Machine behaviour.

Machines powered by artificial intelligence increasingly mediate our social, cultural, economic and political interactions. Understanding the behaviour of artificial intelligence systems is essential to our ability to control their actions, reap their benefits and minimize their harms. Here we argue that this necessitates a broad scientific research agenda to study machine behaviour that incorporates and expands upon the discipline of computer science and includes insights from across the sciences. We first outline a set of questions that are fundamental to this emerging field and then explore the technical, legal and institutional constraints on the study of machine behaviour.

The Moral Psychology of Artificial Intelligence.

Moral psychology was shaped around three categories of agents and patients: humans, other animals, and supernatural beings. Rapid progress in artificial intelligence has introduced a fourth category for our moral psychology to deal with: intelligent machines. Machines can perform as moral agents, making decisions that affect the outcomes of human patients or solving moral dilemmas without human supervision. Machines can be perceived as moral patients, whose outcomes can be affected by human decisions, with important consequences for human-machine cooperation. Machines can be moral proxies that human agents and patients send as their delegates to moral interactions or use as a disguise in these interactions. Here we review the experimental literature on machines as moral agents, moral patients, and moral proxies, with a focus on recent findings and the open questions that they suggest.

The Moral Machine experiment.

With the rapid development of artificial intelligence have come concerns about how machines will make moral decisions, and the major challenge of quantifying societal expectations about the ethical principles that should guide machine behaviour. To address this challenge, we deployed the Moral Machine, an online experimental platform designed to explore the moral dilemmas faced by autonomous vehicles. This platform gathered 40 million decisions in ten languages from millions of people in 233 countries and territories. Here we describe the results of this experiment. First, we summarize global moral preferences. Second, we document individual variations in preferences, based on respondents' demographics. Third, we report cross-cultural ethical variation, and uncover three major clusters of countries. Fourth, we show that these differences correlate with modern institutions and deep cultural traits. We discuss how these preferences can contribute to developing global, socially acceptable principles for machine ethics. All data used in this article are publicly available.

Citizens from 13 countries share similar preferences for COVID-19 vaccine allocation priorities.

How does the public want a COVID-19 vaccine to be allocated? We conducted a conjoint experiment asking 15,536 adults in 13 countries to evaluate 248,576 profiles of potential vaccine recipients who varied randomly on five attributes. Our sample includes diverse countries from all continents. The results suggest that in addition to giving priority to health workers and to those at high risk, the public favors giving priority to a broad range of key workers and to those with lower income. These preferences are similar across respondents of different education levels, incomes, and political ideologies, as well as across most surveyed countries. The public favored COVID-19 vaccines being allocated solely via government programs but were highly polarized in some developed countries on whether taking a vaccine should be mandatory. There is a consensus among the public on many aspects of COVID-19 vaccination, which needs to be taken into account when developing and communicating rollout strategies.

Discovering the unknown unknowns of research cartography with high-throughput natural description.

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.

Universals and variations in moral decisions made in 42 countries by 70,000 participants.

When do people find it acceptable to sacrifice one life to save many? Cross-cultural studies suggested a complex pattern of universals and variations in the way people approach this question, but data were often based on small samples from a small number of countries outside of the Western world. Here we analyze responses to three sacrificial dilemmas by 70,000 participants in 10 languages and 42 countries. In every country, the three dilemmas displayed the same qualitative ordering of sacrifice acceptability, suggesting that this ordering is best explained by basic cognitive processes rather than cultural norms. The quantitative acceptability of each sacrifice, however, showed substantial country-level variations. We show that low relational mobility (where people are more cautious about not alienating their current social partners) is strongly associated with the rejection of sacrifices for the greater good (especially for Eastern countries), which may be explained by the signaling value of this rejection. We make our dataset fully available as a public resource for researchers studying universals and variations in human morality.

Moral artificial intelligence and machine puritanism.

Puritanism may evolve into a technological variant based on norms of delegation of actions and perceptions to artificial intelligence. Instead of training self-control, people may be expected to cede their agency to self-controlled machines. The cost-benefit balance of this machine puritanism may be less aversive to wealthy individualistic democracies than the old puritanism they have abandoned.

Trustworthiness perception at zero acquaintance: Consensus, accuracy, and prejudice.

Research on trustworthiness perception from faces has unfolded in a way that is strikingly reminiscent of Jussim's narrative in his 2012 book. Jussim's analysis warns us against overemphasizing evidence about prejudice over evidence about accuracy, when both are scant; and reminds us to hold all accounts to the same standards, whether they call on societal biases or true signals.

Eye movements disrupt episodic future thinking.

Remembering the past and imagining the future both rely on complex mental imagery. We considered the possibility that constructing a future scene might tap a component of mental imagery that is not as critical for remembering past scenes. Whereas visual imagery plays an important role in remembering the past, we predicted that spatial imagery plays a crucial role in imagining the future. For the purpose of teasing apart the different components underpinning scene construction in the two experiences of recalling episodic memories and shaping novel future events, we used a paradigm that might selectively affect one of these components (i.e., the spatial). Participants performed concurrent eye movements while remembering the past and imagining the future. These concurrent eye movements selectively interfere with spatial imagery, while sparing visual imagery. Eye movements prevented participants from imagining complex and detailed future scenes, but had no comparable effect on the recollection of past scenes. Similarities between remembering the past and imagining the future are coupled with some differences. The present findings uncover another fundamental divergence between the two processes.

Eye movements disrupt spatial but not visual mental imagery.

It has long been known that eye movements are functionally involved in the generation and maintenance of mental images. Indeed, a number of studies demonstrated that voluntary eye movements interfere with mental imagery tasks (e.g., Laeng and Teodorescu in Cogn Sci 26:207-231, 2002). However, mental imagery is conceived as a multifarious cognitive function with at least two components, a spatial component and a visual component. The present study investigated the question of whether eye movements disrupt mental imagery in general or only its spatial component. We present data on healthy young adults, who performed visual and spatial imagery tasks concurrently with a smooth pursuit. In line with previous literature, results revealed that eye movements had a strong disruptive effect on spatial imagery. Moreover, we crucially demonstrated that eye movements had no disruptive effect when participants visualized the depictive aspects of an object. Therefore, we suggest that eye movements serve to a greater extent the spatial than the visual component of mental imagery.

Generative AI as a tool for truth.

Conversation with a trained chatbot can reduce conspiratorial beliefs.

Lie detection algorithms disrupt the social dynamics of accusation behavior.

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 impact of generative artificial intelligence on socioeconomic inequalities and policy making.

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.

Low second-to-fourth digit ratio predicts indiscriminate social suspicion, not improved trustworthiness detection.

Testosterone administration appears to make individuals less trusting, and this effect has been interpreted as an adaptive adjustment of social suspicion, that improved the accuracy of trusting decisions. Here, we consider another possibility, namely that testosterone increases the subjective cost of being duped, decreasing the propensity to trust without improving the accuracy of trusting decisions. In line with this hypothesis, we show that second-to-fourth digit ratio (2D:4D, a proxy for effects of testosterone in the foetus) correlates with the propensity to trust, but not with the accuracy of trusting decisions. Trust game players (n = 144) trusted less when they had lower 2D:4D (high prenatal testosterone), but their ability to detect the strategy of other players was constant (and better than chance) across all levels of digit ratio. Our results suggest that early prenatal organizing effects of testosterone in the foetus might impair rather than boost economic outcomes, by promoting indiscriminate social suspicion.

Can mutualistic morality predict how individuals deal with benefits they did not deserve?

An individual obtains an unfair benefit and faces the dilemma of either hiding it (to avoid being excluded from future interactions) or disclosing it (to avoid being discovered as a deceiver). In line with the target article, we expect that this dilemma will be solved by a fixed individual strategy rather than a case-by-case rational calculation.

Humans feel too special for machines to score their morals.

Artificial intelligence (AI) can be harnessed to create sophisticated social and moral scoring systems-enabling people and organizations to form judgments of others at scale. However, it also poses significant ethical challenges and is, subsequently, the subject of wide debate. As these technologies are developed and governing bodies face regulatory decisions, it is crucial that we understand the attraction or resistance that people have for AI moral scoring. Across four experiments, we show that the acceptability of moral scoring by AI is related to expectations about the quality of those scores, but that expectations about quality are compromised by people's tendency to see themselves as morally peculiar. We demonstrate that people overestimate the peculiarity of their moral profile, believe that AI will neglect this peculiarity, and resist for this reason the introduction of moral scoring by AI.

Trust within human-machine collectives depends on the perceived consensus about cooperative norms.

With the progress of artificial intelligence and the emergence of global online communities, humans and machines are increasingly participating in mixed collectives in which they can help or hinder each other. Human societies have had thousands of years to consolidate the social norms that promote cooperation; but mixed collectives often struggle to articulate the norms which hold when humans coexist with machines. In five studies involving 7917 individuals, we document the way people treat machines differently than humans in a stylized society of beneficiaries, helpers, punishers, and trustors. We show that a different amount of trust is gained by helpers and punishers when they follow norms over not doing so. We also demonstrate that the trust-gain of norm-followers is associated with trustors' assessment about the consensual nature of cooperative norms over helping and punishing. Lastly, we establish that, under certain conditions, informing trustors about the norm-consensus over helping tends to decrease the differential treatment of both machines and people interacting with them. These results allow us to anticipate how humans may develop cooperative norms for human-machine collectives, specifically, by relying on already extant norms in human-only groups. We also demonstrate that this evolution may be accelerated by making people aware of their emerging consensus.