In their thoughtful review, Veissière and colleagues motivate active inference as an integrative framework for understanding and modeling the dynamics of cultural cognition. Conspicuously absent was discussion of the role of emotions in “thinking through other minds.” However, emotions plausibly underwrite the motivation, maintenance, and enforcement of many sociocultural norms and practices (e.g., feeling an automatic aversion to breaking a social norm, or anticipating that others would react with anger/disappointment if one failed to engage in a cultural practice). Here, we argue that there is an important opportunity to expand their model by explicitly incorporating emotion. Motivated by the active inference framework, we will focus on dynamic bidirectional interactions between two broad emotion-related processes: affective response generation (cast as policy selection) and affective response perception (cast as Bayesian state inference).
As elsewhere (Smith et al. Reference Smith, Killgore and Lane2018c), we use the broad term “affective response generation” to denote the flexible engagement of multiple quick/involuntary changes across visceromotor, skeletomotor, and attentional states in response to (current, remembered, or imagined) interoceptive or exteroceptive stimuli of perceived significance to an organism's needs, goals, and values. For example, consider the simultaneous elicitation of unpleasant changes in posture, facial expression, autonomic arousal, threat-biased attention, and avoidance motivation that can quickly ensue in response to a simple social gesture. Although not always, and not always within awareness, these types of multimodal internal responses often occur in response to the perceived thoughts and feelings of others, and they can play an essential role in the sociocultural phenomena that Veissière and colleagues discuss. Within the active inference framework, recent work (Allen et al. Reference Allen, Levy, Parr and Friston2019; Smith et al. Reference Smith, Lane, Parr and Friston2019a; Reference Smith, Parr and Friston2019b) has shown how these responses can be cast in terms of multimodal policy selection. That is, based on a set of (e.g., interoceptive or social) signals, the states of the world, the body, and of the minds of self and others can be (implicitly or explicitly) inferred, which can then engage predictions about how these states will change over time if different policies (i.e., different sets of visceromotor, skeletomotor, and attentional changes) were selected. The “affective response” that is generated then corresponds to the enactment of the policy predicted to lead to the states most consistent with an organism's preferences (e.g., perceiving social approval) – often involving automatic attention to salient sources of information (e.g., the locations of friends and enemies) and visceromotor adjustments based on the predicted metabolic demands associated with navigating the environment so as to attain preferred states (e.g., staying close to friends and away from enemies).
There are many cases in which this type of quick/involuntary policy selection process can play adaptive (and often unintentional) social roles. For example, although typically not intentional, the automatic production of tears (crying) can elicit helpful social support from others. Unintended changes in posture and muscle tension in response to social norm violations can also convey implicit signals to others about the probability that aggressive action will ensue if such violations continue.
However, the presence of an affective response does not entail that it will be perceived/interpreted correctly by oneself or others. The ability to infer the interoceptive and emotional states of self and others, based on the internally/externally observable sensory consequences that follow from the enactment of affective policies (e.g., perceived heart rate or facial expression changes), has also been formulated as Bayesian inference within the active inference framework (e.g., inferring the probability that an individual is sad based on information about their body state and the context [Barrett Reference Barrett2017; Smith et al. Reference Smith, Killgore, Alkozei and Lane2018b]). When interacting with affective response generation, affective response perception can produce iterative bidirectional interactions between the minds and bodies of multiple individuals. For example, an individual could feel an aversion to participate in particular social practices, but then also react with frustration because they don't want to feel that aversion. Or an individual might be happy because they believe they are meeting others’ sociocultural expectations, but then react with disappointment when they perceive that others are displeased.
The relevant inferential dynamics also appear to play out at multiple hierarchical levels. For example, there is evidence that individuals automatically simulate the body states they perceive in others in order to infer what emotions they are feeling (Niedenthal Reference Niedenthal2007). There are also important cases where correctly inferring the emotions of others requires the deployment of additional higher-level knowledge (e.g., “even though I like this restaurant, she will be sad if we go there”). These different levels also plausibly facilitate different social dynamics. For example, simulating the discomfort of another individual's tense/shaky posture could promote automatic empathic responding (i.e., making the other individual feel better would make you feel better [Lamm et al. Reference Lamm, Decety and Singer2011]). In contrast, the explicit semantic inference that what another person is feeling corresponds to the concept FEAR can facilitate important social inferences about both the past and the future by drawing on emotion knowledge, such as the likely causes and consequences of that state (e.g., “it was likely caused by a perceived threat” and “the person is now likely to try to avoid that threat”) – which can then inform subsequent decision-making (Fig. 1).
Figure 1. Depiction of “thinking through others’ emotions” as an extension of the “thinking through other minds” framework. Based on the (implicitly or explicitly) perceived thoughts and feelings of others, quick/involuntary somatovisceral (e.g., valenced changes in facial expression, body posture, and autonomic state) and cognitive (e.g., selective attention) policies are enacted and perceived by both self and others. Subsequent inferences about one's own emotional state and the emotional states of others (both implicit and explicit) then further inform sociocultural decision-making (e.g., conforming to social norms and engaging in cultural practices).
Successfully navigating the hypersocial human niche requires iterative and nested use of these processes – in which one's goals can only be accomplished by predicting the emotions of others in the future under different courses of action (e.g., “he's feeling angry because I'm driving below the speed limit – if I speed up then he will calm down”). This in turn leads to complex and continuous feedback loops that can facilitate both adaptive (Smith et al. Reference Smith, Weihs, Alkozei, Killgore and Lane2019c) and maladaptive (Smith et al. Reference Smith, Alkozei, Killgore and Lane2018a) social dynamics. The resulting dynamics lead to a type of “thinking through others’ emotions” that depends jointly on interoceptive and exteroceptive inference and on bidirectional brain-body interactions across individuals throughout a society. We put forward these additional affective dynamics as essential to completing the authors’ account of thinking through other minds.
In their thoughtful review, Veissière and colleagues motivate active inference as an integrative framework for understanding and modeling the dynamics of cultural cognition. Conspicuously absent was discussion of the role of emotions in “thinking through other minds.” However, emotions plausibly underwrite the motivation, maintenance, and enforcement of many sociocultural norms and practices (e.g., feeling an automatic aversion to breaking a social norm, or anticipating that others would react with anger/disappointment if one failed to engage in a cultural practice). Here, we argue that there is an important opportunity to expand their model by explicitly incorporating emotion. Motivated by the active inference framework, we will focus on dynamic bidirectional interactions between two broad emotion-related processes: affective response generation (cast as policy selection) and affective response perception (cast as Bayesian state inference).
As elsewhere (Smith et al. Reference Smith, Killgore and Lane2018c), we use the broad term “affective response generation” to denote the flexible engagement of multiple quick/involuntary changes across visceromotor, skeletomotor, and attentional states in response to (current, remembered, or imagined) interoceptive or exteroceptive stimuli of perceived significance to an organism's needs, goals, and values. For example, consider the simultaneous elicitation of unpleasant changes in posture, facial expression, autonomic arousal, threat-biased attention, and avoidance motivation that can quickly ensue in response to a simple social gesture. Although not always, and not always within awareness, these types of multimodal internal responses often occur in response to the perceived thoughts and feelings of others, and they can play an essential role in the sociocultural phenomena that Veissière and colleagues discuss. Within the active inference framework, recent work (Allen et al. Reference Allen, Levy, Parr and Friston2019; Smith et al. Reference Smith, Lane, Parr and Friston2019a; Reference Smith, Parr and Friston2019b) has shown how these responses can be cast in terms of multimodal policy selection. That is, based on a set of (e.g., interoceptive or social) signals, the states of the world, the body, and of the minds of self and others can be (implicitly or explicitly) inferred, which can then engage predictions about how these states will change over time if different policies (i.e., different sets of visceromotor, skeletomotor, and attentional changes) were selected. The “affective response” that is generated then corresponds to the enactment of the policy predicted to lead to the states most consistent with an organism's preferences (e.g., perceiving social approval) – often involving automatic attention to salient sources of information (e.g., the locations of friends and enemies) and visceromotor adjustments based on the predicted metabolic demands associated with navigating the environment so as to attain preferred states (e.g., staying close to friends and away from enemies).
There are many cases in which this type of quick/involuntary policy selection process can play adaptive (and often unintentional) social roles. For example, although typically not intentional, the automatic production of tears (crying) can elicit helpful social support from others. Unintended changes in posture and muscle tension in response to social norm violations can also convey implicit signals to others about the probability that aggressive action will ensue if such violations continue.
However, the presence of an affective response does not entail that it will be perceived/interpreted correctly by oneself or others. The ability to infer the interoceptive and emotional states of self and others, based on the internally/externally observable sensory consequences that follow from the enactment of affective policies (e.g., perceived heart rate or facial expression changes), has also been formulated as Bayesian inference within the active inference framework (e.g., inferring the probability that an individual is sad based on information about their body state and the context [Barrett Reference Barrett2017; Smith et al. Reference Smith, Killgore, Alkozei and Lane2018b]). When interacting with affective response generation, affective response perception can produce iterative bidirectional interactions between the minds and bodies of multiple individuals. For example, an individual could feel an aversion to participate in particular social practices, but then also react with frustration because they don't want to feel that aversion. Or an individual might be happy because they believe they are meeting others’ sociocultural expectations, but then react with disappointment when they perceive that others are displeased.
The relevant inferential dynamics also appear to play out at multiple hierarchical levels. For example, there is evidence that individuals automatically simulate the body states they perceive in others in order to infer what emotions they are feeling (Niedenthal Reference Niedenthal2007). There are also important cases where correctly inferring the emotions of others requires the deployment of additional higher-level knowledge (e.g., “even though I like this restaurant, she will be sad if we go there”). These different levels also plausibly facilitate different social dynamics. For example, simulating the discomfort of another individual's tense/shaky posture could promote automatic empathic responding (i.e., making the other individual feel better would make you feel better [Lamm et al. Reference Lamm, Decety and Singer2011]). In contrast, the explicit semantic inference that what another person is feeling corresponds to the concept FEAR can facilitate important social inferences about both the past and the future by drawing on emotion knowledge, such as the likely causes and consequences of that state (e.g., “it was likely caused by a perceived threat” and “the person is now likely to try to avoid that threat”) – which can then inform subsequent decision-making (Fig. 1).
Figure 1. Depiction of “thinking through others’ emotions” as an extension of the “thinking through other minds” framework. Based on the (implicitly or explicitly) perceived thoughts and feelings of others, quick/involuntary somatovisceral (e.g., valenced changes in facial expression, body posture, and autonomic state) and cognitive (e.g., selective attention) policies are enacted and perceived by both self and others. Subsequent inferences about one's own emotional state and the emotional states of others (both implicit and explicit) then further inform sociocultural decision-making (e.g., conforming to social norms and engaging in cultural practices).
Successfully navigating the hypersocial human niche requires iterative and nested use of these processes – in which one's goals can only be accomplished by predicting the emotions of others in the future under different courses of action (e.g., “he's feeling angry because I'm driving below the speed limit – if I speed up then he will calm down”). This in turn leads to complex and continuous feedback loops that can facilitate both adaptive (Smith et al. Reference Smith, Weihs, Alkozei, Killgore and Lane2019c) and maladaptive (Smith et al. Reference Smith, Alkozei, Killgore and Lane2018a) social dynamics. The resulting dynamics lead to a type of “thinking through others’ emotions” that depends jointly on interoceptive and exteroceptive inference and on bidirectional brain-body interactions across individuals throughout a society. We put forward these additional affective dynamics as essential to completing the authors’ account of thinking through other minds.