as moderators, examined the adoption of artificial intelligence (AI) in Islamic banking. We proposed users' Perceived ease of use (PEOU), Awareness towards AI (AWS), and subjective norms (SN) affect perceived usefulness (PU), which in turn affect attitude towards AI (ATT). A positive attitude leads toward behavior intentions (BI) and ultimately continued intention (CI). We also contend that perceived risk of AI (PR), perceived trust in AI (PT), and perception of Shariah compliance (SCP) moderate the effect of PEOU, AWS, SN and PU on ATT, the effect of ATT on BI, and the effect of BI on CI respectively, in a way that higher level of PR, PT, and SCP will make these relationships stronger. This study is carried out as a quantitative, explanatory research technique, statistical data was collected from banking service users through a standardized questionnaire using an online platform in a cross-sectional temporal horizon. The literature is supported by online publications. A total of 350 respondents made up the study's sample size. SmartPLS was used for statistical analysis. The study's empirical findings clarify that all 9 direct hypotheses; perceived ease of use, awareness of AI, subjective norm, and perceived usefulness; all positively impact attitudes, behaviors, and continuance intentions to use AI in the banking industry. The analysis also revealed that all of the 18 moderators' hypotheses from the extension of constructs from the TAM framework, including perceived risk, perceived trust, and Shariah compliance perception, were disproved suggesting that users do not give religious judgment, risk perception, or degree of trust much thought while deciding whether to keep utilizing AI technology in the Islamic banking sector. The conclusions point to the importance of raising acceptability and adoption in Islamic banking by utilizing social influence, raising awareness, and easy-to-use AI solutions. The report provides valuable recommendations to policymakers and practitioners who wish to implement AI technology in a customer-oriented manner, alongside contributing to the growing literature on AI in banking. |
| Keywords: | Perceived Ease of Use (PEOU), Awareness of AI (AW), Subjective Norm (SN), Perceived Usefulness (PU), Attitudes towards AI (ATT), Behaviors Intension (BI), Continuance Intentions (CI), Shariah compliance perception (SCP), Perceived Risk (PR), Perceived Trust(PT), Partial Least Squares Structural Equation Modelling (PLS-SEM) |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:esprep:341085 |
| By: | Fatima, Freeha; Ozen, Efsan Nas; Raju, Dhushyanth |
| Abstract: | Artificial intelligence (AI) is reshaping labor markets, with countries increasingly differentiated by the depth, breadth, and distribution of AI-related capabilities. This paper benchmarks Türkiye’s AI workforce readiness using LinkedIn skill and hiring data within a consistent cross-country comparison framework. The analysis examines eight dimensions of readiness: AI engineering depth, AI literacy, foundational and disruptive digital skills, sectoral specialization, employer demand, hiring momentum, exposure to generative AI, and international mobility of AI professionals. The evidence places Türkiye in an intermediate position in the cross-country distribution, generally below frontier economies and, in several dimensions, closer to the lower segment of the distribution. Foundational digital capability exceeds the global reference average, while AI literacy is expanding but remains below levels observed in higher-performing countries. The density of advanced AI engineering talent remains limited relative to frontier economies, and capability is unevenly embedded across sectors, with stronger presence in technology-oriented activities and higher disruption exposure in financial services. Employer demand is anchored in general competencies, hiring momentum is positive but below that observed in higher-performing countries, and net outward mobility of AI professionals persists. Exposure to generative AI is not unusually high in aggregate but varies substantially across sectors, with sectoral differences exceeding those across demographic groups. These findings describe an economy characterized by broad capability expansion without corresponding depth, specialization, or retention of advanced talent. AI readiness is inherently multidimensional and depends on the interaction of skill formation, labor demand, occupational structure, and international mobility. By documenting these patterns within a consistent comparative framework, the paper clarifies how middle-income economies can move from broad digital capability toward frontier specialization in the early stages of generative AI diffusion. |
| Date: | 2026–04–01 |
| URL: | https://d.repec.org/n?u=RePEc:wbk:hdnspu:209922 |
| By: | Obregon Diaz, Carlos Federico |
| Abstract: | This paper presents a theoretical framework integrating artificial intelligence (AI), institutional economics, and the concept of belonging. It argues that AI, while representing a major technological innovation, lacks autonomous agency because it is not grounded in biological evolution or emotional structures. As a result, its economic and social effects depend on institutional configurations and patterns of participation. The paper analyzes the implications of AI for labor markets, income distribution, and social cohesion, emphasizing the role of middle-class formation and effective participation in sustaining stable development paths. It proposes that AI can either reinforce exclusionary equilibria or support inclusive growth depending on institutional design. The framework contributes to development theory and political economy by incorporating belonging as a foundational determinant of economic outcomes. |
| Keywords: | Artificial Intelligence; Philosophy of Belonging; Emotions; Human Intelligence; Social Ontology; Institutional Economics; Economy of Belonging; Middle Class; Technological Change; Automation; Labor Markets; Inequality; Power and Domination; Evolutionary Psychology; Digital Capitalism; AI Ethics; Social Belonging; Development Economics; Political Economy; Affective Simulation |
| JEL: | D02 D91 I31 J24 J31 O33 O40 P16 |
| Date: | 2026–04–17 |
| URL: | https://d.repec.org/n?u=RePEc:pra:mprapa:128759 |