Incorporating individual-level cognitive priors offers an important route to personalizing neural networks, yet accurately eliciting such priors remains challenging: existing methods either fail to uniquely identify them or introduce systematic biases. Here, we introduce PriorProbe, a novel elicitation approach grounded in Markov Chain Monte Carlo with People that recovers fine-grained, individual-specific priors. Focusing on a facial expression recognition task, we apply PriorProbe to individual participants and test whether integrating the recovered priors with a state-of-the-art neural network improves its ability to predict an individual's classification on ambiguous stimuli. The PriorProbe-derived priors yield substantial performance gains, outperforming both the neural network alone and alternative sources of priors, while preserving the network's inference on ground-truth labels. Together, these results demonstrate that PriorProbe provides a general and interpretable framework for personalizing deep neural networks.

Knowledge sharing is central to strategy and organizational learning, yet its effect on knowledge production remains underexplored. When knowledge diffuses too easily, individuals may free ride on others' costly knowledge production, creating a suboptimal equilibrium in which knowledge sharing persists but the average payoff is no greater than if everyone produced knowledge independently-the so-called Rogers' paradox. We develop a game-theoretic model to re-examine this puzzle. In our baseline model, we reproduce Rogers' paradox; frictionless sharing does not increase performance beyond individual knowledge production alone. We then extend the model to incorporate absorptive capacity-the need for prior investment in one's own knowledge before learning from others. Absorptive capacity discourages pure free riding. Counterintuitively, although it introduces frictions in knowledge sharing, absorptive capacity increases collective knowledge production beyond the level attainable through individual knowledge production alone. This explains why extensive knowledge sharing in domains such as academia does not erode incentives for knowledge production. It also contributes to knowledge-based theories of the firm by showing that although hierarchical control may be crucial in contexts where knowledge is simple and codifiable, it may not be necessary where knowledge is complex and tacit. More broadly, our analysis illustrates how formal modeling can uncover overlooked mechanisms and integrate insights across cultural evolution, organizational learning, and strategy.

Being unpredictable is useful for creativity, exploration, and good decision-making. Decades of research asking people to generate random sequences of numbers have concluded that people systematically deviate from randomness, but individuals do so in very idiosyncratic ways. However, there is no consensus on the cause of these rich individual differences: most theories postulate that people achieve this by spending cognitive resources monitoring their own output and changing the way they say items accordingly, whereas a Local Sampling account postulates that people draw into a general-purpose ability to produce samples, which they would use to make judgments and choices, and which are inherently somewhat unpredictable. Here we distinguish between these possibilities by asking people to generate sequences both at random and as they come to mind, across two experiments at different production speeds, using a non-uniform distribution. We employ several measures of how sequences deviate from randomness. We find that, consistent with the Local Sampling explanation, people deviate from randomness in virtually identical ways in both sequences, with high individual differences pointing to a common cognitive process. We follow up these findings by computationally modelling human performance using several Local Sampling models. We find many participants had the same model as best-fitting in both sequences; with estimated parameters correlating strongly across tasks showing individual differences in how dependent on previous items sampling is. Overall, we conclude that random generation is better understood as employing a general-purpose faculty, Local Sampling, which is stable across time and tasks, with differences in the sampler’s features resulting in differences in random generation performance.

Shiffrin, Stigler, and Keil worry that science frequently falls victim to an illusion of understanding. We take their argument a step further, arguing that psychological scientists often fundamentally mischaracterize human understanding as rooted in internally represented theories of any kind. We propose an alternative, interactive notion of understanding.

Computational process models are pivotal tools for studying the human mind, offering explicit algorithmic accounts of how human behavior arises. However, they are often siloed across individual tasks and domains, lacking a coherent framework for examining computational principles shared across cognition. The challenges are twofold. First, existing models rely heavily on researcher-defined and task-dependent representations of the stimuli. Second, they typically capture only a limited range of behavioral responses. To address these limitations, we propose a two-component framework. A neural network, trained on ecologically valid ground-truth data, functions as an Implicit Bayesian Model (IBM) that maps complex real-world stimuli directly onto posterior probability distributions. A human-like sampling process, the Autocorrelated Bayesian Sampler (ABS), then samples hypotheses from those posteriors to generate diverse behavioral responses while accounting for systematic deviations from normative predictions. This combination of realistic representations with a general-purpose algorithmic decision process enables cognitive mechanisms to be implemented and transferred across tasks at real-world scale. Through three case studies spanning multiple domains and stimulus modalities, we demonstrate the generality of the IBM-ABS framework in explaining human behavior both qualitatively and quantitatively, and explore its capacity to yield new theoretical insights through incorporating realistic priors and predicting the effects of experimental manipulations.

Forecasting from a stream of past observations is a routine judgment task in organizations and markets, and a large literature holds that people systematically overreact when doing so, placing excessive weight on recent observations. Prominent models of expectation formation, grounded in mechanisms such as increased salience for recent signals and limited memory for past information, have predicted overreaction as a near-universal behavioral pattern, successfully capturing prior empirical findings. Here we identify boundary conditions on overreaction in expectation formation. Reanalyzing prior data in which participants viewed the full history of past realizations as a line graph, we confirm that while overreaction is most prevalent, it also varies systematically with task context, diminishing under high persistence and even reversing into underreaction as the process approaches a random walk. To test whether information display format shapes these patterns, we conducted two new experiments in which past realizations were presented sequentially, with only the most recent observation displayed (ruling out visual extrapolation). This design should, under either salience- or memory-based theories, elicit maximal overreaction. However, we find that participants instead exhibited reduced overreaction and, under high persistence, pronounced underreaction. These results establish information display format as a key moderator of forecast bias direction: reducing available history not only attenuates overreaction but can reverse its direction entirely, posing a novel challenge to prevailing theories of expectation formation.

People systematically deviate from rational Bayesian standards in belief updating, displaying biases such as base-rate neglect and conservatism. Two very different types of model aim to explain these biases: simple heuristics and stochastic sampling approximations of the Bayesian solution, like the Bayesian Sampler. However, neither approach alone accounts for the variety of responses. Here we explore a hybrid Heuristic-Anchored Bayesian Sampler (HABS) which integrates simple heuristics with a Bayesian sampler. In this framework, a simple heuristic provides an initial estimate, which may then be refined through a Bayesian sampling process to approximate the ideal Bayesian posterior. Behaviour thus depends on the size of the sample: if no samples are drawn, a heuristic response is produced, but adding samples will introduce noise while increasing average judgment accuracy. Analyses of data from a new experiment (N=200) and re-analysis of data from Stengård et al., 2022 revealed that, in most conditions, this integrative approach outperformed purely heuristic or purely Bayesian models in explaining how people update their beliefs in the medical diagnosis task. Our findings suggest that people flexibly combine mental shortcuts and approximate Bayesian processes, illuminating why some responses appear purely heuristic while others reflect approximate Bayesian updating.

Coordination is central to social life, yet LLMs are poor coordinators. We introduce the Hangman game as a novel experimental paradigm to study why. Our focus is on shared agency, a precondition for coordination, defined as the ability to form shared goals, intentions, and plans. Success in Hangman requires agents to coordinate on word choices across rounds. Human teams reliably develop shared strategies, coordinating on mutually predictable word patterns. LLMs fail to do so. Across Human-Human, Human-LLM, and LLM-LLM teams, humans outperform all tested state-of-the-art LLMs, and humans paired with LLMs perform worse than when paired with other humans. As LLMs are increasingly deployed as collaborative agents, coordination failures undermine their usefulness. Hangman offers a flexible, controlled method for measuring and tracking coordination skills as models develop.

The high level of inequality of income and wealth across individuals, groups, and nations is widely regarded as among the most fundamental problems facing humanity. Yet democracies often elect and reelect politicians who deliver policies that exacerbate rather than reduce inequality. We argue that this disconnect arises from three basic features of human social cognition that evolved for interpersonal interactions in small groups rather than for navigating the vast political, economic, and technological systems that shape contemporary life: a focus on local notions of equality and equity between socially connected individuals rather than across society at large; group identification, which directs attention to inequality between groups rather than "pure" inequality across the broader population; and what we call the "i-frame bias"-a tendency to explain social outcomes, including inequality, as the product of individual behaviors, taking the system within which those individuals operate as a given. We extend our analysis to examine how economic elites exploit these features of social cognition to shape the political and policy landscape and examine the role psychology can play in reducing inequality.

Does the utility of an outcome influence people's assessment of risk and uncertainty? Prominent theories propose that people rely on mental simulation to evaluate probabilities and risky events, yet prior empirical findings is mixed as to whether and how utility biases this process. Across four experiments (total N=206, with Experiment 4 pre-registered), we tested this question using a random generation paradigm, in which participants mentally simulated gamble outcomes and said them out loud. Then we compared these responses with probability judgments and predictions. While we identified individual differences, the majority of participants exhibited neutrality, with no systematic impact of utility on their sampling distributions. Nevertheless, an optimism subgroup emerged, with a larger proportion of optimistic participants when the monetary context was more salient. Additionally, outcome utilities have similar effects on probability judgments, predictions, and random generation tasks, consistent with the proposal that these tasks are related to each other and may all rely on mental simulation. These findings indicate that utility does not uniformly distort mental simulation, but can do so for a stable subgroup under particular motivational contexts, and they motivate cognitive models that capture both unbiased and optimistic forms of mental sampling.

Societal expectations have been found to determine which social roles people should occupy. However, so far, these beliefs have been mainly explored with self-report and response conflict measures where expectation-confirming (vs. violating) judgments elicit faster responding. The present lab study (N = 57) applied a novel approach – the random generation paradigm – to understand how pre-existing social assumptions determine which information is retrieved from memory when prompted by different social categories. Specifically, we asked participants to imagine (hypothetical) people working in certain professions and to say their names out loud. We found that the statistics of the uttered names reflected societal gender stereotypes and environmental statistics of actual people working in these occupations. Importantly, the proportion of female and male names generated for each profession by each participant predicted their performance in a sequential priming task (prime = stereotyped professions, target = female and male faces) better than the environmental statistics or participants’ estimates of gender proportions. Together, these findings offer a new, and widely applicable, method for exploring cultural beliefs and help clarify how social information is sampled from memory when making social judgments.

How has human culture become so complex? We argue that a key process is social tinkering: the gradual accumulation of ad hoc innovations to the social rules that coordinate behavior in response to immediate challenges. Momentary innovations provide precedents that can be reused, entrenched, adapted and recombined to handle future challenges. Interactions between these social rules create rich cultural systems (languages, ethics, political organization) of increasing complexity through processes of spontaneous order, not deliberate design. To explain the historical emergence of cumulative cultural complexity, we distinguish between six overlapping and interacting stages: (1) non-social tinkering to solve problems in the natural world; (2) learning and copying from the tinkering of others; (3) social tinkering involving jointly agreeing on momentary conventions to coordinate interactions, typically for mutual benefit; (4) creating communicative conventions (language) to support more complex social interactions; (5) social tinkering of linguistically-formulated cultural rules leading to laws, organizations, institutions, etc.; and (6) tinkering with linguistically-formulated non-social knowledge, allowing for the creation of science and technology. The rich interplay of innovation across the six stages is crucial for explaining increasing cultural and organizational complexity and our collective mastery of the natural world. Because social and non-social tinkering requires two different kinds of learning, this analysis has important implications for the understanding of human learning and cognition, including moral and evolutionary psychology, theory of mind, and the view of the child-as-scientist. Social tinkering also has substantial implications for current theories of cultural evolution.

How should we identify interesting topics in cognitive science? This paper suggests that one useful research strategy is to hunt for, and attempt to resolve, paradoxes: that is, apparent or real contradictions in our understanding of the mind and of thought. The rationale for this strategy is the assumption that our current thinking, and our various partial theories, of any topic are typically ill-defined, inconsistent or both. Thus, contradictions and confusions abound. Isolating paradoxes helps us expose vagueness and contradictions and demands that we formulate our ideas more precisely. From this point of view, finding a robust and puzzling contradiction in our current thinking should be celebrated as an achievement in itself. Ideally, of course, we then make further progress by clarifying how the paradox may be resolved, by clarifying our theories or finding new data that may decide between inconsistent assumptions. This approach is illustrated through examples from the author's research over several decades, which seems in retrospect to involve a repeated, if largely unwitting, application of this strategy.

How we should design and interact with social artificial intelligence depends on the socio-relational role the AI is meant to emulate or occupy. In human society, relationships such as teacher-student, parent-child, neighbors, siblings, or employer-employee are governed by specific norms that prescribe or proscribe cooperative functions including hierarchy, care, transaction, and mating. These norms shape our judgments of what is appropriate for each partner. For example, workplace norms may allow a boss to give orders to an employee, but not vice versa, reflecting hierarchical and transactional expectations. As AI agents and chatbots powered by large language models are increasingly designed to serve roles analogous to human positions - such as assistant, mental health provider, tutor, or romantic partner - it is imperative to examine whether and how human relational norms should extend to human-AI interactions. Our analysis explores how differences between AI systems and humans, such as the absence of conscious experience and immunity to fatigue, may affect an AI's capacity to fulfill relationship-specific functions and adhere to corresponding norms. This analysis, which is a collaborative effort by philosophers, psychologists, relationship scientists, ethicists, legal experts, and AI researchers, carries important implications for AI systems design, user behavior, and regulation. While we accept that AI systems can offer significant benefits such as increased availability and consistency in certain socio-relational roles, they also risk fostering unhealthy dependencies or unrealistic expectations that could spill over into human-human relationships. We propose that understanding and thoughtfully shaping (or implementing) suitable human-AI relational norms will be crucial for ensuring that human-AI interactions are ethical, trustworthy, and favorable to human well-being.

Noise in behavior is often considered a nuisance: Although the mind aims for the best possible action, it is let down by unreliability in the sensory and response systems. Researchers often represent noise as additive, Gaussian, and independent. Yet a careful look at behavioral noise reveals a rich structure that defies easy explanation. First, in both perceptual and preferential judgments sensory and response noise may potentially play only minor roles, with most noise arising in the cognitive computations. Second, the functional form of the noise is both non-Gaussian and nonindependent, with the distribution of noise being better characterized as heavy-tailed and as having substantial long-range autocorrelations. It is possible that this structure results from brains that are, for some reason, bedeviled by a fundamental design flaw, albeit one with intriguingly distinctive characteristics. Alternatively, noise might not be a bug but a feature. Specifically, we propose that the brain approximates probabilistic inference with a local sampling algorithm, one using randomness to drive its exploration of alternative hypotheses. Reframing cognition in this way explains the rich structure of noise and leads to the surprising conclusion that noise is not a symptom of cognitive malfunction but plays a central role in underpinning human intelligence.

Behaving randomly can be advantageous: it prevents others from capitalizing on patterns in our behavior. Unfortunately, the consensus from sixty years of psychological research is that people cannot do so: when attempting to be random, people’s responses exhibit systematic patterns. Random phenomena, however, are not instantaneously random. They require sufficient time between observations (e.g. the weather) or for enough iterations of a randomizing process to have occurred (e.g. card shuffling). It is unknown whether human sequences can be random if afforded such delays. Here we show that a modest temporal separation between items can make human sequences indistinguishable from random ones. We carried out our own experiment (N = 54) and analyzed ten existing datasets with different production rates, response sets, and response modalities. We found that when the delay between items was between two and four seconds, differences between human and random sequences disappeared. Furthermore, by comparing sequences produced by the same participants at different speeds, we confirmed that when participants make an effort, the needed delay is independent of production rate, akin to the weather. Our results show that people are able to generate randomness, and within a few seconds, giving us an accessible protection against potential exploits.

We present a novel theory of moral cognition organized around resource-rational contractualism. From a contractualist perspective, ideal moral judgments are those that would be agreed to by rational bargaining agents-an idea with widespread support in philosophy, psychology, economics, biology, and cultural evolution. As a practical matter, however, investing time and effort in negotiating every interpersonal interaction is unfeasible. Instead, we propose, people use abstractions and heuristics to efficiently identify mutually beneficial arrangements. We argue that many well-studied elements of our moral minds, such as reasoning about others' utilities ("consequentialist" reasoning) or evaluating intrinsic ethical properties of certain actions ("deontological" reasoning), can be naturally understood as resource-rational approximations of a contractualist ideal. It is widely agreed upon that morality guides people with conflicting interests towards agreements of mutual benefit. We therefore might expect numerous proposals for organizing human moral cognition around the logic of bargaining, negotiation, and agreement. Yet, while "contractualist" ideas play an important role in moral philosophy, they are starkly underrepresented in the field of moral psychology. From a contractualist perspective, ideal moral judgments are those that would be agreed to by rational bargaining agents-an idea with wide-spread support in philosophy, psychology, economics, biology, and cultural evolution. As a practical matter, however, investing time and effort in negotiating every interpersonal interaction is unfeasible. Instead, we propose, people use abstractions and heuristics to efficiently identify mutually beneficial arrangements. We argue that many well-studied elements of our moral minds, such as reasoning about others' utilities ("consequentialist" reasoning) or evaluating intrinsic ethical properties of certain actions ("deontological" reasoning), can be naturally understood as resource-rational approximations of a contractualist ideal. Moreover, this view explains the flexibility of our moral minds-how our moral rules and standards get created, updated and overridden and how we deal with novel cases we have never seen before. Thus, the apparently fragmentary nature of our moral psychology-commonly described in terms of systems in conflict-can be largely unified around the principle of finding mutually beneficial agreements under resource constraint. Our resulting "triple theory" of moral cognition naturally integrates contractualist, consequentialist and deontological concerns.