Chris Frith

The human dorsal anterior cingulate facilitates acceptance of unfair offers and regulates inequity aversion

Functional MRI examines how the brain manages unfair offers in a bargaining game, suggesting implications for social behavior.

AUG 2025FrithAnnual Review of Psychology

Scenes from a Marriage: How We Found Our Way from Experimental Psychology to Social Neuroscience

Reflecting on cognitive processes and their social impact highlights culture's role in experimental psychology.

APR 2025FrithCurrent Directions in Psychological Science

What Autism Taught Us About Our Social Nature

Autism has influenced social-cognitive neuroscience in important ways. It has provided the impetus to look for the brain basis of mentalizing and encouraged the search for the brain bases of other social abilities. A fundamental aspect of social interaction is the ability to predict what other agents are going to do. We propose a hierarchy of three worlds—the world of objects, the world of agents, and the world of ideas—that respectively present their own challenges and solutions to such predictions. The world of ideas provides a direct interface between individual minds and other minds (i.e., culture). We highlight the power of culture to change subjective experiences and the power of subjective experiences to influence culture. The example of autism shows these mutual influences at work. These influences have led to dramatic changes in the concept of “autism” since its first use in child psychiatry.

Overcoming pride via the dorsal ACC underlies acceptance of unfair offers

Bargaining is a fundamental social behavior in which individuals often accept unfair offers. Traditional behavioral models, based solely on choice data, typically interpret this acceptance as simple reward-maximization. However, the suppression of emotions such as inequity aversion or pride may also play a critical role in this decision. Incorporating response time alongside choice data provides a means to quantify participants’ internal conflict in suppressing these emotions and deciding to accept unfair offers. In this study, we conducted functional magnetic resonance imaging (fMRI) of the ultimatum game, where participants decided within 10 seconds whether to accept or reject monetary distribution offers from a proposer. Using the drift diffusion model (DDM), we quantified decision-making dynamics based on both choice and response time. Participants who suppressed disadvantageous inequity (DI)-driven rejection (reflected by a lower DDM weight for DI) exhibited heightened dorsal anterior cingulate cortex (dACC) activity in response to DI. Functional connectivity analysis revealed a negative correlation between the dACC and the ventrolateral prefrontal cortex (vlPFC) when DI was large, which encoded both the rejection rates, and the response times associated with accepting DI offers. Furthermore, vlPFC activity was significantly correlated with amygdala activity during high DI conditions, specifically encoding response time for accepting DI offers but not rejection rates. Importantly, these findings could not be captured using standard value-based models that rely solely on choice data. Our results underscore the dACC’s critical role in mediating the suppression of emotional responses to DI, enabling the acceptance of unfair offers in a dynamic bargaining process.

JAN 2025FrithOpen Mind

Sharing the World—A Social Aspect of Consciousness

Moving through our environment generates multiple changes in my sensations. But I do not experience the environment as changing. My conscious perceptual experience is of a stable environment through which I move. This perception is created by intricate neural computations that automatically take account of my movements. The stable environment that I experience is independent of my actions. As a result, I experience it as objective: a set of facts about the world that constrain my movements. Because it is objective I expect that it will also constrain the movements of others in the same way, whether these are rocks rolling down a hill or animals foraging for food. This experience of objectivity creates a shared understanding of the world that enhances our interactions with others. Our perceptual experiences, while personal, are shaped by our model of the world, and since others are modelling the same world, their models will be very similar. Interactions with others will further increase this similarity. The models create a form of common knowledge. This common knowledge is an inherent feature of our basic conscious perception, even when we're not actively reflecting on or deliberately sharing our experiences. The common knowledge created by our conscious perception of the world enables the coordination of behaviour which is a critical precursor for the evolution of cooperative behaviour.

SEP 2024Frith

Sharing the world – a social aspect of consciousness

I suggest that shared knowledge is an inherent feature of conscious experience, even when we're not actively reflecting on or sharing our experiences. When we navigate the physical world, we perceive ourselves as existing within a stationary environment independent of our actions. This perception arises from intricate neural computations that take account of our movements as well as sensory input. This shared understanding of the world is essential for successful social interaction, as we assume that others perceive the same reality. Our subjective experiences, while personal, are shaped by our shared understanding of the world, which becomes a form of common knowledge.

Defining key concepts for mental state attribution

The terminology used in discussions on mental state attribution is extensive and lacks consistency. In the current paper, experts from various disciplines collaborate to introduce a shared set of concepts and make recommendations regarding future use.

Selfish behavior requires top-down control of prosocial motivation

Individuals must regularly choose between prosocial and proself behaviors. While past neuroscience research has revealed the neural foundations for prosocial behaviors, many studies have oversimplified proself behaviors, viewing them merely as a reward-maximization process. However, recent behavioral evidence suggests that response times for proself behaviors are often slower than those for prosocial behaviors, suggesting a more complex interdependence between prosocial and proself neural computations. To address this issue, we conducted an fMRI experiment with the ultimatum game, where participants were requested to accept (money distributed as proposed) or reject (both sides receive none) offers of money distribution. In the decisions, the participants could maximize self-interest by accepting the offer (i.e., proself), while by rejecting it, they could punish unfair proposers and promote the “equity” social norm (i.e., prosocial). We constructed a drift diffusion model (DDM) that considers both behavioral choices and response times and used the DDM parameters in our fMRI analysis. We observed that participants who suppressed inequity-driven rejection behaviors displayed heightened dACC activity in response to disadvantageous inequity. Importantly, our functional connectivity analysis demonstrated that the dACC exhibited negative functional connectivity with the amygdala when unfair offers were presented. Furthermore, the PPI connectivity encoded the average reaction time for accepting unfair offers (i.e., proself behaviors). Considering that the amygdala also responded to disadvantageous inequity in these experiments and previous studies, these results show that the top-down control of prosocial motives (i.e., aversion to disadvantageous inequity) plays a key role in implementing proself behaviors.

JAN 2024FrithCognitive Neuropsychiatry

What makes us social and what does it tell us about mental disorders?

DEC 2023FristonNeuroscience & Biobehavioral Reviews

Federated inference and belief sharing

This paper concerns the distributed intelligence or federated inference that emerges under belief-sharing among agents who share a common world-and world model. Imagine, for example, several animals keeping a lookout for predators. Their collective surveillance rests upon being able to communicate their beliefs-about what they see-among themselves. But, how is this possible? Here, we show how all the necessary components arise from minimising free energy. We use numerical studies to simulate the generation, acquisition and emergence of language in synthetic agents. Specifically, we consider inference, learning and selection as minimising the variational free energy of posterior (i.e., Bayesian) beliefs about the states, parameters and structure of generative models, respectively. The common theme-that attends these optimisation processes-is the selection of actions that minimise expected free energy, leading to active inference, learning and model selection (a.k.a., structure learning). We first illustrate the role of communication in resolving uncertainty about the latent states of a partially observed world, on which agents have complementary perspectives. We then consider the acquisition of the requisite language-entailed by a likelihood mapping from an agent's beliefs to their overt expression (e.g., speech)-showing that language can be transmitted across generations by active learning. Finally, we show that language is an emergent property of free energy minimisation, when agents operate within the same econiche. We conclude with a discussion of various perspectives on these phenomena; ranging from cultural niche construction, through federated learning, to the emergence of complexity in ensembles of self-organising systems.

A Standard Framework for Social Cognition: interoperable algorithms for inference and representation

Interpersonal relationships are a central feature of what it is to be human. Theory of Mind (ToM), or mentalising, is the ability to represent the hidden thoughts and beliefs of the self and others to navigate these relationships. However, to date there is no common set of formal principles that can be used to understand interpersonal relationships across context, disorder, and diversity, and in humans and artificial agents. Here we synthesise previously distinct components of social representation and inference from a broad range of past works into a common computational framework – the Standard Framework for Social Cognition - to allow translation across previously disparate processes, and to provide a starting point for local theories. We discuss applications of our standard framework to problems regarding dysfunctional theory of mind, neural implementation, artificial intelligence, and discuss potential future directions for the field.

The Integrated Information Theory of Consciousness as Pseudoscience

The media, including news articles in both Nature and Science, have recently celebrated the Integrated Information Theory (IIT) as a ‘leading’ and empirically tested theory of consciousness. We are writing as researchers with some relevant expertise to express our concerns.

Consciousness in Artificial Intelligence: Insights from the Science of Consciousness

Theoretical review reveals an absence of conscious traits in current AI systems, suggesting future architectures could feasibly achieve computational markers of consciousness.

MAR 2023FrithQuarterly Journal of Experimental Psychology

Consciousness, (meta)cognition, and culture

Our conscious experience is determined by a combination of top-down processes (e.g., prior beliefs) and bottom-up processes (e.g., sensations). The balance between these two processes depends on estimates of their reliability (precision), so that the estimate considered more reliable is given more weight. We can modify these estimates at the metacognitive level, changing the relative weights of priors and sensations. This enables us, for example, to direct our attention to weak stimuli. But there is a cost to this malleability. For example, excessive weighting of top-down processes, as in schizophrenia, can lead to perceiving things that are not there and believing things that are not true. It is only at the top of the brain's cognitive hierarchy that metacognitive control becomes conscious. At this level, our beliefs concern complex, abstract entities with which we have limited direct experience. Estimates of the precision of such beliefs are more uncertain and more malleable. However, at this level, we do not need to rely on our own limited experience. We can rely instead on the experiences of others. Explicit metacognition plays a unique role, enabling us to share our experiences. We acquire our beliefs about the world from our immediate social group and from our wider culture. And the same sources provide us with better estimates of the precision of these beliefs. Our confidence in our high-level beliefs is heavily influenced by culture at the expense of direct experience.

Animacy and the prediction of behaviour

To survive, all animals need to predict what other agents are going to do next. We review neural mechanisms involved in the steps required for this ability. The first step is to determine whether an object is an agent, and if so, how sophisticated it is. This involves brain regions carrying representations of animate agents. The movements of the agent can then be anticipated in the short term based solely on physical constraints. In the longer term, taking into account the agent's goals and intentions is useful. Observing goal directed behaviour activates the neural action observation network, and predicting future goal directed behaviour is helped by the observer's own action generating mechanisms. Intentions are critically important in determining actions when interacting with other agents, as several intentions can lie behind an action. Here, interpretation is helped by prior beliefs about the agent and the brain's mentalising system is engaged. Biologically-constrained computational models of action recognition exist, but equivalent models for understanding intentional agents remain to be developed.

Animacy and the prediction of behaviour

To survive, all animals need to predict what other agents are going to do next. We review neural mechanisms involved in the steps required for this ability. The first step is to determine whether an object is an agent, and if so, how sophisticated it is. This involves brain regions carrying representations of animate agents. The movements of the agent can then be anticipated in the short term based solely on physical constraints. In the longer term, taking into account the agent’s goals and intentions is useful. Observing goal directed behaviour activates the neural action observation network, and predicting future goal directed behaviour is helped by the observer’s own action generating mechanisms. Intentions are critically important in determining actions when interacting with other agents, as several intentions can lie behind an action. Here, interpretation is helped by prior beliefs about the agent and the brain’s mentalising system is engaged. Biologically-constrained computational models of action recognition exist, but equivalent models for understanding intentional agents remain to be developed.

Helping the waiter to hold his tray: Rigid haptic linkage promotes inter-personal motor coordination

When a glass is lifted from a tray, there is a challenge for the waiter. He must quickly compensate for the reduction in the weight of the tray to keep it balanced. This compensation is easily achieved if the waiter lifts the glass himself. Because he has, himself, initiated the action, he can predict the timing and the magnitude of the perturbation of the tray and respond (via the holding hand) accordingly. In this study, we examined coordination when either one or two people hold the tray while either one of them or a third person removes the glass. Our results show that there is exquisite coordination between the two people holding the tray. We suggest that this coordination depends upon the haptic link provided by the rigid platform that both people are holding. We conclude that the guest at a reception should not lift his drink from the waiter's tray until they have the waiter's attention but, if too thirsty to wait, should lend a hand holding the tray.

Awareness and confidence in perceptual decision-making

Perceptual decision-making employs a range of higher order metacognitive processes. Two of the most important of these are perceptual awareness; or the clarity with which one reports seeing a perceptual stimulus, and response confidence; or the certainty one has about the correctness of one's own perceptual categorizations. We used a novel false feedback paradigm to investigate the relationships between these two processes. We asked people to perform a standard psychophysical detection task. We used feedback to selectively intervene either on our participants’ trust in their own perceptual awareness of the stimulus, or on their confidence in their own responses. We measured the effects of these interventions on response accuracy; on reports of perceptual awareness; and on response confidence. We found that by undermining people's trust in their awareness of the sensory stimulus, we could reliably reduce their accuracy on the task. We suggest that the reason this occurred is that people came to rely less on evidence from their senses when making perceptual decisions. We conclude by suggesting that there is a not a one-to-one mapping between content in conscious experience and how that content is used in perceptual decision making, and that one's perception of the reliability of content also plays a role. This paper explores how different kinds of metacognitive state are related to one another and to perceptual decision making. Our focus is on the states of metacognitive confidence and perceptual awareness. We examine how an intervention on the reliability of these states influences performance in a perceptual detection task. We also examine how the intervention influences reports of the states themselves. The intervention we use is false feedback. For one group of participants, we tell them their perceptual judgement is wrong whenever they report they are uncertain in their choice (confidence intervention). For another group, we tell them their judgement is wrong whenever they report that their experience of the stimulus is unclear (awareness intervention). We find that both interventions reduce the accuracy of people's judgements, but that the awareness intervention is more effective. Also, we find that only the awareness intervention reduces reports of both metacognitive confidence in the response, and awareness of the stimuli. The confidence intervention does not influence either metacognitive state. These results suggest that we should understand confidence and awareness as separate higher level cognitive states, and that we should understand awareness as having a stronger causal role than confidence in perception and performance.

Animacy and the prediction of behaviour

To survive, all animals need to predict what other agents are going to do next. Here, we review neural mechanisms involved in the steps required for this ability. The first step is to determine whether an object is an agent and, if so, how sophisticated it is. This involves brain regions carrying representations of animate agents. Once identified, the movements of an agent can be anticipated in the short term on the basis of purely physical constraints. In the longer term, taking into account the agent’s goals and intentions is useful. Goal directed agents are marked by the rationality of their movements, reaching their goals by the shortest or least effortful path. Observing goal directed behaviour activates the brain’s action observation/mirror neuron network, and predicting future goal directed behaviour is helped by the observer’s own action generating mechanisms. Intentions have a critical role in determining actions when agents interact with other agents, as several intentions can lie behind a given action. In these cases, interpretation is helped by prior beliefs about the agent and the brain’s mentalising system is engaged. Biologically-constrained computational models of action recognition are available, but equivalent models for understanding intentional agents remain to be developed.

Knowing Ourselves Together: The Cultural Origins of Metacognition

Metacognition - the ability to represent, monitor and control ongoing cognitive processes - helps us perform many tasks, both when acting alone and when working with others. While metacognition is adaptive, and found in other animals, we should not assume that all human forms of metacognition are gene-based adaptations. Instead, some forms may have a social origin, including the discrimination, interpretation, and broadcasting of metacognitive representations. There is evidence that each of these abilities depends on cultural learning and therefore that cultural selection might shape human metacognition. The cultural origins hypothesis is a plausible and testable alternative that directs us towards a substantial new programme of research.