|
on Artificial Intelligence |
| By: | Jon Frost; Rudraksh Kansal; Kumar Rishabh; Vatsala Shreeti; Leanne Si Ying Zhang |
| Abstract: | Between 2021 and 2025, 28.7% of artificial intelligence (AI) firms’ investment deals (by deal value) involved a target company that was also an AI firm, while 55.2% of incoming investments in AI firms came from other AI firms.Of all the AI-to-AI investment deals between 2021 and 2025, 16.1% (by deal count) and 46.4% (by deal value) also involved commercial supply chain relationships between the investor and target firms.Circular investment relationships reflect key economic features such as the need to secure critical inputs and the presence of information asymmetries, yet they entail macroeconomic risks and increase opacity. |
| Date: | 2026–10–01 |
| URL: | https://d.repec.org/n?u=RePEc:bis:bisblt:137 |
| By: | Kexin Chen; Tao Lin; Jianwei Huang |
| Abstract: | AI advice increasingly enters human decision workflows before costly information acquisition and final action. This paper studies how such advice affects decision quality, measured by the gross utility of the final action. Advice shifts beliefs before the decision maker chooses an acquisition experiment with uniformly posterior-separable (UPS) costs. The analysis characterizes gross effects through the continuation gross value generated by net-optimal acquisition. Convexity of this value function is necessary and sufficient for every AI signal to weakly improve gross utility at every prior, while nonconvexity allows a binary signal to cause gross harm. The criterion gives tight state-space results. Every AI signal is weakly gross-improving in binary-state problems under arbitrary actions, payoffs, and UPS costs. A minimal three-state problem can generate gross harm through concentrated shutdown, where valuable acquisition stops. For classification problems with Shannon acquisition costs, the criterion becomes an active-set slope test. It guarantees weak gross improvement for homogeneous classification and all three-state weighted classification problems, while a minimal four-state failure arises from attention dilution, where acquisition shifts away from high-stake distinctions. Along a Blackwell-increasing mixture path, gross utility can first decrease and then increase. With endogenous interpretation, cheaper interpretation can reduce gross utility. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.35137 |
| By: | Yvonne Jie Chen (Asian Development Bank); Jie Gong (University of Hong Kong); Jin Li (University of Hong Kong); Zibo Zhao (Arizona State University) |
| Abstract: | While generative artificial intelligence (AI) promises productive efficiency, it can paradoxically lead to lower quality work. We conducted an experiment with professional illustrators and found that AI assistance flattens the quality curve—it accelerates initial gains but sharply diminishes the returns on sustained effort. Faced with this, a significant number of professionals made a strategic choice: they sacrificed the final quality to save time. Our finding highlights a critical challenge for generative AI, which can weaken the motivation required for creative excellence and innovation. In developing economies, where governments may seek to boost growth by using generative AI to substitute for scarce skills, our results suggest that such gains may come at the cost of the sustained effort through which higher-quality work and expertise are built. |
| Keywords: | AI;artificial intelligence;generative AI;motivation;labor productivity;technology adoption |
| JEL: | D91 O31 O33 J24 M50 |
| Date: | 2026–09–17 |
| URL: | https://d.repec.org/n?u=RePEc:ris:adbewp:023629 |
| By: | Benjamin T. Leyden; Daniela Scur |
| Abstract: | When a knowledge worker is given an AI assistant, they stop executing tasks and start delegating them. Thus, we propose that the use of AI is an act of management, which helps explain the heterogeneity in returns to AI use observed across firms. We model the return to AI as moderated by two managerial abilities: direction (how well the worker specifies what they delegate) and verification (how reliably they catch the tool's errors). As AI becomes more reliable, the skill with the higher marginal return shifts from verification to direction. To the extent that use builds skill, workers develop their managerial abilities in the normal course of work, but as the tool improves, direction will continue to build while verification will not. As increasingly fewer tasks are worth checking, the skill collapses. Workers will reach that point sooner if they only bear part of the cost of their errors. In equilibrium, the strongest verifiers tend to sort into the jobs where collapse is most costly. |
| Keywords: | artificial intelligence, delegation, managerial ability, skill formation |
| JEL: | M54 J24 O33 D24 D83 D86 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_13030 |
| By: | Hattori, Keisuke |
| Abstract: | Replacing a human follower with AI that supplies more input can reduce team output. With complementary effort, a leader works harder because a human follower observes this effort and responds with greater effort. AI that matches the follower's contribution removes this reason to work hard, so the leader works less and team output falls. In larger teams, the remaining human followers also work less. Profit-maximizing firms may nevertheless adopt AI that contributes more while the team produces less. Falling AI costs can trigger an output drop at adoption, followed by recovery. Replacing followers can mean discouraging leaders. |
| Keywords: | Artificial intelligence, Team production, Leadership, Strategic complementarity, Automation |
| JEL: | D23 J23 M54 O33 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:esprep:344041 |
| By: | Binelli, Chiara; Cossu, Elena |
| Abstract: | Participation in AI research shapes which type of AI is built, whose societal needs it addresses, and how it is used in society. Despite increasing attention to gender diversity in AI, we lacksystematic evidence on the research topics, methods used, and collaboration structures associated with female AI researchers. This paper addresses this significant gap by developing the first large-scale, validated mapping of women’s contributions to both generative and non-generative AI research across data science and the main social science disciplines (economics, political science, and sociology). Using a sample of over 500, 000 papers published between 2023 and 2026, we document three key findings. First, women are under-represented among academic AI authors across all four disciplines and recent years, and least represented in data science. Second, conditional on participating, women’s AI research is thematically distinctive: it is more applied, social and purpose-driven, concentrated in health, education and fairness, and, in data science, oriented towards using rather than building AI. Third, this distinctive work is heard less: conditional on working on the same AI topic women and men are cited equally, but the topics women disproportionately work on are themselves cited less, so women’s ideas receive systematically fewer citations. This voice gap is therefore a topic-composition penalty rather than a same-topic penalty, and it reduces efficiency by giving a distinctive, society-oriented research agenda too little voice in how AI develops. |
| Date: | 2026–09–10 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:cxmk6_v2 |
| By: | Radoslaw Kotkowski; Michal Polasik; Andrzej Meler |
| Keywords: | Generative AI; Technology adoption; Digital divide; Gender gap; Diffusion of innovations |
| JEL: | C25 D83 J16 O33 |
| Date: | 2026–08 |
| URL: | https://d.repec.org/n?u=RePEc:cxu:wpaper:64 |
| By: | Kevin Rinz |
| Abstract: | AI mentions in US job advertisements have increased substantially since the beginning of 2024. The increase is concentrated in occupations more intensely exposed to the capabilities of large language models: one additional standard deviation of exposure is associated with a 3.1 percentage point increase in the rate at which job ads mention AI. Alongside this increase in AI mentions, more exposed occupations have seen job postings stabilize after a period of relative decline, posted wages rise, and hires and separations rise, all relative to less exposed occupations. Realized wages for new hires have risen consistent with posted wages, while wages of continuously employed workers have declined, possibly because exposed occupations have seen labor markets become less tight. |
| Keywords: | artificial intelligence; labor demand; skills; job postings |
| JEL: | J23 J24 O33 |
| Date: | 2026–09–24 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedcwq:103850 |
| By: | Donghang Li; Yunhan Zheng; Alok Prakash; Shenhao Wang; Jinhua Zhao |
| Abstract: | Artificial intelligence capabilities are often measured as resources accumulated within firms, yet they also circulate across organizational boundaries through worker mobility. Drawing on a database of approximately 535 million employment records across 58 countries, we reconstruct the inter-firm network of AI talent flows over 2010--2022. We find that AI talent inflows are becoming increasingly concentrated among a small set of leading firms, with stronger preferential attachment than in general labor mobility. Yet the network core remains contestable: central positions in AI networks change hands far more often than in non-AI networks. Network position also carries information that goes beyond workforce size. This network position is economically consequential: AI network centrality is associated with higher enterprise value, after accounting for workforce scale, assets, and fixed effects. Event-study estimates further indicate that sharp improvements in network position are followed by higher enterprise value relative to matched firms with similar pre-event trajectories. Together, these findings suggest that corporate AI advantage depends not only on how much talent firms accumulate, but also on where they sit in its circulation. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.08264 |
| By: | Maria D. Tito |
| Abstract: | Artificial intelligence (AI) technologies have the potential to transform production processes and labor markets, yet their ultimate effects on employment, occupations, and skill demand remain uncertain (Acemoglu and Restrepo, 2018). While research continues to explore implications of AI adoption for productivity and employment, firms' hiring decisions provide a timely measure of employers' evolving skill requirements (Deming and Kahn, 2018; Hershbein and Kahn, 2018; Acemoglu et al., 2022). |
| Date: | 2026–09–30 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedgfn:103871 |
| By: | Mühlbauer, Sabrina (Institute for Employment Research (IAB), Nuremberg, Germany); Ziethmann, Paula (Institute for Employment Research (IAB), Nuremberg, Germany); Weber, Enzo (Institute for Employment Research (IAB), Nuremberg, Germany) |
| Abstract: | "Despite widespread recognition of intersectionality in discrimination research, most algorithmic fairness evaluations remain limited to single-attribute group comparisons or individual-level prediction errors. This paper introduces a distributional approach to intersectional fairness that evaluates how machine learning models reshape the allocation of outcomes across intersecting social groups. Using large-scale administrative labour-market data from public employment services, we analyse an AI-supported job matching system with 141 occupational classes and more than three million observations. We measure intersectional disparities via Jensen–Shannon divergence and compare observed occupational distributions to model-implied allocations. We show that unconstrained models do not merely reproduce existing labour-market structures but systematically amplify intersectional disparities across occupations. Fairness-aware regularization reduces aggregate disparities but introduces a pronounced trade-off with predictive performance and reallocates distortions across occupations rather than eliminating them. Counterfactual averaging achieves substantial reductions in occupation-level disparities while preserving most predictive performance, may provide a more favourable fairness-performance trade-off in this application setting than directly constraining model outcomes. To identify localized fairness risks, we propose an occupation-level audit mechanism that flags cases with unusually large distributional shifts. Even under strong fairness constraints, a small number of occupations exhibit substantial residual distortions, indicating that fairness interventions compress rather than remove inequality. Our results demonstrate that fairness in large-scale decision-support systems must be evaluated at the level of outcome distributions. More broadly, they highlight the limits of purely technical mitigation and the need to combine fairness-aware modelling with explainability, monitoring, and human oversight in real-world deployment." (Author's abstract, IAB-Doku) ((en)) |
| JEL: | J64 J71 C55 |
| Date: | 2026–09–16 |
| URL: | https://d.repec.org/n?u=RePEc:iab:iabdpa:202606 |
| By: | Coyle, Diane; Nguyen, Nghi; Poquiz, John Lourenze; Riley, Rebecca |
| Abstract: | This paper provides early evidence that AI adoption by firms is linked to management practices in unique ways. Using panel data from the UK Management and Expectations Survey (MES), we find that firms with higher baseline management scores are significantly more likely to adopt AI. This relationship is specific to AI: the same management practices show no relationship with the adoption of other technologies such as robotics, specialized equipment, or specialized software. We also find that management capabilities related to performance measurement appear strongly related to AI adoption. These results suggest that organisational complementarities are specific to different technologies, with implications for policies to encourage AI adoption by firms. |
| Keywords: | AI; technology adoption; management practices; technology diffusion; organisational capital; decentralisation; ICT |
| JEL: | D24 L23 O31 O33 |
| Date: | 2026–09–28 |
| URL: | https://d.repec.org/n?u=RePEc:eoe:escoed:escoe-dp-2026-13 |
| By: | Alex Blumenfeld; Jonathon Hazell; Chen Lian; Andreas Schaab |
| Abstract: | We measure how artificial intelligence (AI) affects the economy through its impact on software engineering productivity. We use information from financial markets to develop a forward-looking measure that is available in real time. We estimate the sensitivity of each firm’s stock return to an AI stock market index, and how this sensitivity depends on the share of firm payroll in software engineering. We use a model to map this cross-sectional relationship into software engineering productivity gains. From November 2022 to December 2025, AI increased the market’s expected present value of software engineering productivity by the equivalent of a permanent 32.6% productivity increase. The corresponding effect on the level of GDP is 3.6% in the baseline and 6.5% when higher software engineering productivity also raises R&D productivity. By mid-2026, amid rapid progress in coding agents, the effect of AI on productivity and GDP had more than doubled relative to the end of 2025. |
| JEL: | E22 G12 O33 O47 |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35793 |
| By: | Maria Sole Pagliari |
| Abstract: | This paper identifies AI news shocks from financial markets and traces their macroe-conomic effects in the United States and the euro area. We combine a daily equity factor based on differences in firms’ AI exposures with sign, magnitude, and narrative restrictions in a Bayesian VAR to isolate revisions in AI’s expected economic value. Monthly local projections reveal an investment-led adjustment in the US: capital-goods orders rise early and persistently, while aggregate activity responds more gradually and with greater uncertainty. Producer prices increase before consumer prices, and the expansion spreads from sectors closely connected to the AI production network to the broader economy. In the euro area, equity prices rise and the euro depreciates; import and energy costs feed into inflation before real activity strengthens. Inflation expectations and nominal rates also increase. The results highlight a monetary policy dilemma: financial spillovers and imported price pressures arrive before the broader productive gains from AI adoption. |
| Keywords: | AI news shocks; macroeconomic dynamics; international spillovers; pro-duction networks |
| JEL: | C32 E32 E44 F41 O33 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:dnb:dnbwpp:871 |
| By: | Leidorf-Tidå, B.J.A.; König, Pascal |
| Abstract: | Trust in artificial intelligence (AI) applications is deemed crucial for uptake of the technology and the development of an AI market. In the present article, we argue that the institutional design of regulatory regimes matters for whether citizens trust AI systems. We study how trust in AI systems depends on who the responsible regulator is, the scope of the regulator’s scrutiny of AI systems, and the involvement of external stakeholders in regulatory oversight. Based on a conjoint design, we examine how respondents value these institutional design choices for the case of high-risk AI systems used in health care. Using samples from Lithuania, Romania, Sweden, and the Netherlands, with 4001 respondents overall, the pre-registered experiment shows that citizens have more trust in AI systems if vendors, civil society and especially independent researchers are involved in oversight, and if regulators actively scrutinize AI systems rather than relying solely on self-reporting. At least in two countries, citizens also seem to have more trust in AI systems if the sectoral health care regulator is the responsible regulator instead of a cross-sectoral regulator such as the national Data Protection Authority or a specific AI regulator. |
| Date: | 2026–10–01 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:by96n_v1 |
| By: | König, Pascal; Leidorf-Tidå, B.J.A. |
| Abstract: | This article examines EU citizens’ support for EU policy to attain digital sovereignty regarding Artificial intelligence (AI). Promoting domestic industries in AI is crucially important for attaining AI sovereignty. Yet, AI sovereignty policy ultimately hinges on support by the European public, and protective policy intervening in the EU AI market may be felt directly – and negatively – by citizens. Using survey data from Lithuania, the Netherlands, Romania, and Sweden, we examine how much citizens support AI sovereignty policy even if this comes with costs and study how political and AI attitudes are associated with this trade-off evaluation. The analysis shows that EU-centrism is positively associated with support for AI sovereignty policy whereas positive attitudes toward AI and higher trust in US and Chinese tech companies show negative associations. The results have important consequences for the EU’s possibilities to garner public support for an AI sovereignty policy agenda. |
| Date: | 2026–10–02 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:263yj_v1 |
| By: | Dang, Hai-Anh (World Bank); Do, Minh (University of Economics and Business, Vietnam National University, Hanoi) |
| Abstract: | Can artificial intelligence (AI) make scarce expertise more widely available in poorer countries and improve household welfare? Can it help measure welfare where surveys are infrequent? This structured critical review examines both questions and finds that the answers depend on the application and setting. Across nine selected evaluations, AI assistance raised output per unit of time by 29% on average, and the average remains positive in every reported sensitivity check. Eight tutoring comparisons give an average gain of 0.21 standard deviations on tests taken without the tool, including in poorer-country schools, but this result is less robust when higher-risk trials are excluded. Business and labor-market effects vary, and we found no qualifying causal study of a generative-AI tool’s effect on household consumption poverty in a poorer country. Measured use rises steeply with national income. AI-based models predict wealth levels more accurately than changes, and their value for targeting depends on the alternative and its cost. The review identifies promising applications for evaluated expansion, proposes a survey module, and sets out conditions for validating AI-based statistics and for assessing costs, persistence, and equity. |
| Keywords: | artificial intelligence, poverty, inequality, developing countries, measurement, critical review |
| JEL: | O33 I32 D63 J24 C81 |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18984 |
| By: | Harry Lyu; Neil Thompson |
| Abstract: | LLM judges are increasingly used to assess whether AI outputs meet workplace requirements, but agreement on response rankings does not establish agreement on acceptance rates or occupational aggregates. We introduce O*NET-BENCH, an audit suite derived from an existing survey of 45, 796 worker ratings, and evaluate 33 pre-existing judge configurations across six model families on 4, 501 test ratings. Twenty-five configurations achieve tie-aware pair accuracy of at least 0.60, although a train-fitted response-only TF-IDF baseline nearly matches the strongest judge. Despite this ordering agreement, judges estimate that 3.0%-97.9% of responses are acceptable, compared with 61.1% for occupation-matched workers. In one fine-tuned lineage, changing from pointwise scoring to a bundled few-shot/listwise protocol improves response ordering while reducing agreement with worker means at the task and occupation levels; this reversal replicates on a task- and worker-disjoint validation split under prespecified criteria. Cross-validated calibration largely removes mean bias, but calibrated scores explain at most 8.5% of individual worker-rating variance. Prediction-assisted estimation yields at most small precision gains at the studied label budgets. These results show that ranking agreement alone is insufficient for occupational measurement. Judges should be validated against the acceptance rates and aggregates their scores will be used to estimate. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.02492 |
| By: | Jie Gong; Jiayi Hou; Jin Li; Fei Pu; Xinjue Yao |
| Abstract: | We study gaming and adaptive incentive design in AI adoption. A large medical-device company required roughly 5, 000 employees to submit at least 200 queries per month to an internal AI assistant. First-time use increased significantly after the mandate, but usage patterns suggested gaming: query counts bunched at the threshold, and 31 percent of queries were repeated or off-task. The firm subsequently revised incentive design and lowered the target to 100. Using staggered implementation across branches, we estimate that the adjustment reduced query volume by 30 percent, with repeated or off-task queries accounting for about 90 percent of the decline. The estimated change in non-repeated, work-related queries was small and statistically insignificant. Among sales employees, monthly sales increased by 7 percent. The findings identify incentive adaptation as a consequential margin of technology adoption: recalibrating usage requirements can reduce gaming and improve employee performance. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.32859 |
| By: | Haochen Luo; Yifan Li; Binh Minh An; Xiaolong Luo; Zhengzhao Lai; Yuan Zhang; Chen Liu |
| Abstract: | Large language models (LLMs) and multi-agent systems (MAS) have shown promise in financial decision-making, yet existing evaluations focus on equity trading and primarily assess directional prediction, overlooking the structural complexity of derivative markets. Option trading introduces fundamentally different challenges, including nonlinear payoffs and multi-leg strategy construction, requiring structured decisions rather than simple directional bets. We introduce LiveOption, an evaluation framework for LLM-based agents in option trading. LiveOption formulates the problem as structured sequential decision-making under realistic execution and capital constraints, and provides a reproducible environment with standardized interaction protocols. The framework includes three task suites covering portfolio overlays, event-driven earnings trading, and 0DTE intraday trading. We further propose a hierarchical metric suite that evaluates action validity, decision quality, risk characteristics, and outcome-level performance. Experiments show that current agents often fail to achieve competitive returns in most scenarios. LiveOption offers a principled testbed for evaluating structured decision-making beyond outcome-based metrics. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.33470 |
| By: | Joshua S. Gans |
| Abstract: | How much liability should AI providers bear when their services enable both attack and defence? Liability can improve welfare while increasing harm. Providers sell a common input to productive users, attackers and defenders. Within a defended contest, a higher common price reduces effort without changing attack success or attacker profits, saving resources and improving the target's security payoff. Compensation weakens defence and raises attacker profits. Optimal liability balances these effects against productive exclusion. Greater competition can lower optimal liability; every such decline must end at an outcome retaining defence. With cybersecurity access fixed, monopoly can warrant partial liability but never full liability when provision is worthwhile. When guardrails preserving productive uses are available, strong competition favours universal guarding socially but encourages unilateral removal at insufficient liability. At a fixed provider count, sufficiently many productive users ensure a pure equilibrium with universal guarding under high liability. A universal-guarding requirement makes liability redundant. Under monopoly, adoption follows a unique liability threshold, while zero liability remains uniquely optimal for a range of parameters with sufficiently many productive users. |
| JEL: | D62 K13 L13 L51 L86 |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35828 |