A foundation model to predict and capture human cognition

Binz, M., Akata, E., et al. (2025).

Nature.

https://doi.org/10.1038/s41586-025-09215-4

Abstract

Establishing a unified theory of cognition has been an important goal in psychology. A first step towards such a theory is to create a computational model that can predict human behaviour in a wide range of settings. Here we introduce Centaur, a computational model that can predict and simulate human behaviour in any experiment expressible in natural language. We derived Centaur by fine-tuning a state-of-the-art language model on a large-scale dataset called Psych-101. Psych-101 has an unprecedented scale, covering trial-by-trial data from more than 60,000 participants performing in excess of 10,000,000 choices in 160 experiments. Centaur not only captures the behaviour of held-out participants better than existing cognitive models, but it also generalizes to previously unseen cover stories, structural task modifications and entirely new domains. Furthermore, the model’s internal representations become more aligned with human neural activity after fine-tuning. Taken together, our results demonstrate that it is possible to discover computational models that capture human behaviour across a wide range of domains. We believe that such models provide tremendous potential for guiding the development of cognitive theories, and we present a case study to demonstrate this.

The article is here.

Here are some thoughts:

This article is important because it introduces Centaur, a novel computational model that represents a major step toward a unified theory of cognition. By fine-tuning a large language model on a vast dataset of human behavior, the researchers created a model with superior predictive power that can generalize across different cognitive domains. This model not only outperforms existing, specialized cognitive models but also demonstrates an alignment with human neural activity, suggesting it captures fundamental principles of human thought. Ultimately, the paper proposes that Centaur can serve as a powerful tool for scientific discovery, guiding the development and refinement of new psychological theories.

Cognitive computational model reveals repetition bias in a sequential decision-making task

Legler, E., Rivera, D. C.,  et al. (2025).

Communications Psychology, 3(1).

https://doi.org/10.1038/s44271-025-00271-0

Abstract

Humans tend to repeat action sequences that have led to reward. Recent computational models, based on a long-standing psychological theory, suggest that action selection can also be biased by how often an action or sequence of actions was repeated before, independent of rewards. However, empirical support for such a repetition bias effect in value-based decision-making remains limited. In this study, we provide evidence of a repetition bias for action sequences using a sequential decision-making task (N = 70). Through computational modeling of choices, we demonstrate both the learning and influence of a repetition bias on human value-based decisions. Using model comparison, we find that decisions are best explained by the combined influence of goal-directed reward seeking and a tendency to repeat action sequences. Additionally, we observe significant individual differences in the strength of this repetition bias. These findings lay the groundwork for further research on the interaction between goal-directed reward seeking and the repetition of action sequences in human decision making.

Here are some thoughts:

This research on "repetition bias in a sequential decision-making task" offers valuable insights for psychologists, impacting both their own professional conduct and their understanding of patient behaviors. The study highlights that human decision-making is not solely driven by the pursuit of rewards, but also by an unconscious tendency to repeat previous action sequences. This finding suggests that psychologists, like all individuals, may be influenced by these ingrained patterns in their own practices, potentially leading to a reliance on familiar methods even when alternative, more effective approaches might exist. An awareness of this bias can foster greater self-reflection, encouraging psychologists to critically evaluate their established routines and adapt their strategies to better serve patient needs.

Furthermore, this research provides a crucial framework for understanding repetitive behaviors in patients. By demonstrating the coexistence of repetition bias with goal-directed reward seeking, the study helps explain why individuals might persist in actions that are not directly rewarding or may even be detrimental, a phenomenon often observed in conditions like obsessive-compulsive disorder or addiction. This distinction between the drivers of behavior can aid psychologists in more accurate patient assessment, allowing them to discern whether a patient's repetitive actions stem from a strong, non-reward-driven bias or from deliberate, goal-oriented choices. The research also notes significant individual differences in the strength of this bias, implying the need for personalized treatment approaches. Moreover, the study's suggestion that frequent repetition contributes to habit formation by diminishing goal-directed control offers insights into how maladaptive habits develop and how interventions can be designed to disrupt these cycles or bolster conscious control.

Confidence in Moral Decision-Making

Schooler, L.,  et al. (2024).

Collabra Psychology, 10(1).

https://doi.org/10.1525/collabra.121387

Abstract

Moral decision-making typically involves trade-offs between moral values and self-interest. While previous research on the psychological mechanisms underlying moral decision-making has primarily focused on what people choose, less is known about how an individual consciously evaluates the choices they make. This sense of having made the right decision is known as subjective confidence. We investigated how subjective confidence is constructed across two moral contexts. In Study 1 (240 U.S. participants from Amazon Mechanical Turk, 81 female), participants made hypothetical decisions between choices with monetary profits for themselves and physical harm for either themselves or another person. In Study 2 (369 U.S. participants from Prolific, 176 female), participants made incentive-compatible decisions between choices with monetary profits for themselves and monetary harm for either themselves or another person. In both studies, each choice was followed by a subjective confidence rating. We used a computational model to obtain a trial-by-trial measure of participant-specific subjective value in decision-making and related this to subjective confidence ratings. Across all types of decisions, confidence was positively associated with the absolute difference in subjective value between the two options. Specific to the moral decision-making context, choices that are typically seen as more blameworthy – i.e., causing more harm to an innocent person to benefit oneself – suppressed the effects of increasing profit on confidence, while amplifying the dampening effect of harm on confidence. These results illustrate some potential cognitive mechanisms underlying subjective confidence in moral decision-making and highlighted both shared and distinct cognitive features relative to non-moral value-based decision-making.

Here are some thoughts:

The article explores how individuals form a sense of confidence in their moral choices, particularly in situations involving trade-offs between personal gain and causing harm. Rather than focusing solely on what people choose, the research delves into how confident people feel about the decisions they make—what is known as subjective confidence. Importantly, this confidence is not only influenced by the perceived value of the options but also by the moral implications of the choice itself. When people make decisions that benefit themselves at the expense of others, particularly when the action is considered morally blameworthy, their sense of confidence tends to decrease. Conversely, decisions that are morally neutral or praiseworthy are associated with greater subjective certainty. In this way, the moral weight of a decision appears to shape how individuals internally evaluate the quality of their choices.

For mental health professionals, these findings carry significant implications. Understanding how confidence is constructed in the context of moral decision-making can deepen insight into clients’ struggles with guilt, shame, indecision, and moral injury. Often, clients question not just what they did, but whether they made the "right" decision—morally and personally. This research highlights that moral self-evaluation is complex and sensitive to both the outcomes and the perceived ethical nature of one’s actions. It also suggests that people are more confident in decisions that affect themselves than those that impact others, which may help explain patterns of self-doubt or moral rumination in therapy. Additionally, for clinicians themselves—who frequently navigate ethically ambiguous situations—recognizing how subjective confidence is shaped by moral context can support reflective practice, supervision, and ethical decision-making. Ultimately, this research adds depth to our understanding of how people process and live with the choices they make, and how these internal evaluations may guide future behavior and psychological well-being.