Source-linked AI summary
The impact of generative artificial intelligence on socioeconomic inequalities and policy making
Valerio Capraro, Austin Lentsch, Daron Acemoglu, Selin Akgun, Aisel Akhmedova, Ennio Bilancini, Jean-François Bonnefon, Pablo Brañas-Garza, Luigi Butera, Karen M. Douglas, Jim A. C. Everett, Gerd Gigerenzer, Christine Greenhow, Daniel A. Hashimoto, Julianne Holt-Lunstad, Jolanda Jetten, Simon Johnson, Chiara Longoni, Pete Lunn, Simone Natale, Iyad Rahwan, Neil Selwyn, Vivek Singh, Siddharth Suri, Jennifer Sutcliffe, Joe Tomlinson, Sander van der Linden, Paul A. M. Van Lange, Friederike Wall, Jay J. Van Bavel, Riccardo Viale
TL;DR
Generative AI may both worsen and mitigate socioeconomic inequalities across information, work, education, healthcare, and policymaking. This article synthesizes interdisciplinary research, identifies gaps and trade-offs, and finds that existing EU, US, and UK regulatory approaches do not adequately address these challenges.
Problem
Generative AI creates mixed socioeconomic effects, including misinformation, uneven workplace benefits, a widening digital divide, and potentially deeper healthcare inequalities.
Method
The article provides a state-of-the-art interdisciplinary overview, evaluates existing research across information-intensive domains, identifies gaps, and recommends research directions and policies.
Results
Existing regulatory approaches in the EU, US, and UK sometimes fail to adequately address generative AI’s socioeconomic challenges.
Takeaways & Limitations
A dynamic regulatory framework is needed to keep pace with rapid AI advances while harnessing benefits and mitigating risks.
Takeaways & Limitations
AI companionship may reduce loneliness, but overreliance can inhibit human interaction and increase isolation without fulfilling deep social needs.
Abstract
from arXiv · showhide
Generative artificial intelligence has the potential to both exacerbate and ameliorate existing socioeconomic inequalities. In this article, we provide a state-of-the-art interdisciplinary overview of the potential impacts of generative AI on (mis)information and three information-intensive domains: work, education, and healthcare. Our goal is to highlight how generative AI could worsen existing inequalities while illuminating how AI may help mitigate pervasive social problems. In the information domain, generative AI can democratize content creation and access, but may dramatically expand the production and proliferation of misinformation. In the workplace, it can boost productivity and create new jobs, but the benefits will likely be distributed unevenly. In education, it offers personalized learning, but may widen the digital divide. In healthcare, it might improve diagnostics and accessibility, but could deepen pre-existing inequalities. In each section we cover a specific topic, evaluate existing research, identify critical gaps, and recommend research directions, including explicit trade-offs that complicate the derivation of a priori hypotheses. We conclude with a section highlighting the role of policymaking to maximize generative AI's potential to reduce inequalities while mitigating its harmful effects. We discuss strengths and weaknesses of existing policy frameworks in the European Union, the United States, and the United Kingdom, observing that each fails to fully confront the socioeconomic challenges we have identified. We propose several concrete policies that could promote shared prosperity through the advancement of generative AI. This article emphasizes the need for interdisciplinary collaborations to understand and address the complex challenges of generative AI.
Regulation of AI
The paper argues that existing AI regulatory approaches in the EU, US, and UK insufficiently address socioeconomic inequality. It recommends policies spanning misinformation, government expertise, taxation, worker and consumer voice, human-complementary research, and professional training.
- Regulation of AI: Current regulatory responses in the EU, US, and UK differ in structure but do not sufficiently address socioeconomic inequalities.The EU uses a risk-based AI Act, the US relies on fragmented federal and state initiatives, and the UK favors non-statutory soft principles.
- Regulation of AI: A more symmetric tax structure should reduce incentives that favor algorithmic automation over hiring and training labor.The proposed approach would address the lower labor share associated with tax advantages for capital investment.
- Regulation of AI: Policies should strengthen worker and civil-society voice, consumer control over information, human-complementary research, and professional AI training.The recommendations include data unions, support for AI that augments human expertise, and training on capabilities, limitations, ethics, and technical skills.
- Regulation of AI: Governments should develop tools, standards, and educational campaigns to combat AI-generated misinformation.The recommendations cover AI-generated text, images, audio, and video while improving public fact-checking strategies.
- Regulation of AI: AI expertise within government and consultative bodies could support more timely and effective regulatory decision-making.The paper points to EU, UK, and US initiatives as examples of developing this capacity.
Regulation using AI
The paper examines generative AI as a potential policymaking aid while emphasizing alignment and implementation problems. It concludes that AI’s mixed effects require research, dynamic regulation, and supporting institutions capable of monitoring alignment and regulatory change.
- Regulation using AI: Generative AI could support policymaking by analyzing data, recognizing patterns, forecasting trends, and simulating policy outcomes.The paper also stresses that ethical and practical concerns may be prohibitive with current tools.
- Regulation using AI: Chatbots can simulate human behavior in decision-making contexts, but GPT-4 underestimates self-interest and inequity aversion while overestimating altruism.These findings complicate the use of language models as decision-making assistants.
- Regulation using AI: Aligning generative AI with culturally diverse and competing human values is challenging, especially for policy recommendations.The paper identifies non-maleficence, justice, and cultural sensitivity as high-level alignment constraints.
- Regulation using AI: Alignment procedures based on homogeneous or unrepresentative informants can produce socially biased AI outputs, even when more diverse preferences are used.The paper discusses balanced human feedback, consensus-making chatbots, and ecosystems of value-diverse chatbots as alternatives.
- Regulation using AI: Policymakers often lack the expertise to evaluate AI systems’ embedded preferences and systematic biases, creating an implementation problem.The paper recommends supporting organizations that frequently evaluate alignment with legal requirements and signal regulatory changes to companies.
- Regulation using AI: The paper finds that generative AI may democratize information and improve outcomes while also creating misinformation, access, inequality, and human-interaction concerns.It calls for urgent research questions and dynamic regulatory frameworks that keep pace with technological change.
Supplementary Information
The supplementary information lays out example research questions and experimental designs across information, work, education, and healthcare, while emphasizing trade-offs that complicate predictions. It also identifies policy and organizational questions concerning access, competition, privacy, misinformation, cooperation, creativity, and clinical reliability.
- Accessibility: AI accessibility research examines audio descriptions, mobility assistants, and summarization tools for users with visual, physical, or cognitive disabilities.Proposed comparisons assess comprehension, enjoyment, navigation, independence, and mobility with and without AI support.
- Competition and policy: Competition research asks whether open-source AI, regulation, and venture capital can help smaller firms innovate against large firms without sacrificing growth or sustainable innovation.The proposed designs compare open-source and proprietary firms, evaluate regulatory frameworks, and relate venture funding to innovation and market performance.
- Privacy and data governance: Privacy and personalization research tests transparency, anonymization, and consent models because stronger safeguards may reduce misuse while limiting data sharing or personalization.The proposed outcomes include privacy breaches, data sharing, trust, consent quality, and personalized content filtering.
- Information integrity: Misinformation research considers detection, verification, counter-misinformation dialogue, and user education, while recognizing that subtle AI content, misclassification, and reduced platform trust complicate interventions.The agenda includes distinguishing human- and AI-generated text and testing verification in one-to-one communications.
- Cross-domain trade-offs: Across education, healthcare, cooperation, creativity, and social interaction, proposed benefits remain conditional because AI may reinforce biases, omit critical details, reduce originality, weaken human cooperation, or increase isolation.The cited trade-offs include algorithmic bias, oversimplification, clinical overreliance, and insufficient genuine human interaction.