Generative artificial intelligence (AI) 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.

Experiments using economic games are becoming a major source for the study of human social behavior. These experiments are usually conducted with university students who voluntarily choose to participate. Across the natural and social sciences, there is some concern about how this “particular” subject pool may systematically produce biased results. Focusing on social preferences, this study employs data from a survey-experiment conducted with a representative sample of a city's population (N = 765). We report behavioral data from five experimental decisions in three canonical games: dictator, ultimatum and trust games. The dataset includes students and non-students as well as volunteers and non-volunteers. We separately examine the effects of being a student and being a volunteer on behavior, which allows a ceteris paribus comparison between self-selected students (students*volunteers) and the representative population. Our results suggest that self-selected students are an appropriate subject pool for the study of social behavior.

In an anonymous 4-person economic game, participants contributed more money to a common project (i.e., cooperated) when required to decide quickly than when forced to delay their decision (Rand, Greene & Nowak, 2012), a pattern consistent with the social heuristics hypothesis proposed by Rand and colleagues. The results of studies using time pressure have been mixed, with some replication attempts observing similar patterns (e.g., Rand et al., 2014) and others observing null effects (e.g., Tinghög et al., 2013; Verkoeijen & Bouwmeester, 2014). This Registered Replication Report (RRR) assessed the size and variability of the effect of time pressure on cooperative decisions by combining 21 separate, preregistered replications of the critical conditions from Study 7 of the original article (Rand et al., 2012). The primary planned analysis used data from all participants who were randomly assigned to conditions and who met the protocol inclusion criteria (an intent-to-treat approach that included the 65.9% of participants in the time-pressure condition and 7.5% in the forced-delay condition who did not adhere to the time constraints), and we observed a difference in contributions of −0.37 percentage points compared with an 8.6 percentage point difference calculated from the original data. Analyzing the data as the original article did, including data only for participants who complied with the time constraints, the RRR observed a 10.37 percentage point difference in contributions compared with a 15.31 percentage point difference in the original study. In combination, the results of the intent-to-treat analysis and the compliant-only analysis are consistent with the presence of selection biases and the absence of a causal effect of time pressure on cooperation.