COMPUTATIONAL SOCIAL AFFECTIVE NEUROSCIENCE

Virtually every action requires some degree of social consideration. These considerations could be the beliefs, feelings, or actions of a particular individual, or more broadly, codes of conduct informed by cultural customs and social norms. My research seeks to understand the psychological and neurobiological mechanisms sustaining these social processes and focuses primarily on three questions pertaining to social cognition and emotions: (1) How do emotions influence social decision-making?; (2) How do we represent others beliefs and social norms?; and (3) How can social interactions regulate our emotions and somatic pain? To approach these problems, I use a computational framework that is grounded in cognitive neuroscience. For example, beliefs about other's actions/feelings/beliefs can be formally described as probability distributions and updated via feedback received through interpersonal interactions. Cognitive and affective motivations can be quantified in terms of subjective value and integrated into utility functions, which can then be used to describe behavioral actions such as decisions. These types of computational tools provide a formal framework to describe the complex processes involved in social cognition and emotions and serve as organizing principles for my approach to studying the neurobiology of social interactions.

Emotions

Making a decision involves selecting the option that best maximizes the benefits while minimizing the costs with respect to a particular goal. Emotions can modulate these value signals, but can also serve as independent motivational signals in the decision-making process (Chang & Sanfey, 2008). These motivational signals can take the form of expected emotions, which refer to anticipated emotional states associated with a given decision that are never actually experienced (e.g., guilt; Chang, Smith, Dufwenberg, & Sanfey), or immediate emotions that are experienced at the time of decision and occur either directly in response to a specific event (e.g., anger or fear; Chang & Sanfey, 2013; van't Wout, Chang, & Sanfey, 2010) or as a result of a transitory fluctuation in mood (e.g., sadness; Harle, Chang, van't Wout, & Sanfey, 2012). In collaboration with a number of colleagues, I have sought to characterize how these distinct classes of emotions are processed in the brain using an interdisciplinary approach that combines theory and methods from psychology, economics, and neuroscience.

Social Norms

One of the things that we have found is that emotions provide a mechanism at an individual level to sustain social norms. Social emotions such as guilt (Chang et al., 2011) and anger (Chang and Sanfey, 2013) can be conceptualized as value signals which motivate social actions and formally described as deviations from social expectations. For example, guilt involves disappointing a relationship partner and can be mathematically defined as the difference between a possible action and what you believe your relationship partner expects you to do. Operationalizing these emotions in an economic utility function provides testable predictions for understanding how these value signals influence behavior and also how they might be represented in the brain. The results of this work suggest that a coherent network in the brain previously associated with both conflict and emotion (e.g., insula and ACC) underlies decisions to behave consistently with a social norm. Social emotions can thus be construed as negative affective states resulting from deviations from shared social expectations and we appear to be motivated to behave consistently with these expectations in order to avoid experiencing any discomfort (Chang, et al., 2011). Emotions can also motivate us to enforce a social norm following another's transgression by this same process (Chang & Sanfey, 2013). This work provides a computational and neurobiological foundation for understanding amorphous constructs such as moral sentiments.

Neuropsychometrics of Emotion

An important caveat to my work on social emotions, is that the insula turns out to be one of the most frequently activated regions in all of cognitive neuroscience (activated in approximately 40% of all studies) and thus is likely not specific to emotions. Machine learning tools from computer science (e.g., unsupervised and supervised classifcation) can be useful in assessing the psychometric properties of brain patterns in describing psychological constructs. For example, the insula appears to be functionally heterogeneous and can be parcellated into at least 3 anatomically distinct subregions (e.g., dorsal anterior, ventral anterior, and posterior) based on shared patterns of connectivity with the rest of the brain. The consistency and specificity of the function of these subregions (e.g., processing cognitive, affective-chemosensory, and sensorimotor information) can be assessed using large-scale meta-analytic decoding using the neurosynth database, which contains activations from approximately 3,500 studies (Chang, Yarkoni, Khaw, & Sanfey, 2013). My current work on emotion attempts to define patterns of brain activity that are predictive of negative affective states. In these studies I use machine-learning algorithms to identify multivariate patterns of brain activation that can predict how much guilt a given participant is experiencing (Chang, Smith, Dufwenberg, Wager, & Sanfey, In preparation) or how they might rate a negative picture they are viewing (Chang, Gianaros, Manuck, Krishnan, & Wager, In preparation). Using this approach we can begin to assess the psychometric properties of brain patterns such as the sensitivity, specificity, and even validity of a pattern in predicting a psychological state. Brain patterns, just like any other psychological or medical test, must demonstrate adequate reliability and validity before they can be used to characterize and diagnose clinical pathology.

Social Learning

This line of research seeks to understand how trust is developed and maintained via social learning. From a psychological perspective, trust can be considered the degree to which an individual believes a relationship partner will assist in attaining a specific interdependent goal. This work has combined reinforcement learning models with fMRI to understand the psychological and neural mechanism underlying how initial beliefs about a relationship partner's trustworthinesss impacts their ability to trust them after repeated interactions in an economic game framework (Chang et al., 2010; Fareri, Chang, and Delgado, 2012; Ratala, et al., In Preparation). The results suggest that trust can be operationalized as a probability of reciprocation that is dynamically updated based on experience (Chang, et al., 2010). The computational process of updating beliefs about trustworthiness appears to recruit the ventral striatum which suggests that social learning utilizes more basic learning systems (Fareri, Chang, & Delgado, 2012; Ratala et al., In Preparation). Interestingly, initial impressions may bias how beliefs are updated from experience consistent with predictions from instructed learning.

Nonspecific Psychotherapeutic Effects

Considerable effort spanning multiple decades has revealed that psychotherapeutic interventions are effective in treating a variety of psychological symptoms. Meta-analyses of treatment effects across therapeutic modalities indicate that many treatments appear to be similarly effective at alleviating psychological distress. This has lead to the popular notion of "common factors", which refer to aspects of psychotherapy that transcend therapeutic modalities. While it seems to be generally accepted that nonspecific factors such as expectations and the therapeutic relationship are important in treatment outcomes, there is surprisingly little known about why they work (Chang, Sanfey, & Wager, In preparation). My current work is focused on understanding the psychological, computational, and neurobiological mechanisms sustaining expectation and relationship based treatment effects in pain processing. This work aims to (1) apply computational modeling approaches to understand the expectation effects associated with placebos, (2) use experimental manipulations of provider characteristics (e.g., trustworthiness, expectations, and emotions) to reveal how aspects of the social interaction influence perceptions of pain, (3) understand how emotions and pain are regulated in the context of romantic relationships, and (4) use pharmacological manipulations to test underlying neurobiological mechanisms of expectation and relationship based effects on pain processing. I believe that this type of mechanistic translational approach will pave the way in bridging cognitive neuroscience with clinical psychology and will hopefully inspire future innovations.