Jean-François Bonnefon’s research while affiliated with Toulouse School of Economics and other places

The frontier of artificial intelligence (AI) is constantly moving, raising fears and concerns whenever AI is deployed in a new occupation. Some of these fears are legitimate and should be addressed by AI developers—but others may result from psychological barriers, suppressing the uptake of a beneficial technology. Here, we show that country-level variations across occupations can be predicted by a psychological model at the individual level. Individual fears of AI in a given occupation are associated with the mismatch between psychological traits people deem necessary for an occupation and perceived potential of AI to possess these traits. Country-level variations can then be predicted by the joint cultural variations in psychological requirements and AI potential. We validated this preregistered prediction for six occupations (doctors, judges, managers, care workers, religious workers, and journalists) on a representative sample of 500 participants from each of 20 countries (total N = 10,000). Our findings may help develop best practices for designing and communicating about AI in a principled yet culturally sensitive way, avoiding one-size-fits-all approaches centered on Western values and perceptions.

Climate skepticism remains a significant barrier to public engagement with accurate climate information, because skeptics actively engage in information avoidance to escape exposure to climate facts. Here we show that generative AI can enhance engagement with climate science among skeptical audiences by subtly modifying headlines to align better with their existing perspectives, without compromising factual integrity. In a controlled experiment (N = 2000) using a stylized social media interface, headlines of climate science articles modified by an open-source large language model (Llama3 70B, version 3.0) led to more bookmarks and more upvotes, and these effects were strongest among the most skeptical participants. Skeptics who engaged with climate science as a result of this intervention showed a shift in beliefs toward the scientific consensus by the end of the study. These results show that generative AI can alter the information diet skeptics consume, with the promise that scalable, sustained engagement will promote better epistemic health. They highlight the potential of generative AI as a tool for truth, showing that while it can be misused by bad actors, it also holds promise for advancing public understanding of science when responsibly deployed by well-intentioned actors.

While artificial intelligence (AI) enables significant productivity gains from delegating tasks to machines, it can also facilitate the delegation of unethical behaviour. Here, we demonstrate this risk by having human principals instruct machine agents to perform a task with an incentive to cheat. Principals’ requests for cheating behaviour increased when the interface implicitly afforded unethical conduct: Machine agents programmed via supervised learning or goal specification evoked more cheating than those programmed with explicit rules. Cheating propensity was unaffected by whether delegation was mandatory or voluntary. Given the recent rise of large language model-based chatbots, we also explored delegation via natural language. Here, cheating requests did not vary between human and machine agents, but compliance diverged: When principals intended agents to cheat to the fullest extent, the majority of human agents did not comply, despite incentives to do so. In contrast, GPT4, a state-of-the-art machine agent, nearly fully complied. Our results highlight ethical risks in delegating tasks to intelligent machines, and suggest design principles and policy responses to mitigate such risks.

Conversation with a trained chatbot can reduce conspiratorial beliefs

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.

Humans, aware of the social costs associated with false accusations, are generally hesitant to accuse others of lying. Our study shows how lie detection algorithms disrupt this social dynamic. We develop a supervised machine-learning classifier that surpasses human accuracy and conduct a large-scale incentivized experiment manipulating the availability of this lie-detection algorithm. In the absence of algorithmic support, people are reluctant to accuse others of lying, but when the algorithm becomes available, a minority actively seeks its prediction and consistently relies on it for accusations. Although those who request machine predictions are not inherently more prone to accuse, they more willingly follow predictions that suggest accusation than those who receive such predictions without actively seeking them.

The deployment of AI systems for welfare benefit allocation allows for accelerated decision-making and faster provision of critical help, but has already led to an increase in unfair benefit denials and false fraud accusations. Collecting data in the US and the UK (N = 2 449), we explore the public acceptability of such speed-accuracy trade-offs in populations of claimants and non-claimants. We observe a general willingness to trade off speed gains for modest accuracy losses, but this aggregate view masks notable divergences between claimants and non-claimants. Although welfare claimants comprise a relatively small proportion of the general population (e.g., 20% in the US representative sample), this vulnerable group is much less willing to accept AI deployed in welfare systems, raising concerns that solely using aggregate data for calibration could lead to policies misaligned with stakeholder preferences. Our study further uncovers asymmetric insights between claimants and non-claimants. The latter consistently overestimate claimants' willingness to accept speed-accuracy trade-offs, even when financially incentivized for accurate perspective-taking. This suggests that policy decisions influenced by the dominant voice of non-claimants, however well-intentioned, may neglect the actual preferences of those directly affected by welfare AI systems. Our findings underline the need for stakeholder engagement and transparent communication in the design and deployment of these systems, particularly in contexts marked by power imbalances.

The deployment of AI systems for welfare benefit allocation allows for accelerated decision-making and faster provision of critical help, but has already led to an increase in unfair benefit denials and false fraud accusations. Collecting data in the US and the UK (N = 2449), we explore the public acceptability of such speed-accuracy trade-offs in populations of claimants and non-claimants. We observe a general willingness to trade off speed gains for modest accuracy losses, but this aggregate view masks notable divergences between claimants and non-claimants. Although welfare claimants comprise a relatively small proportion of the general population (e.g., 20% in the US representative sample), this vulnerable group is much less willing to accept AI deployed in welfare systems, raising concerns that solely using aggregate data for calibration could lead to policies misaligned with stakeholder preferences. Our study further uncovers asymmetric insights between claimants and non-claimants. The latter consistently overestimate claimant willingness to accept speed-accuracy trade-offs, even when financially incentivized for accurate perspective-taking. This suggests that policy decisions influenced by the dominant voice of non-claimants, however well-intentioned, may neglect the actual preferences of those directly affected by welfare AI systems. Our findings underline the need for stakeholder engagement and transparent communication in the design and deployment of these systems, particularly in contexts marked by power imbalances.

To succeed, we posit that research cartography will require high-throughput natural description to identify unknown unknowns in a particular design space. High-throughput natural description, the systematic collection and annotation of representative corpora of real-world stimuli, faces logistical challenges, but these can be overcome by solutions that are deployed in the later stages of integrative experiment design.