|
on Artificial Intelligence |
| By: | Bernd Irlenbusch (University of Cologne) |
| Abstract: | Artificial intelligence increasingly shapes economic decisions, yet its value depends on whether humans rely on it appropriately. This survey selectively reviews experimental economic evidence (2020 – 2026) on trust in AI, with a focus on privacy, transparency, accountability, fairness, and efficiency. The evidence challenges simple accounts of algorithm aversion or algorithm appreciation. Individuals may underuse beneficial AI because of opacity, autonomy concerns, or institutional distrust, but may also over-rely on deficient systems, disclose excessive data, or delegate responsibility strategically. The survey suggests that trust in AI is best understood as calibrated reliance under informational and institutional constraints. Effective governance should structure informational and institutional environments that help humans calibrate reliance on AI to its actual capabilities, limitations, and social consequences. |
| Keywords: | Trust in AI, calibrated reliance, algorithm aversion, algorithm appreciation, privacy, transparency, accountability, fairness, efficiency |
| JEL: | C90 C91 C92 C93 O33 |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:ajk:ajkdps:417 |
| By: | Alex Chan |
| Abstract: | Participants acted as loan officers deciding whether to approve real $10, 000-loans issued by a private U.S. lender using an AI’s default-risk predictions. When explanations revealed that the AI penalized non-White or female borrowers, participants were more likely to override the AI’s profit-maximizing recommendation. When their bonuses depended on repayment, however, they sought predictions but avoided explanations, consistent with willful ignorance; this effect faded when explanations were framed as purely financial or demographics were hidden. A secondary experiment reveals a novel bias: participants failed to reason contingently and undervalued explanations even when these complemented private information and improved decision accuracy. |
| JEL: | B4 C1 C91 C92 D12 D14 D81 G21 G41 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35240 |
| By: | Domenico Di Prisco (IÉSEG School Of Management [Puteaux]); Silvia Dello Russo (LUISS - Libera Università Internazionale degli Studi Sociali Guido Carli [Roma], LUISS Business School, Università LUISS Guido Carli, Rome, Italy.) |
| Abstract: | Artificial intelligence (AI) technologies promise to transform how people perform tasks and make decisions within organizations. Yet, their impact on human reasoning processes remains poorly understood. When encountering an unexpected AI suggestion, individuals may either attempt to understand the reasoning behind it or blindly accept or reject it. What drives these different reactions, however, remains unexplored. Unpacking these factors is essential to advance our understanding of augmentation and prevent major decision-making failures. This study addresses this gap through three experimental studies. In study 1 we find that, when performing a task, the unexpected failure of one's own frames increases the likelihood of individuals blindly accepting AI suggestions and effortfully trying to explain them. In study 2 we shed light on the underlying reasons for the results, by analyzing qualitative insights. We find that the unexpected failure of frames promotes "problematization pivoting", a phenomenon wherein individuals anchor their reasoning to opaque AI suggestions ignoring other available cues. In study 3, we add evidence of potential negative performance implications associated with effects documented before. Overall, these findings contribute to the literature on human-AI augmentation and sensemaking theory, while also alerting managers and policymakers on the perils associated with AI use. |
| Keywords: | Artificial intelligence Experiment AI opacity Sensemaking theory Human-AI interaction Augmentation |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:hal:journl:hal-05622528 |
| By: | David Arbour; Eli Ben-Michael; Avi Feller; Apoorva Lal; Lo-Hua Yuan |
| Abstract: | Generative AI and large language models can produce realistic predictions of human behavior from rich, unstructured inputs with little to no task-specific training data. Recent work uses these ``digital twin'' predictions to supplement human responses in surveys and experiments. We study the special case of using AI-generated predictions to reduce variance in randomized experiments. We argue that doing so requires no new estimators and that researchers can simply include AI predictions as covariates in standard regression adjustment, analogous to adjusting for a prognostic score. A benefit of this approach is a ``do no harm'' property whereby the adjusted estimator reverts to the unadjusted difference in means when predictions are uninformative. Other methods, such as variants of prediction-powered inference, do not have this guarantee. We provide implementation guidance, including how to obtain continuous scores from discrete LLM outputs and how to use LLMs to featurize unstructured inputs as auxiliary covariates. We demonstrate these ideas in simulations and three empirical applications: a survey mega-study, an email marketing A/B test, and a large-scale technology platform experiment. Overall, efficiency gains are real if modest, with greater benefits in studies that contain substantial text and other unstructured data. We also confirm the do no harm property empirically. Given these gains and limited costs, we recommend adjusting for AI-generated predictions as a regular empirical practice. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.08853 |
| By: | Fangyan Wang; Zaiyan Wei; Yang Wang |
| Abstract: | Generative artificial intelligence (AI) is expected to transform work, but less is known about how firms reorganize labor demand as the technology diffuses. Existing research has largely focused on which occupations are exposed to AI or whether exposed jobs decline. We extend this debate by examining whether firms adjust by changing where they hire, what jobs contain, or both. Using a nationwide dataset of job postings in the United States, covering all sectors of the economy, we construct a dynamic, posting-level measure of generative AI exposure with a two-stage large language model pipeline. The pipeline identifies the tasks described in each posting and classifies the extent to which generative AI can perform or assist them. We then decompose changes in aggregate exposure into two margins: reallocation of demand across jobs and redesign of tasks within jobs. We document three main findings. First, generative AI exposure is dynamic rather than fixed, changing substantially over time. Second, labor demand adjusts through both margins. Hiring reallocation explains the largest share of the aggregate decline in exposure, accounting for 52% on average, while within-job redesign becomes increasingly important, accounting for 39.5%. A complementary Oaxaca-Blinder decomposition shows that shifts in occupational composition account for about 90% of the exposure change attributable to observable job characteristics. Third, adjustment differs across the job ladder. Senior jobs adjust earlier and mainly through reallocation, whereas junior jobs adjust through a broader mix of reallocation, redesign, and their interaction. These findings suggest that labor-market adjustment to generative AI is a process of organizational reconfiguration, in which firms reshape both hiring demand and the task architecture of work. |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2605.23159 |
| By: | Pawel Adrjan (Indeed); Yusuke Aoki (Indeed); Gabriele Ciminelli (Asian Development Bank); Robin Döttling (Erasmus University of Rotterdam); Sílvia Garcia-Mandicó (Asian Development Bank) |
| Abstract: | We study the evolution of digital and artificial intelligence (AI) skill demand across six economies in Asia and the Pacific between 2019 and 2024 using millions of online job postings from Indeed and a large language model to classify them by their required level of digital and AI proficiency. Digital skill requirements are widespread across the occupational distribution and have expanded most rapidly in traditionally low and mid digital jobs, pointing to broad technological diffusion. Jobs requiring higher digital proficiency command substantially higher pay, even when comparing job postings within the same job title and controlling for cognitive, interpersonal, organizational, and technical skill requirements. We estimate wage premia of approximately 4%, 11%, and 26% for basic, intermediate, and advanced digital skills, respectively, with intermediate and advanced digital skills commanding larger premia than many other highly rewarded skills. We also document a sharp increase in demand for AI-related competencies, particularly those related to the use of AI tools. |
| Keywords: | digital skills;artificial intelligence;online job postings;labor market;wage premium;large language models;Asia and the Pacific |
| JEL: | C55 J24 J30 O15 O33 |
| Date: | 2026–06–08 |
| URL: | https://d.repec.org/n?u=RePEc:ris:adbewp:022642 |
| By: | Alex Chan |
| Abstract: | I study medical liability when artificial intelligence acts as a doctor rather than as a passive clinical tool. The central object is the legally usable medical record: the inputs, logs, warnings, prescriptions, follow-up instructions, and outcomes on which courts, contracts, insurers, and regulators can condition responsibility. I show that AI medical liability is an institutional design problem under imperfect legal information. If the record separates AI-controllable error from patient nonadherence and natural disease progression, high-powered AI-fault liability implements the standard accident-law ideal. If the record is coarse, the first best may be infeasible: the same transfer that disciplines the AI also insures the patient's hidden action. With joint causation, the relevant object is a marginal-responsibility score rather than a posterior cause label. I characterize the feasible set of liability incentives generated by the record and show when the optimal rule is no liability, strict liability, negligence, a safe harbor, comparative fault, or a continuous warranty. I then study algorithmic defensive design, through which AI developers can design not only medical recommendations but also the record on which future liability depends. Adoption, learning, enterprise liability, insurance, no-fault compensation, and regulation enter as ways to change the record, the liable entity, or the financing of compensation. The framework yields conditional implications rather than a one-size-fits-all rule. |
| JEL: | D47 D82 D86 D9 G22 I1 I13 I18 K12 K13 L51 |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35321 |
| By: | Dana Golden; Aruna Balasubramanian; Niranjan Balasubramanian |
| Abstract: | Data centers now account for 4.4% of United States electricity demand, yet the grid-level effectiveness of the renewable energy certificates (RECs) and power purchase agreements (PPAs) hyperscalers use to claim carbon neutrality remains unclear. We develop a game-theoretic model in which a data center operator chooses among RECs, PPAs, and behind-the-meter colocation while generators make entry decisions under endogenous financing costs. The model identifies a timing wedge -- the mismatch between consumption and credited renewable generation -- as a central mechanism through which AI demand degrades reliability, raises prices, and increases emissions even when RECs cover 100% of annual consumption. Colocation with storage addresses this wedge directly and induces the greatest renewable entry by eliminating generator revenue risk. We test these predictions by exploiting the staggered release of large language models as a natural experiment, using difference-in-differences on a novel dataset linking AI activity to local grid outcomes. AI demand significantly increases fossil generation, wholesale prices (up to 25% in treated PJM zones), and outage frequency (0.5--1 additional outages per year) near data centers, with impacts scaling in model size. Data centers with on-site generation exhibit a sign reversal in power-quality effects, consistent with the model's prediction that behind-the-meter capacity absorbs demand spikes. Counterfactual analyses show that edge inference, spatial reallocation, and colocated storage each substantially mitigate grid impacts, while REC-only strategies do not. Together, our results demonstrate that the externalities of AI to the grid are tightly coupled to procurement design and the spatial organization of data center infrastructure. |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.00811 |
| By: | Wabenga Yango, James |
| Abstract: | This paper develops a general-equilibrium overlapping-generations model with endogenous fertility, in which firms accumulate both physical and artificial intelligence (AI) capital, and uses it to study the macroeconomic transmission of two structural disturbances: an AI technology shock and a longevity shock. The AI shock acts as a capital-demand disturbance: it raises all rates of return, most sharply the return to AI capital, reallocates investment from physical to AI capital, and produces a frontloaded expansion of output that decays monotonically. The longevity shock acts as a saving-supply disturbance: it deepens the aggregate capital stock, compresses returns and the real interest rate, and generates hump-shaped, persistent dynamics. The two shocks move fertility in opposite directions: AI raises it modestly through an income effect, while longevity lowers it by strengthening the life-cycle saving motive and the opportunity cost of child-rearing. A forecast-error variance decomposition attributes the bulk of the volatility in most aggregate variables to the longevity shock, while the AI shock accounts for the largest share of the variance in the return to AI capital. Fertility is strongly countercyclical and almost perfectly negatively correlated with hours worked, placing the household time-allocation margin at the center of the transmission mechanism. A robustness analysis confirms that these conclusions reflect structural properties of the model: variation in the capital share and in the persistence of the AI shock leaves the signs of the wage, fertility, output, and consumption responses unchanged, and only the labor–AI elasticity of substitution can reverse them, beyond a threshold that lies well above standard empirical estimates. |
| Keywords: | Artificial intelligence; endogenous fertility; longevity; general equilibrium; life-cycle model; capital accumulation; demographic transition. |
| JEL: | E22 E32 J11 J13 J26 O33 O41 |
| Date: | 2026–06–01 |
| URL: | https://d.repec.org/n?u=RePEc:pra:mprapa:129480 |
| By: | t'Serstevens, François; Oschatz, Corinna; , Abdul.Sittar; Trilling, Damian; Guček, Alenka |
| Abstract: | Online social networks have become a central arena for political and social discourse, yet interactions on these platforms are frequently characterized by hostile interactions. While disagreement is a normal and required feature of democratic debate, research suggests that disrespectful communication discourages users from engaging in political discussions and may negatively affect both participants and the broader audience exposed to such interactions. In response, previous interventions have attempted to improve online discourse through behavioral nudges and interface design changes, though their effectiveness has often been limited. This study examines whether AI-mediated paraphrasing interventions can reduce uncivil expression while preserving substantive disagreement in online political discussions. Using an experimental setting that simulates social media interactions, we analyse how AI-generated paraphrases influence the tone of conversations and assess their effects not only on the direct participants of a debate but also on external observers who encounter these exchanges. The findings provide insights into the potential of AI-assisted communication tools to foster healthier online discourse. |
| Date: | 2026–05–31 |
| URL: | https://d.repec.org/n?u=RePEc:osf:socarx:jhbuf_v1 |
| By: | Qianan Wang; Zen Chen |
| Abstract: | The rapid expansion of artificial intelligence (AI) investment has revived a recurrent question in financial economics: are AI-related assets experiencing a bubble, or is the market capitaliz- ing a durable general-purpose technology? This paper develops a hybrid review and diagnostic framework for evaluating whether AI is in an ongoing financial bubble as of May 2026. The analysis begins from asset-pricing foundations in state prices, stochastic discount factors, martingale valuation, and pricing kernels, then connects these foundations to rational bubbles, behavioral bubbles, technology manias, and modern econometric bubble-detection methods. Current evidence shows both genuine fundamentals and bubble-like fragilities. On the fundamental side, realized revenue growth, enterprise adoption, and productivity evidence support a nontrivial share of AI valuations. On the fragile side, capital expenditure has accelerated faster than observed monetization in some layers, private- market valuations are concentrated in a small number of firms, and investor narratives often capitalize future productivity gains before they have appeared in cash flows. The paper proposes a five-pillar diagnostic framework that combines fundamental valuation, residual-exuberance tests, SADF/GSADF explosive-root procedures, LPPL/HLPPL price-pattern diagnostics, sen- timent and issuance measures, and capex-payback analysis. The central conclusion is that AI is best understood as a real technological revolution with localized bubble dynamics rather than as either a pure speculative mania or a bubble-free productivity miracle. |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.01575 |
| By: | Yuqi Li; Siyuan Liu; Bingjun Liu |
| Abstract: | While deep learning has excelled in various domains, its application to sequential decision-making in finance remains challenging due to the low Signal-to-Noise Ratio (SNR) and non-stationarity of financial data. Leveraging the reasoning capabilities of Large Language Models (LLMs), we propose \textbf{PandaAI}, a closed-loop neuro-symbolic LLM agent with market regime modeling and constrained alpha generation, which bridges general LLM reasoning with financial rigor and suppresses the financial toxicity of LLM-generated outputs. To bridge the gap between general linguistic capability and financial rigor, we fine-tune a domain-specific LLM. Furthermore, we integrate this LLM into a modular architecture and form a closed-loop system. Unlike traditional models that optimize isolated prediction metrics, \textbf{PandaAI} is designed as a neuro-symbolic agent that navigates the complex, real-world financial environment with explicit risk awareness. Extensive experiments on CSI 300 stock data show that \textbf{PandaAI} achieves a $18.2\%$ higher Rank IC and $25.7\%$ lower maximum drawdown than state-of-the-art time-series models. Our constrained LLM generation and dual-channel adaptation method provide a general paradigm for LLM deployment in high-stakes sequential decision-making scenarios. |
| Date: | 2026–06 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.06823 |
| By: | Shuang Chen; Clemens Sialm; David X. Xu |
| Abstract: | We examine the growth and performance of AI-driven investing. Using investment advisers' regulatory disclosures, labor market data, and fund strategy descriptions, we document that AI-driven investing has grown steadily since the early 2010s and is concentrated among hedge funds. AI hedge funds outperformed non-AI hedge funds in the early years, but this outperformance declined over time, even among early adopters. Contrary to concerns about AI-driven strategy homogeneity, AI hedge funds exhibit lower return comovement than non-AI peers. Our findings highlight both the alpha-generating potential and the limitations of AI as a source of investment performance. |
| JEL: | G11 G23 G24 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35273 |
| By: | Eugene Park |
| Abstract: | We study how frontier large language models (LLMs) behave as financial forecasters during boom-bust market cycles when made progressively aware of Soros's theory of reflexivity. Standard AI-assisted forecasting treats the market as an exogenous system. Reflexivity theory holds otherwise: prices shape fundamentals, and every forecaster is a participative agent in the loop it analyzes. We evaluate three frontier models - GPT5, Claude Sonnet 4.6, and Gemini 3 Pro - under four accumulating zero-shot conditions across two historically distinct episodes: the dot-com bubble (1996-2001) and the global financial crisis (2004-2009). The primary metric is directional forecasting accuracy; we also report the Sharpe ratio of an implied long/cash strategy to capture the risk-adjusted economic value of the forecasts. All inputs are anonymized and normalized to guard against memorization. We find that conditions incorporating reflexivity awareness improve forecasting accuracy differently across models and context windows, revealing that the same theoretical awareness can produce qualitatively different forecasting behavior across frontier LLMs. |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2606.00061 |
| By: | Christopher Clayton; Antonio Coppola |
| Abstract: | We study whether AI methods applied to large-scale portfolio holdings data can improve macroprudential financial regulation. We build a graph-based deep learning model tailored to security-level data on the holdings of financial intermediaries. The architecture incorporates economic priors and learns latent representations of both assets and investors from the network structure of portfolio positions. Applied to the universe of non-bank financial intermediaries, covering nearly $40 trillion in wealth, the model substantially outperforms existing approaches in out-of-sample forecasts of intermediary trading behavior, including in crisis episodes. The model has more than ten times the explanatory power for the cross-sectional variation in asset returns during stress events compared to traditional approaches, and it outperforms existing systemic risk metrics at the institution level. Its learned representations show that the holdings network encodes rich, economically interpretable information about fire- sale vulnerability. The architecture is fully inductive, producing informative estimates even when entire asset classes or investors are withheld from training. We embed our empirical approach into a macroprudential optimal policy framework to formalize why these objects matter for policy and welfare. We show that even in an equilibrium environment subject to the Lucas critique, the predictive information from the model improves welfare by sharpening the cross-sectional targeting of policy interventions, and we demonstrate a complementarity between prediction and structural knowledge. |
| JEL: | C4 G1 G2 |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35227 |
| By: | Ferrara, Andreas (University of Pittsburgh, Department of Economics, and NBER) |
| Abstract: | Large language models (LLMs) are lowering the entry barriers to working with exciting data sources that used to require strong data science skills, such as handwritten ledgers, text, images, or sound recordings. This guide provides an introduction for researchers who are new to LLMs. It sets out a step-by-step workflow for turning a research idea into working code and data, and describes the four main ways of interacting with an LLM: the chat window, editor-integrated assis tants, agentic coding tools, and the API. It then works through the decisions a practitioner meets in sequence, beginning with whether an LLM is the right tool and whether the data are allowed to be sent to one, then how to select models, write prompts, manage context limits, and control costs, and finally how to validate, reproduce, document, and correct LLM-generated measures in regression settings. A review of recent research shows how these tools already extract, link, har monize, and classify historical data at scale. Four worked examples with replication files illustrate the use of LLMs. They classify emotions in paintings, link census records without names, measure newspaper salience and sentiment around the 1882 Chinese Exclusion Act, and score the emotional delivery of Franklin D. Roosevelt's wartime speeches. The guide also condenses the workflow, the best-practice recommendations, and the preparation of replication packages into summary tables and checklists to aid applied economists. |
| Keywords: | Large Language Models, Artificial Intelligence, Economic History, Practitioner's Guide JEL Classification: C8, N0, C55 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:cge:wacage:810 |