|
on Artificial Intelligence |
| By: | von Zahn, Moritz; Liebich, Lena; Jussupow, Ekaterina; Hinz, Oliver; Bauer, Kevin |
| Abstract: | The use of explainable AI (XAI) methods to render the prediction logic of black-box AI interpretable to humans is becoming more popular and more widely used in practice, among other things due to regulatory requirements such as the EU AI Act. Previous research on human-XAI interaction has shown that explainability may help mitigate black-box problems but also unintentionally alter individuals' cognitive processes, e.g., distorting their reasoning and evoking informational overload. While empirical evidence on the impact of XAI on how individuals "think" is growing, it has been largely overlooked whether XAI can even affect individuals' "thinking about thinking", i.e., metacognition, which theory conceptualizes to monitor and control previously-studied thinking processes. Aiming to take a first step in filling this gap, we investigate whether XAI affects confidence calibrations, and, thereby, decisions to transfer decision-making responsibility to AI, on the meta-level of cognition. We conduct two incentivized experiments in which human experts repeatedly perform prediction tasks, with the option to delegate each task to an AI. We exogenously vary whether participants initially receive explanations that reveal the AI's underlying prediction logic. We find that XAI improves individuals' metaknowledge (the alignment between confidence and actual performance) and partially enhances confidence sensitivity (the variation of confidence with task performance). These metacognitive shifts causally increase both the frequency and effectiveness of human-to-AI delegation decisions. Interestingly, these effects only occur when explanations reveal to individuals that AI's logic diverges from their own, leading to a systematic reduction in confidence. Our findings suggest that XAI can correct overconfidence at the potential cost of lowering confidence even when individuals perform well. Both effects influence decisions to cede responsibility to AI, highlighting metacognition as a central mechanism in human-XAI collaboration. |
| Keywords: | Explainable Artificial Intelligence, Metacognition, Metaknowledge, Delegation, Machine Learning, Human-AI Collaboration |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:safewp:334511 |
| By: | Nobuyuki Hanaki; Bolin Mao; Tiffany Tsz Kwan Tse; Wenxin Zhou |
| Abstract: | This study investigates participants’ willingness to pay for stock forecasting advice from algorithms, financial experts, and peers. Contrary to prior findings on “algorithm aversion, ” participants valued algorithmic advice as much as expert advice and relied on it heavily, even though its performance was not superior. This algorithm appreciation reflects a shift in perceived reliability among students. However, it led to lower payoffs, as participants overpaid for advice that failed to significantly improve outcomes. These findings highlight the importance of developing tools and policies that help individuals better evaluate the actual value of algorithmic advice. |
| Date: | 2024–12 |
| URL: | https://d.repec.org/n?u=RePEc:dpr:wpaper:1268r |
| By: | Hangcheng Zhao; Ron Berman |
| Abstract: | Large language models (LLMs) change how consumers acquire information online; their bots also crawl news publishers' websites for training data and to answer consumer queries; and they provide tools that can lower the cost of content creation. These changes lead to predictions of adverse impact on news publishers in the form of lowered consumer demand, reduced demand for newsroom employees, and an increase in news "slop." Consequently, some publishers strategically responded by blocking LLM access to their websites using the robots.txt file standard. Using high-frequency granular data, we document four effects related to the predicted shifts in news publishing following the introduction of generative AI (GenAI). First, we find a consistent and moderate decline in traffic to news publishers occurring after August 2024. Second, using a difference-in-differences approach, we find that blocking GenAI bots can have adverse effects on large publishers by reducing total website traffic by 23% and real consumer traffic by 14% compared to not blocking. Third, on the hiring side, we do not find evidence that LLMs are replacing editorial or content-production jobs yet. The share of new editorial and content-production job listings increases over time. Fourth, regarding content production, we find no evidence that large publishers increased text volume; instead, they significantly increased rich content and use more advertising and targeting technologies. Together, these findings provide early evidence of some unforeseen impacts of the introduction of LLMs on news production and consumption. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.24968 |
| By: | Walter, Johannes |
| Abstract: | Rising political polarization generates significant negative externalities for democratic institutions and economic stability, yet scalable interventions to reduce polarization remain scarce. In this paper, I study whether AI chatbots can reduce political polarization. In two preregistered online RCTs with representative U.S. samples, I find that AI significantly reduces polarization on the Ukraine war and immigration policy. In Experiment 1, AI reduced polarization by 20 percentage points, with effects persisting for one month. Experiment 2 pits AI against incentivized human persuaders and Static Text. I find no significant difference in effectiveness: all three reduced polarization by roughly 10 percentage points. While AI conversations were rated as more enjoyable, mechanism analysis reveals that persuasion is driven by learning and trust, not enjoyment. These results demonstrate AI's scalable persuasive power, highlighting its dual-use potential: it can be deployed to effectively reduce polarization, but also poses risks of misuse. |
| Keywords: | Political Polarization, AI Persuasion, Experimental Economics, Information Provision |
| JEL: | D72 D83 D91 O33 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:zbw:zewdip:334534 |
| By: | Christos Makridis; Christos A. Makridis |
| Abstract: | Technological change has repeatedly disrupted creative labor markets, raising concerns about whether new tools substitute for artists or shift the organization of creative work. This paper studies how occupational exposure to generative AI (genAI) maps into employment and earnings outcomes for U.S. artists following the unanticipated release of ChatGPT. I combine an occupation-level LLM task exposure index with establishment-based occupational outcomes from the Occupational Employment and Wage Statistics (OEWS) and individual microdata from the American Community Survey (ACS), estimating event-study specifications that compare more versus less exposed artistic occupations from 2017 to 2024. Across datasets, I find little evidence of short-run earnings declines associated with LLM exposure through 2023, with point estimates near zero and in some specifications modestly positive. Evidence on employment is more mixed, with weaker employment growth in 2023 for more exposed artistic occupations in some specifications. To investigate mechanisms, I use the Gallup Workplace Panel from 2023 to 2025 to measure AI use directly and relate changes in AI use to job satisfaction and burnout. Within-person estimates show limited average well-being effects of adoption, but suggest heterogeneous responses for artists and a use pattern concentrated in ideation and creative-support tasks. The results are consistent with early task reallocation than immediate labor-market harm, while leaving open the possibility of medium-run adjustment as adoption deepens and complementary investments accumulate. |
| Keywords: | artificial intelligence, arts, culture, creative economy, large language models, generative AI, skills dynamics, workforce |
| JEL: | J23 J24 L82 O33 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_12368 |
| By: | Melanie Arntz; Myriam Baum; Eduard Brüll; Ralf Dorau; Matthias Hartwig; Britta Matthes; Sophie-Charlotte Meyer; Oliver Schlenker; Anita Tisch; Sascha Wischniewski |
| Abstract: | Artificial intelligence (AI) is diffusing rapidly in the workplace, yet aggregate productivity gains remain limited. This paper examines the dual diffusion of AI – through both formal, employer-led and informal, employee-initiated adoption – as potential explanation. Using a representative survey of nearly 10, 000 employees in Germany, we document a high extensive but low intensive margin of usage: while 64 percent use AI tools, only 20 percent use them frequently. This diffusion is strongly skill-biased and depends less on establishment and regional characteristics. While formality is associated with more frequent usage, training, AI-based supervision, and higher perceived productivity gains, it does not broaden access. These patterns suggest that widespread informal usage can coexist with limited productivity effects when complementary investments and organizational integration lag behind. |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ifowps:_422 |
| By: | Zoë B. Cullen; Ester Faia; Elisa Guglielminetti; Ricardo Perez-Truglia; Concetta Rondinelli |
| Abstract: | We present the first large-scale field experiment test of strategic complementarities in firms’ technology adoption. Our experiment was embedded in a Bank of Italy survey covering around 3, 000 firms. We elicited firms’ beliefs about competitors’ adoption of two advanced technologies: Artificial Intelligence (AI) and robotics. We randomly provided half of the sample with accurate information about adoption rates. Most firms substantially underestimated competitors’ current adoption, and when provided with information, they updated their expectations about competitors’ future adoption. The information increased firms’ own intended future adoption of robotics, although we do not observe a significant effect on AI adoption. Our findings provide causal evidence on coordination in innovation and illustrate how information frictions shape technology diffusion. |
| JEL: | C93 D22 L21 O33 |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:34532 |
| By: | Lapo Santarlasci; Armando Rungi; Loredana Fattorini; Nestor Maslej |
| Abstract: | Artificial intelligence has become a key arena of global technological competition and a central concern for Europe's quest for technological sovereignty. This paper analyzes global AI patenting from 2010 to 2023 to assess Europe's position in an increasingly bipolar innovation landscape dominated by the United States and China. Using linked patent, firm, ownership, and citation data, we examine the geography, specialization, and international diffusion of AI innovation. We find a highly concentrated patent landscape: China leads in patent volumes, while the United States dominates in citation impact and technological influence. Europe accounts for a limited share of AI patents but exhibits signals of relatively high patent quality. Technological proximity reveals global convergence toward U.S. innovation trajectories, with Europe remaining fragmented rather than forming an autonomous pole. Gravity-model estimates show that cross-border AI knowledge flows are driven primarily by technological capability and specialization, while geographic and institutional factors play a secondary role. EU membership does not significantly enhance intra-European knowledge diffusion, suggesting that technological capacity, rather than political integration, underpins participation in global AI innovation networks. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.19569 |
| By: | Daniel Bj\"orkegren |
| Abstract: | Market expectations about AI's economic impact may influence interest rates. Previous work has shown that US bond yields decline around the release of a sample of mostly proprietary AI models (Andrews and Farboodi 2025). I extend this analysis to include also open weight AI models that can be freely used and modified. I find long-term bond yields shift in opposite directions following the introduction of open versus closed models. Patterns are similar for treasuries, corporate bonds, and TIPS. This suggests that the movement of bond yields around AI models may be a function of not only technological advances but also factors such as licensing. The different movements suggest that markets may anticipate openness to have important economic implications. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.14969 |
| By: | Filippo Gusella; Eugenio Vicario |
| Abstract: | Results in the Heterogeneous Agent Model (HAM) literature determine the proportion of fundamentalists and trend followers in the financial market. This proportion varies according to the periods analyzed. In this paper, we use a large language model (LLM) to construct a generative agent (GA) that determines the probability of adopting one of the two strategies based on current information. The probabilities of strategy adoption are compared with those in the HAM literature for the S&P 500 index between 1990 and 2020. Our findings suggest that the resulting artificial intelligence (AI) expectations align with those reported in the HAM literature. At the same time, extending the analysis to artificial market data helps us to filter the decision-making process of the AI agent. In the artificial market, results confirm the heterogeneity in expectations but reveal systematic asymmetry toward the fundamentalist behavior. |
| Keywords: | Heterogeneous Expectations, Large Language Model, Generative Agent, Funda mentalists, Trend Followers |
| JEL: | E30 E70 D84 |
| Date: | 2025 |
| URL: | https://d.repec.org/n?u=RePEc:frz:wpaper:wp2025_18.rdf |
| By: | Yuhan Hou; Tianji Rao; Jeremy Tan; Adler Viton; Xiyue Zhang; David Ye; Abhishek Kodi; Sanjana Dulam; Aditya Paul; Yikai Feng |
| Abstract: | The Federal Open Market Committee (FOMC) sets the federal funds rate, shaping monetary policy and the broader economy. We introduce \emph{FedSight AI}, a multi-agent framework that uses large language models (LLMs) to simulate FOMC deliberations and predict policy outcomes. Member agents analyze structured indicators and unstructured inputs such as the Beige Book, debate options, and vote, replicating committee reasoning. A Chain-of-Draft (CoD) extension further improves efficiency and accuracy by enforcing concise multistage reasoning. Evaluated at 2023-2024 meetings, FedSight CoD achieved accuracy of 93.75\% and stability of 93.33\%, outperforming baselines including MiniFed and Ordinal Random Forest (RF), while offering transparent reasoning aligned with real FOMC communications. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.15728 |
| By: | Zhenyu Gao; Wenxi Jiang; Yutong Yan |
| Abstract: | We develop a statistical test to detect lookahead bias in economic forecasts generated by large language models (LLMs). Using state-of-the-art pre-training data detection techniques, we estimate the likelihood that a given prompt appeared in an LLM's training corpus, a statistic we term Lookahead Propensity (LAP). We formally show that a positive correlation between LAP and forecast accuracy indicates the presence and magnitude of lookahead bias, and apply the test to two forecasting tasks: news headlines predicting stock returns and earnings call transcripts predicting capital expenditures. Our test provides a cost-efficient, diagnostic tool for assessing the validity and reliability of LLM-generated forecasts. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.23847 |
| By: | Alina Voronina; Oleksandr Romanko; Ruiwen Cao; Roy H. Kwon; Rafael Mendoza-Arriaga |
| Abstract: | This paper investigates how Large Language Models (LLMs) from leading providers (OpenAI, Google, Anthropic, DeepSeek, and xAI) can be applied to quantitative sector-based portfolio construction. We use LLMs to identify investable universes of stocks within S&P 500 sector indices and evaluate how their selections perform when combined with classical portfolio optimization methods. Each model was prompted to select and weight 20 stocks per sector, and the resulting portfolios were compared with their respective sector indices across two distinct out-of-sample periods: a stable market phase (January-March 2025) and a volatile phase (April-June 2025). Our results reveal a strong temporal dependence in LLM portfolio performance. During stable market conditions, LLM-weighted portfolios frequently outperformed sector indices on both cumulative return and risk-adjusted (Sharpe ratio) measures. However, during the volatile period, many LLM portfolios underperformed, suggesting that current models may struggle to adapt to regime shifts or high-volatility environments underrepresented in their training data. Importantly, when LLM-based stock selection is combined with traditional optimization techniques, portfolio outcomes improve in both performance and consistency. This study contributes one of the first multi-model, cross-provider evaluations of generative AI algorithms in investment management. It highlights that while LLMs can effectively complement quantitative finance by enhancing stock selection and interpretability, their reliability remains market-dependent. The findings underscore the potential of hybrid AI-quantitative frameworks, integrating LLM reasoning with established optimization techniques, to produce more robust and adaptive investment strategies. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.24526 |
| By: | Hongshen Sun; Juanjuan Zhang |
| Abstract: | Large language models (LLMs) are increasingly used to simulate human behavior, but common practices to use LLM-generated data are inefficient. Treating an LLM's output ("model choice") as a single data point underutilizes the information inherent to the probabilistic nature of LLMs. This paper introduces and formalizes "model belief, " a measure derived from an LLM's token-level probabilities that captures the model's belief distribution over choice alternatives in a single generation run. The authors prove that model belief is asymptotically equivalent to the mean of model choices (a non-trivial property) but forms a more statistically efficient estimator, with lower variance and a faster convergence rate. Analogous properties are shown to hold for smooth functions of model belief and model choice often used in downstream applications. The authors demonstrate the performance of model belief through a demand estimation study, where an LLM simulates consumer responses to different prices. In practical settings with limited numbers of runs, model belief explains and predicts ground-truth model choice better than model choice itself, and reduces the computation needed to reach sufficiently accurate estimates by roughly a factor of 20. The findings support using model belief as the default measure to extract more information from LLM-generated data. |
| Date: | 2025–12 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2512.23184 |