|
on Artificial Intelligence |
| By: | Federico Atzori (Sapienza University); Luca Corazzini (University of Milan - Bicocca); Valeria Maggian (Ca’ Foscari University of Venice); Filippo Pavesi (LIUC University); Massimo Scotti (LIUC University) |
| Abstract: | We investigate how generative AI shapes creative performance and human-AI interaction in an open-ended writing task that employs a laboratory experiment in which participants are randomly assigned to either receive access to a large language model (ChatGPT-4.2) or not. Creative performance is measured by the average score assigned by independent evaluators recruited through the Prolific platform, and detailed logs of human-AI interaction are analyzed to measure AI use, prompting intensity, ideation requests, and the textual overlap between AI outputs and participants' final writings. Three main results emerge. First, AI access increases performance, but the gain is entirely driven by active use: participants with access who do not submit queries perform no better than those without AI. Second, the relationship between interaction intensity and performance is concave, peaking at roughly eight queries, consistent with iterative exploration rather than mechanical copying. Third, structural mediation analyses show that ideation requests affect performance primarily indirectly, by increasing downstream incorporation of AI-generated language; the direct effect of requesting an idea from the AI is negligible once execution-stage reliance is accounted for. We further document heterogeneity in AI reliance: cultural capital (proxied by books owned) predicts lower AI use, while prior AI exposure predicts higher use. By contrast, incentive schemes have limited effects on both outcomes and AI-related behaviors. |
| Keywords: | Human-AI Interaction; Creativity; Generative AI; Laboratory Experiment |
| JEL: | C91 D83 J24 O33 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ven:wpaper:2026:16 |
| By: | Yuhao Fu; Nobuyuki Hanaki; Haitao Wang |
| Abstract: | Artificial intelligence increasingly participates in economic interactions not only as a tool, but also as an autonomous bargaining counterpart negotiating on behalf of firms, platforms, and consumers. Yet little is known about how humans respond psychologically and strategically when bargaining with such agents in dynamic settings. We study this question in a laboratory experiment using a three-stage alternating-offer bargaining game in which participants negotiate in real time with either another human or a GPT-based AI agent. We also introduce a human-beneficiary condition in which the AI agent’s earnings may affect another participant’s payment. Agreements are not reached earlier in human–human bargaining than in human–AI bargaining, but they are reached significantly earlier when the AI’s payoff has human consequences. Human proposers offer more to human opponents than to AI agents, whereas responders become significantly more willing to accept unfair AI offers when AI earnings may benefit another human. These findings suggest that fairness and reciprocity toward AI are weaker and more conditional than toward humans, but partially re-emerge when AI outcomes affect real people. The results have implications for the design of AI negotiation systems and broader human–AI economic interactions. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:dpr:wpaper:1311 |
| By: | Jonathan Shaki; Eden Hartman; Sarit Kraus; Yonatan Aumann |
| Abstract: | Large language models (LLMs) are increasingly used to provide instructions to many agents who interact with one another. Such shared reliance couples agents who appear to act independently: they may in fact be guided by a common model. This coupling can change the prospects for cooperation among agents with misaligned incentives. We study settings in which multiple LLMs each advise a population of clients who participate in instances of an underlying game, creating strategic interaction at the level of the LLMs themselves. This induces a meta-game among the LLMs, mediated through clients. We first analyze the one-shot setting, where shared instructions can change equilibrium behavior only when an LLM may influence more than one role in the same interaction; in such cases, cooperation may emerge, and the effect of client share can be beneficial, harmful, or non-monotone, depending on the base game. Our main result concerns the repeated setting. We prove a folk theorem for LLMs: despite indirect observation and the clients' inability to identify which LLM advised their opponents, all feasible and individually rational outcomes can be sustained as $\varepsilon$-equilibria. The result does not follow from the standard folk theorem and requires new proof techniques. Together, these results show that shared LLM guidance can sustain cooperation among populations of agents even when the underlying incentives are misaligned. |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2605.06525 |
| By: | Sean S. Cao; Wei Jiang; Hui Xu |
| Abstract: | This research explores how human-defined goals influence the behavior of Large Language Models (LLMs) through purpose-conditioned cognition. Using financial prediction tasks, we show that revealing the downstream use (e.g., predicting stock returns or earnings) of LLM outputs leads the LLM to generate biased sentiment and competition measures, even though these measures are intended to be downstream task–independent. Goal-aware prompting shifts these intermediate measures toward the disclosed downstream objective, producing in-sample overfitting. Specifically, purpose leakage improves performance on data prior to the LLM’s knowledge cutoff, but provides no advantage after the cutoff. This bias is strong enough that regularization of prompt instructions cannot fully address this form of overfitting. We further show that the bias can arise from users’ unintentional conversational context that hints at the purpose. Overall, we document that AI bias due to “seeing the goal” is not an algorithmic flaw, but stems from human accountability in research design. |
| JEL: | G14 G17 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35142 |
| By: | Erin McGurk; David Khachaturov |
| Abstract: | We argue that AI-saturated markets are likely to create Veblen-good premiums, which we term human-provenance premiums, for verified human presence, and hence AI governance should treat human-provenance verification as labor infrastructure. Generative and agentic AI systems lower the cost of many standardized cognitive, creative, and coordination tasks, weakening the scarcity premiums that have supported much middle-tier knowledge work. We argue that this pressure may produce an asymmetric barbell-shaped structure of value capture in advanced economies: high-volume synthetic production controlled by owners of AI infrastructure at one pole, and scarce, high-status human labor valued for verified human presence at the other. We advance three claims. First, AI compresses the value of standardized middle-tier labor by making good-enough synthetic substitutes scalable at low marginal cost, hollowing out the middle of the skill distribution currently categorized by knowledge work. Second, this compression reallocates demand for human labor toward work valued for its visible human character. We term this performative humanity and distinguish three forms of labor: relational presence, aesthetic provenance, and accountability. Third, as these premiums depend on credible verification, AI governance should treat human-provenance systems as labor infrastructure rather than as luxury authenticity labels. To evaluate hybrid human-AI work, we propose constitutive human presence as the relevant standard: human labor retains premium value when human judgment, attention, accountability, authorship, or relational participation is not incidental to the output but constitutive of what is being purchased. |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2605.03210 |
| By: | Tom Davidson; Basil Halperin; Thomas Houlden; Anton Korinek |
| Abstract: | AI labs are increasingly using AI itself to accelerate AI research, creating a feedback loop that could lead to an intelligence explosion. We develop a general semi-endogenous growth model with an innovation network, where research and automation in one sector increase the productivity of research in other sectors, and derive a clean analytical condition under which growth becomes superexponential (``explosive''). We find that automating research can offset diminishing returns to ideas by activating two reinforcing channels: a technological feedback loop across research sectors, and an economic feedback loop in which higher output finances further research. Growth becomes explosive if the combined strength of technological and economic feedback loops overcomes diminishing returns. In a simple simulation calibrated to trends in AI progress, fully automating software research and modest (5%) automation in other sectors generates a singularity within six years. Bottlenecks do not overturn the result if task automation advances sufficiently fast. |
| JEL: | O31 O33 O40 O41 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35155 |
| By: | Fahn, Matthias (University of Hong Kong); Li, Jin (University of Hong Kong); Sun, Chang (University of Hong Kong) |
| Abstract: | We study how AI affects compensation and job design when performance depends on workers’ non-contractible effort. In a principal–agent model with limited liability, AI reduces effort costs but disproportionately lowers the cost of achieving satisfactory performance. This raises the incentive cost of sustaining high effort and can induce firms to replace high-wage, high-effort good jobs with low-wage, low-effort bad jobs, even when good jobs create more total surplus. As a result, AI can lower wages, reduce worker welfare, and even depress profits. If workers can adopt AI unilaterally, adoption occurs even when the resulting equilibrium harms both parties; when adoption requires worker cooperation, resistance is strongest where AI erodes rents embodied in good jobs. In a search-and-matching extension, endogenous outside options amplify these forces, reinforcing a bad-job economy and potentially reducing employment. |
| Keywords: | artificial intelligence, agency costs, job design, labor contracts, limited liability, incentives, search and matching |
| JEL: | D86 J41 O33 L23 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:iza:izadps:dp18574 |
| By: | Gmyrek, Pawel,; Berg, Janine,; Kaminski, Karol,; Konopczyński, Filip,; Ładna, Agnieszka,; Nafradi, Balint,; Rosłaniec, Konrad,; Troszyński, Marek, |
| Keywords: | artificial intelligence, automation, ISCO, occupational classification, employment, labour market analysis, survey |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:ilo:ilowps:995653377802676 |
| By: | Kathryn Bonney; Cory L. Breaux; Emin Dinlersoz; Lucia S. Foster; John C. Haltiwanger; Aditya A. Pande |
| Abstract: | Using novel, nationally representative data from the 2026 AI supplement to the U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS), we characterize AI diffusion across three layers: firm-wide adoption, business-function deployment, and worker-task use. During Nov 2025–Jan 2026, 18% of firms used AI in at least one function (32%, employment-weighted), with adoption expected to reach 22% within six months. Use is concentrated in large firms and knowledge-intensive sectors, reaching 50%–60% (60%–70%, employment-weighted) among very large firms in Information, Professional Services, and Finance. Among adopters, scope remains limited: 57% use AI in three or fewer functions, most often Sales and Marketing (52%), Strategy (45%), and IT (41%). Worker-level use appears in 23% (41%, employment-weighted) of firms, primarily for writing, document analysis, and information search; 65% restrict use to three or fewer tasks. Evidence suggests both top-down and bottom-up diffusion: worker use can occur without firm adoption, and vice versa. Most firms (66%) use AI for task augmentation, while employment reductions are rare (2%). Regression results show a positive relationship between firm performance and AI integration breadth. However, functional deployment and operational investment are associated with employment declines, while worker-task use is not once these factors are controlled for. |
| JEL: | L23 O33 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35141 |
| By: | Eduardo Levy Yeyati |
| Abstract: | Artificial intelligence (AI) is increasingly shaping economic structures, governance, and global power dynamics. Yet existing AI readiness indexes often provide a distorted view of countries’ capabilities—rewarding formal strategies and patents while overlooking deployment-first innovations, informal economies, and adaptive governance capacities. These biases particularly disadvantage Latin America and the Caribbean (LAC). This paper makes two contributions to the AI readiness agenda. First, it empirically documents the divergence and conceptual inconsistencies across leading AI readiness and preparedness indexes. Second, it proposes a two-tier measurement framework: a Composite Readiness–Preparedness Index (CoRPI), providing a transparent baseline diagnostic, and an Adaptive AI Readiness Index (AARI), capturing context-specific capacities and policy learning. Together, these frameworks aim to balance comparability with relevance. By piloting the AARI in LAC, the region can serve as a testbed for a model of AI governance and measurement with global applicability. |
| Date: | 2025–11 |
| URL: | https://d.repec.org/n?u=RePEc:udt:wpgobi:wp_gob_2025_2 |
| By: | Gmyrek, Pawel,; Viollaz, Mariana,; Winkler, Hernan, |
| Abstract: | This paper examines how Generative Artificial Intelligence (GenAI) may affect labour markets across 135 countries, covering around two-thirds of global employment. It focuses on how exposure differs between advanced and developing economies, and how digital infrastructure and task composition shape the balance between automation risks and productivity gains. |
| Keywords: | artificial intelligence, labour market analysis, access to information technology, labour productivity |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ilo:ilowps:995694369002676 |
| By: | Bruce I. Carlin; Ryan D. Israelsen; Christopher F. Wazzan |
| Abstract: | How does AI manage household stock portfolios? We collect a prospective, daily time-series of stock recommendations using several LLM's and study AI's investment style. AI recommends undiversified portfolios that positively load on momentum, large companies, and low book-to-market firms. AI primarily recommends stocks based on how much media attention firms receive. Using multiple versions of queries and requests, we find that buy-and-hold and actively managed AI portfolios do not appear to earn statistically-significant abnormal returns based on the methodology of Daniel et al (1997). |
| JEL: | G11 G12 G14 |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35153 |
| By: | Andrey Fradkin; Rohit Krishnan |
| Abstract: | Markets are a promising way to coordinate AI agent activity for similar reasons to those used to justify markets more broadly. In order to effectively participate in markets, agents need to have informative signals of their own ability to successfully complete a task and the cost of doing so. We propose MarketBench, a benchmark for assessing whether AI agents have these capabilities. We use a 93-task subset of SWE-bench Lite, a software engineering benchmark, with six recently released LLMs as a demonstration. These LLMs are miscalibrated on both success probability and token usage, and auctions built from these self-reports diverge from a full-information allocation. A follow-up intervention where we add information about capabilities from prior experiments to the context improves calibration, but only modestly narrows the gap to a full-information benchmark. We also document the performance of a market-based scaffolding with these LLMs. Our results point to self-assessment as a key bottleneck for market-style coordination of AI agents. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2604.23897 |
| By: | Olivia Zhang; Zhilin Zhang |
| Abstract: | Large language models (LLMs) are increasingly deployed in quantitative finance for stock price forecasting. This review synthesizes recent applications of LLMs in this domain, including extracting sentiment from financial news and social media, analyzing financial reports and earnings-call transcripts, tokenizing or symbolizing stock price series, and constructing multi-agent trading systems. Particular attention is paid to practical pitfalls that are often understated in the literature, such as fragility in sentiment analysis, dataset and horizon design, performance evaluation metrics, data leakage, illiquidity premia, and limits of stock price predictability. Organized from a hedge-fund perspective, the review is intended to guide both academic researchers and hedge fund managers in integrating LLMs into real-world trading pipelines and in stress-testing their robustness under realistic market frictions. |
| Date: | 2026–04 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2605.05211 |
| By: | Damm, Cara; Bauer, Kevin; Hett, Florian; Pelizzon, Loriana |
| Abstract: | The shift from information retrieval (keyword-based search engines) to information synthesis (generative AI) constitutes a fundamental change in how people inform themselves online. We investigate how this shift impacts investment behavior using an incentivized online experiment (N = 374), in which we vary whether participants have access to keyword-based search engines, an LLM-based chatbot, or no additional information source. We find that LLMs facilitate participation in the stock market. Participants with access to an LLM when making investment decisions are significantly more likely to enter the stock market and to remain invested compared to those with access to keyword-based search engines or no further information. Our experiment suggests that perceived difficulty of stock market participation decreases and confidence in these choices increases when using an LLM. However, we also document a substantial risk. Access to LLMs enables individuals to confirm and strengthen experimentally induced beliefs. Even when the chatbot itself is not biased, users can prompt the model to validate beliefs they want to hold. Overall, our findings suggest that while LLMs can reduce participation frictions and encourage stock market investments, their effectiveness in confirmation-seeking can also have detrimental consequences. Consequently, these results highlight the critical need for consumer protection frameworks and financial literacy programs that specifically address the unique dynamics of human-AI interaction in modern retail investing. |
| Keywords: | Large Language Models, Belief Formation, Motivated Reasoning, Financial Decision Making, Robo-Advisors, Stock Market Participation |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:safewp:340833 |
| By: | Maria Eugenia Heyaca; Andrea Pallotta |
| Abstract: | This paper examines how public audit institutions are exploring the use of artificial intelligence (AI) to strengthen oversight and improve audit processes. Drawing on consultations with 15 institutions across 14 countries and the European Union, it reviews emerging AI applications in areas such as anomaly detection, document processing, knowledge management and predictive risk assessment. The findings show that while AI adoption in public audit remains at an early stage, experimentation is expanding and many institutions are integrating AI within broader digital transformation efforts. However, a gap remains between pilot projects and scalable operational deployment. Key challenges include fragmented data systems, limited internal technical expertise and evolving governance frameworks. Strengthening data governance, digital infrastructure and internal development capacity will be critical for audit institutions seeking to responsibly scale AI while maintaining transparency, accountability and public trust. |
| Keywords: | Artificial Intelligence, Audit, Oversight, Public integrity |
| Date: | 2026–05–07 |
| URL: | https://d.repec.org/n?u=RePEc:oec:comaaa:58-en |
| By: | Doellgast, Virginia,; Appalla, Shruti,; Ginzburg, Dina,; Kim, Jeonghun,; Thian, Wen Li, |
| Abstract: | Employers are adopting and refining artificial intelligence (AI) and algorithm-based tools in the workplace, with wide-ranging implications for work and employment. This working paper examines case studies of social dialogue on AI at national, regional, sectoral, company, and workplace levels in Europe, North America, Asia, South America and the Caribbean, and Africa. Findings are organized around three distinct ‘action fields’ in which worker representatives have sought to influence strategies and outcomes associated with the growing use of AI and algorithms in the workplace. These include the employment and skill impacts of AI, algorithmic management practices, and working conditions and rights in AI value chains. Across these action fields, social dialogue is playing a crucial role in encouraging an alternative, high road approach to AI investments and uses, based on complementing rather than replacing worker skills, empowering rather than controlling the workforce, and embedding rather than displacing new jobs in labor and social protections. Comparative findings suggest that these social dialogue initiatives are more effective where there are constraints on employer exit, support for collective worker voice, and strategies of inclusive labor solidarity. |
| Keywords: | artificial intelligence, algorithmic management, social dialogue |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:ilo:ilowps:995660377202676 |
| By: | Arbaaz Karim |
| Abstract: | Aesthetic qualities command measurable premiums in traditional goods markets. However, it remains unclear whether users are willing to pay for such qualities in AI-generated text. This paper estimates the willingness to pay for aesthetic attributes in large language model outputs using an online experiment with N = 117 participants. Participants evaluated responses from four anonymized models across academic, professional, and personal contexts, rated outputs along multiple dimensions, and submitted bids for access using a Becker-DeGroot-Marschak (BDM) mechanism. We find no statistically significant relationship between perceived aesthetic quality and willingness to pay. While participants systematically distinguish between outputs and exhibit consistent preferences over stylistic features, these differences do not translate into higher monetary valuation. Further analysis shows that aesthetic and functional attributes load onto a single latent factor, suggesting that users perceive quality as a unified construct rather than a separable aesthetic dimension. These results imply that, in current large language model (LLM) markets, aesthetic improvements function as baseline expectations rather than sources of price differentiation. |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2605.05578 |