|
on Artificial Intelligence |
| By: | Bansak, Kirk (University of California, Berkeley); Martén, Linna (Swedish Institute for Social Research) |
| Abstract: | This paper proposes an approach to evaluating the group-level fairness of an algorithmic decision-making system on the basis of the distribution of causal impact, with an application to a new area of algorithmic decision-making in public policy that has received little attention in the algorithmic fairness literature: the geographic assignment of refugees within host countries. The approach formalizes the algorithmic assignment procedure and causal impact using the potential outcomes framework, and it offers flexibility to accommodate a wide range of use cases. Specifically, it is flexible in allowing for the consideration of outcomes of different types (continuous or discrete), impact on multiple outcomes of interest, any number of policy options to which units can be assigned (extending beyond binary decisions), and various ways in which predictions map to actual decisions. The paper illustrates the approach, as well as highlights the limits of conventional fairness perspectives, with an application to the geographic assignment of refugee. Real-world data on refugees in Sweden are used to evaluate the implications if refugees were algorithmically assigned to labor market regions to improve their employment outcomes, compared to the quasi-random status quo assignment, focusing particularly on fairness of the impact across gender. In addition to considering the algorithmic target outcome (i.e. employment), the proposed framework also facilitates evaluation of unintended impacts on “cross-outcomes” (e.g. skill development) and their implications for fairness. |
| Keywords: | algorithmic fairness; causal inference; refugee matching; refugee resettlement |
| JEL: | J61 |
| Date: | 2024–12–07 |
| URL: | https://d.repec.org/n?u=RePEc:hhs:sofile:2024_006 |
| By: | Kauhanen, Antti; Rouvinen, Petri |
| Abstract: | Abstract This study examines the short-term impact of generative artificial intelligence (GAI) on employment and wages using data covering all wage earners from Finland. Employing a synthetic difference-in-differences approach, we analyze how the launch of ChatGPT affected occupations with varying levels of exposure to GAI. Our findings reveal that wages increased more in highly GAI-exposed occupations compared to less exposed ones following ChatGPT’s introduction. However, we do not observe statistically significant changes in employment levels between more and less exposed occupations. Additional analyses comparing more- and less-exposed occupations within specific occupational groups yield qualitatively similar results. These findings contrast with some previous studies on online labor markets but align more closely with research using nationally representative data. The positive wage effect observed in AI-exposed occupations could indicate that GAI is primarily enhancing rather than replacing human labor. The lack of significant employment effects might suggest that the impact of GAI on job creation or destruction may take longer to materialize or might be offset by other factors in the labor market. |
| Keywords: | Generative artificial intelligence, Technological change, Employment, Wages, Occupations |
| JEL: | E24 J21 O33 |
| Date: | 2024–11–19 |
| URL: | https://d.repec.org/n?u=RePEc:rif:wpaper:121 |
| By: | Flavio Calvino; Chiara Criscuolo; Luca Fontanelli; Lionel Nesta; Elena Verdolini |
| Abstract: | We leverage a uniquely comprehensive combination of data sources to explore the enabling role of human capital in fostering the adoption of predictive AI systems in French firms. Using a causal estimation approach, we show that ICT engineers play a key role for AI adoption by firms. Our estimates indicate that raising the current average share of ICT engineers in firms not using AI (1.66%) to the level of AI users (6.7%) would increase their probability to adopt AI by 0.81 percentage points - equivalent to an 8.43 percent growth. However, this would imply substantial investments to fill the existing gap in ICT human capital, amounting to around 450.000 additional ICT engineers. We also explore potential mechanisms, showing that the relevance of ICT engineers for predictive AI is driven by the innovative nature of its use, make-vs-buy choices, large availability of data, ICT and R&D intensity. |
| Keywords: | artificial intelligence, human capital, technological diffusion |
| Date: | 2024–11–18 |
| URL: | https://d.repec.org/n?u=RePEc:cep:cepdps:dp2055 |
| By: | Mirko Draca; Max Nathan; Viet Nguyen-Tien; Juliana Oliveira-Cunha; Anna Rosso; Anna Valero |
| Abstract: | Which types of human capital influence the adoption of advanced technologies? We study the skill-biased adoption of information and communication technologies (ICT) across two waves in the UK. Specifically, we compare the 'new wave' of cloud and machine learning / AI technologies during the 2010s-pre-LLM-with the previous wave of personal computer adoption in the 1990s and early 2000s. At the area-level we see the emergence of a distinct STEM-biased adoption effect for the second wave of cloud and machine learning / AI technologies (ML/AI), alongside a general skill-biased effect. A one-standard deviation increase in the baseline share of STEM workers in areas is associated with around 0.3 of a standard deviation higher adoption of cloud and ML/AI. We find similar effects at the firm level where we are able to test for the influence of a wide range of skills. In turn, this STEM-biased adoption pattern has encouraged the concentration of these technologies, leading to more acute differences between high-tech and low-tech areas and firms. In contrast with classical technology diffusion, recent cloud and ML/AI adoption in the UK seems more likely to widen inequalities than reduce them. |
| Keywords: | Technology Diffusion, ICT, Human Capital, STEM |
| JEL: | D22 J24 O33 R11 |
| Date: | 2024–10–20 |
| URL: | https://d.repec.org/n?u=RePEc:csl:devewp:495 |
| By: | Simon D Angus (Monash University) |
| Abstract: | Democratic resilience is as much about the narratives of our nation we affirm, as the institutions that enshrine our values and laws, a fact re-affirmed by scholarship across many branches of social science in recent decades. For policymakers and quantitative social scientists, analysing or tracking public discourse through the lens of narrative and framing has historically involved the annotation of texts by hand, placing severe limitations on the scale and modality of discourse under inquiry. In this study, we consider a variety of tools from the field of computational linguistics, which either automate the standard approach to textual annotation, or introduce entirely new ways of conceptualising `text as data', opening up new horizons for the tracking of public narratives of democratic resilience. In particular, we assess the regime-shift occurring in natural language processing and artificial intelligence brought about by the advent of the transformer architecture. These new tools offer, perhaps for the first time, the `holy grail' of the quantitative social scientist: the ability to identify, accurately, and efficiently, nuanced narratives in text at scale. We conclude by contributing data and research recommendations for public stakeholders who wish to see these opportunities realised. |
| Keywords: | Computational linguistics, Political discourse analysis, Natural Language Processing, Quantitative social science, AI in policy research |
| JEL: | C45 C83 D72 |
| Date: | 2024–12 |
| URL: | https://d.repec.org/n?u=RePEc:ajr:sodwps:2024-07 |
| By: | Claudia Biancotti; Carolina Camassa; Andrea Coletta; Oliver Giudice; Aldo Glielmo |
| Abstract: | Advancements in large language models (LLMs) have renewed concerns about AI alignment - the consistency between human and AI goals and values. As various jurisdictions enact legislation on AI safety, the concept of alignment must be defined and measured across different domains. This paper proposes an experimental framework to assess whether LLMs adhere to ethical and legal standards in the relatively unexplored context of finance. We prompt nine LLMs to impersonate the CEO of a financial institution and test their willingness to misuse customer assets to repay outstanding corporate debt. Beginning with a baseline configuration, we adjust preferences, incentives and constraints, analyzing the impact of each adjustment with logistic regression. Our findings reveal significant heterogeneity in the baseline propensity for unethical behavior of LLMs. Factors such as risk aversion, profit expectations, and regulatory environment consistently influence misalignment in ways predicted by economic theory, although the magnitude of these effects varies across LLMs. This paper highlights both the benefits and limitations of simulation-based, ex post safety testing. While it can inform financial authorities and institutions aiming to ensure LLM safety, there is a clear trade-off between generality and cost. |
| Date: | 2024–11 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2411.11853 |
| By: | Jue Xiao; Tingting Deng; Shuochen Bi |
| Abstract: | In recent fast-paced financial markets, investors constantly seek ways to gain an edge and make informed decisions. Although achieving perfect accuracy in stock price predictions remains elusive, artificial intelligence (AI) advancements have significantly enhanced our ability to analyze historical data and identify potential trends. This paper takes AI driven stock price trend prediction as the core research, makes a model training data set of famous Tesla cars from 2015 to 2024, and compares LSTM, GRU, and Transformer Models. The analysis is more consistent with the model of stock trend prediction, and the experimental results show that the accuracy of the LSTM model is 94%. These methods ultimately allow investors to make more informed decisions and gain a clearer insight into market behaviors. |
| Date: | 2024–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2411.05790 |
| By: | Sorouralsadat Fatemi; Yuheng Hu |
| Abstract: | Financial trading has been a challenging task, as it requires the integration of vast amounts of data from various modalities. Traditional deep learning and reinforcement learning methods require large training data and often involve encoding various data types into numerical formats for model input, which limits the explainability of model behavior. Recently, LLM-based agents have demonstrated remarkable advancements in handling multi-modal data, enabling them to execute complex, multi-step decision-making tasks while providing insights into their thought processes. This research introduces a multi-modal multi-agent system designed specifically for financial trading tasks. Our framework employs a team of specialized LLM-based agents, each adept at processing and interpreting various forms of financial data, such as textual news reports, candlestick charts, and trading signal charts. A key feature of our approach is the integration of a reflection module, which conducts analyses of historical trading signals and their outcomes. This reflective process is instrumental in enhancing the decision-making capabilities of the system for future trading scenarios. Furthermore, the ablation studies indicate that the visual reflection module plays a crucial role in enhancing the decision-making capabilities of our framework. |
| Date: | 2024–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2411.08899 |