Search

Pesnot Lerousseau and Summerfield compare humans to transformer neural networks in a learning task designed to distinguish ‘in-weights’ and ‘in-context’ learning processes.

Using artificial neural networks applied to human data, Eckstein et al. show that good models of reinforcement learning require memory components that track representations of the past.

This Perspective from Summerfield et al. considers the impacts of advanced artificial intelligence systems on the process and function of democracy. The authors explore a wide range of potential risks and opportunities.

In this Perspective, the authors discuss chatbot interactions, highlighting potential risks of emotional dependence and mental health impacts, necessitating coordinated public health responses.

When learning new tasks, both humans and artificial neural networks face a trade-off between reusing prior knowledge to learn faster and avoiding the disruption of earlier learning. This study shows that people and artificial neural networks have similar patterns of transfer and interference and vary in how they balance this trade-off.

The clinical course of chronic lymphocytic leukaemia (CLL) is variable and difficult to predict. Here, the authors conduct a genome wide association study meta-analysis for time to first treatment in CLL patients and report two loci associating with progressive disease.

Visual stimuli can often be predicted by other stimuli in the environment — for example, a barking sound would predict the sight of a dog but not a cat. In this Review, Summerfield and de Lange discuss how expectation modulates neural signals and behaviour in response to visual stimuli.

Santiago Herce Castañón and colleagues show that people are blind to mental errors that arise when combining multiple pieces of discordant information. This blindness helps explain why cognitive judgements often are suboptimal.

Errors are typically followed by more cautious responses. A study now provides evidence that remarkably conserved neural dynamics underlie these post-error adjustments to behavior in rodents and humans.

Spitzer et al. investigate the neural and computational mechanisms involved in weighting, integrating and comparing numbers. They find systematic overweighting of larger numbers, which is reflected in stronger neural signals over the parietal cortex.

Koster et al introduce a deep reinforcement learning (RL) mechanism designed to manage common-pool resources successfully encourages sustainable cooperation among human participants by dynamically adjusting resource allocations based on the current state of the resource pool. The RL-derived policy outperforms traditional allocation methods by balancing generosity when resources are abundant and applying temporary sanctions to discourage free-riding, ultimately maximizing social welfare and fairness.

Moiré field-effect transistors based on graphene/hexagonal boron nitride heterostructures are promising for their high room-temperature carrier mobilities and magnetotransport properties. Here, high-temperature molecular beam epitaxy growth of graphene/hBN gives rise to a moiré-fringed hexagonal superlattice with Hofstadter butterfly electronic band structure and quantum magneto-oscillations above room temperature.

When making economic decisions, our choices are often influenced by irrelevant information. One prominent explanation appeals to normalisation in neural circuits. A new paper by Gluth and colleagues suggests that instead, attentional processes may be responsible.

Repetition suppression, the reduction in neural activity with repeated stimuli, is usually thought to be a result of automatic sensory processes. This study instead finds that this reduction results from high stimulus predictability, a more 'top-down' process.

Koster, Balaguer et al. show that an AI mechanism is able to learn to produce a redistribution policy which is preferred to alternatives by humans in an incentivized game.

Deep neural networks may offer theories of perception, cognition and action for biological brains. Here, Saxe, Nelli and Summerfield offer a road map of how neuroscientists can use deep networks to model and understand biological brains.

Little is known about the brain’s computations that enable the recognition of faces. Here, the authors use unsupervised deep learning to show that the brain disentangles faces into semantically meaningful factors, like age or the presence of a smile, at the single neuron level.

A ‘differentiable neural computer’ is introduced that combines the learning capabilities of a neural network with an external memory analogous to the random-access memory in a conventional computer.

Bang et al. use behavioural data in culturally distinct settings (United Kingdom and Iran) and computational modelling to show that, when making decisions in pairs, people adopt a confidence-matching heuristic to combine their opinions.

To study cognition, researchers have traditionally used laboratory-based experiments, but games offer a valuable alternative: they are intuitive and enjoyable. In this Perspective, Schulz et al. discuss the advantages and drawbacks of games and give recommendations for researchers.

Structural analysis of two human monoclonal antibodies that conform the antibody cocktail AZD7442, in complex with the RBD of SARS-CoV-2, reveal strong neutralization of SARS-CoV-2 variants of concern.

Humans strive to design safe AI systems that align with our goals and remain under our control. However, as AI capabilities advance, we face a new challenge: the emergence of deeper, more persistent relationships between humans and AI systems. We explore how increasingly capable AI agents may generate the perception of deeper relationships with users, especially as AI becomes more personalised and agentic. This shift, from transactional interaction to ongoing sustained social engagement with AI, necessitates a new focus on socioaffective alignment—how an AI system behaves within the social and psychological ecosystem co-created with its user, where preferences and perceptions evolve through mutual influence. Addressing these dynamics involves resolving key intrapersonal dilemmas, including balancing immediate versus long-term well-being, protecting autonomy, and managing AI companionship alongside the desire to preserve human social bonds. By framing these challenges through a notion of basic psychological needs, we seek AI systems that support, rather than exploit, our fundamental nature as social and emotional beings.