Aamir Sohail

BackgroundLearning from feedback provided by others is important for navigating complex social environments. Aberrant processing of such social feedback is implicated in the psychopathology of several mental health disorders including social anxiety and depression. Recent psychiatric research increasingly adopts transdiagnostic approaches that examine mechanisms and constructs cutting across traditional diagnostic categories. MethodsHere, we applied reinforcement learning models together with transdiagnostic measures of psychopathology to formally assess social evaluation learning. Human participants (n = 193) completed the Social Evaluation Learning Task, learning whether computer personas ‘liked,’ ‘disliked,’ or were ‘neutral’ toward them. Participants also completed six questionnaires assessing socially relevant psychiatric traits. ResultsComputational modeling revealed that a model with separate learning parameters for positive and negative feedback best accounted for the learning behavior. Exploratory factor analysis identified three transdiagnostic factors. A factor reflecting social avoidance predicted enhanced learning from negative feedback and a reduced positive learning bias. Socially avoidant individuals with heightened negative learning rates additionally showed faster reaction times specifically in positive social contexts. A second factor linking with emotional insensitivity for others predicted a lower choice accuracy in both positively and negatively valanced conditions and was associated with faster reaction times overall. No significant associations emerged for a third factor reflecting depressive and mood-related symptoms. ConclusionsThese results demonstrate how transdiagnostic trait dimensions shape social learning mechanisms through specific behavioral and computational processes.

Mentalization - the ability to infer others’ beliefs and intentions to guide one’s own choices - is a key cognitive function underlying human social interactions. Large language models (LLMs) demonstrate behaviour consistent with humans on theory-of-mind tasks, yet whether these models can guide adaptive behaviour through mentalization is unknown. Here we use two economic games with cognitive computational modeling to uncover the latent strategies underlying mentalization in LLMs. We tested individual LLM agents across four model families, DeepSeek, GPT-4.1, GPT-5 and Gemini 2.0 Flash (N = 2,099), against opponents of varying sophistication and examined whether a prompting strategy designed to elicit strategic reasoning improved performance. We benchmarked results against human participants (N = 251) as a comparative measure. Across both games, LLMs showed clear behavioural and computational signatures of mentalizing that differed markedly by model provider and size. Strategic prompting generally improved performance by inducing more sophisticated reasoning, yet the extent of the benefit differed across the two tasks. Last, GPT-5 agents flexibly adapted their recursive depth of reasoning to increasingly sophisticated opponents, demonstrating superior performance to human participants. Collectively, we demonstrate different capacities for mentalization across LLMs, and highlight cognitive computational modeling as a formal method for assessing comparative intelligence across humans and machines.

Background: Learning from feedback provided by others is important for navigating complex social environments. Aberrant processing of such social feedback is implicated in the psychopathology of several mental health disorders including social anxiety and depression. Recent psychiatric research increasingly adopts transdiagnostic approaches that examine mechanisms and constructs cutting across traditional diagnostic categories. Methods: Here, we applied reinforcement learning models together with transdiagnostic measures of psychopathology to formally assess social evaluation learning. Human participants ( n = 193) completed the Social Evaluation Learning Task, learning whether computer personas ‘liked’, ‘disliked’, or were ‘neutral’ toward them. Participants also completed six questionnaires assessing socially relevant psychiatric traits. Results: Computational modeling revealed that a model with separate learning parameters for positive and negative feedback best accounted for the learning behavior. Exploratory factor analysis identified three transdiagnostic factors. A factor reflecting social avoidance predicted enhanced learning from negative feedback and a reduced positive learning bias. Socially avoidant individuals with heightened negative learning rates additionally showed slower decision times specifically in positive social contexts. A second factor linking with emotional insensitivity for others predicted a lower choice accuracy in both positively and negatively valenced conditions. No significant associations emerged for a third factor reflecting depressive and mood-related symptoms. Conclusions: These results demonstrate how transdiagnostic trait dimensions shape social learning mechanisms through specific behavioral and computational processes.

University placements provide valuable hands-on research experience for students, whilst allowing labs to engage in meaningful outreach. However, building and structuring these placements for college students (ages 16 to 18) may present challenges. Firstly, many hosts lack teaching experience towards this specific group, which differ significantly from university students in their subject knowledge and research skills. Additionally, hosts must balance creating engaging content while ensuring students develop meaningful practical skills and knowledge. Drawing from our collective experience, we provide practical guidelines in structuring placements in the fields of psychology and cognitive neuroscience. We discuss the importance of advance planning, tailoring content to student interests, incorporating practical elements, and promoting interdisciplinary engagement. In doing so, we aim to encourage more labs to host student placements, subsequently widening access for psychology and neuroscience research.

We present BayesCog, an openly-available online course for the computational modeling of human behaviour (i.e., cognitive modeling) using Bayesian inference, with reinforcement learning as a core example throughout the course. Assuming little to no prior experience, audience of this course will be formally grounded in key concepts including Bayesian statistics and reinforcement learning, and practically, will build, assess, compare, and validate models using the R interface to the Stan programming language, RStan. Starting with binary choice models, the audience will learn to estimate parameters representing latent components of behaviour by fitting reinforcement learning models, both at the individual and group-level, eventually with hierarchical modeling. The course is generally suitable for those interested in developing models of human cognition at any level of experience. In making the course openly available, we aim for computational modeling under the Bayesian approach to be more strongly represented in the psychological sciences.

The integration of GenAI into academic workflows represents a fundamental shift in scientific practice. While these tools can amplify productivity, they risk eroding the cognitive foundations of expertise by simulating the very tasks through which scientific competence is developed, from synthesis to experimental design to writing. Uncritical reliance can lead to skill atrophy and AI complacency. We propose a framework of essential AI meta-skills: strategic direction, critical discernment, and systematic calibration. These constitute a new form of scientific literacy that builds on traditional critical thinking. Through domain-specific examples and a pedagogical model based on situated learning, we show how these meta-skills can be cultivated to ensure that researchers, particularly those in training, maintain intellectual autonomy. Without deliberate cultivation of these meta-skills, we risk creating the first generation of researchers who serve their tools rather than direct them.

University placements provide valuable hands-on research experience for students, whilst allowing labs to engage in meaningful outreach. However, building and structuring these placements for college students (ages 16-18) may present challenges. Firstly, many hosts lack teaching experience towards this specific group, which differ significantly from university students in their subject knowledge and research skills. Additionally, hosts must balance creating engaging content while ensuring students develop meaningful practical skills and knowledge. Drawing from our collective experience, we provide practical guidelines in structuring placements in the fields of psychology and cognitive neuroscience. We discuss the importance of advance planning, tailoring content to student interests, incorporating practical elements, and promoting interdisciplinary engagement. In doing so, we aim to encourage more labs to host student placements, subsequently widening access for psychology and neuroscience research.

The philosopher Soren Kierkegaard argued in his 1844 work ‘The Concept of Anxiety’ that anxiety is a vital aspect of the human condition, a consequence of freedom and choice. In this article I argue that this, along with other key concepts of anxiety identified by Kierkegaard including the avoidance of past negative experiences, and education in choice outcomes, are supported by modern theories of anxiety using evidence from social and experimental psychology as well as psychiatry. A proposed future body of work involving anxiety and free will stemming from Kierkegaard’s philosophy is also discussed.