A Role for Uncertainty in the Neural Distinction Between Social and Nonsocial Thought

Social Cognition Engages a Consistent Set of Brain Regions: The Mentalizing Network

An abundance of research has been interpreted as evidence that different cognitive processes might support thinking, reasoning, and problem-solving in social compared with nonsocial contexts. Behaviorally, some neuropsychological conditions, such as autism spectrum disorder (ASD), have been associated with atypical responses on social-cognitive tasks despite typical performance in nonsocial tasks that are otherwise thought to have a similar structure (Baron-Cohen, 1990; Baron-Cohen et al., 1985; Fein et al., 1986), whereas others, such as Williams syndrome, are associated with atypicalities in cognition primarily outside the social domain (Karmiloff-Smith et al., 1995). At the neural level, studies using functional MRI (fMRI) have repeatedly observed preferential activation during social compared with nonsocial tasks in a particular set of brain regions, including aspects of the DMPFC and ventromedial prefrontal cortex (VMPFC), the temporal-parietal junction (TPJ), the superior temporal sulcus (STS), and the precuneus or posterior cingulate (Amodio & Frith, 2006; Gallagher et al., 2002; Gallagher & Frith, 2003; Jenkins & Mitchell, 2011; Molenberghs et al., 2016; Saxe, 2006; Spreng et al., 2009; Van Overwalle, 2009), even when the nature of the tasks is otherwise highly similar. For instance, these regions respond more when participants make inferences about others’ beliefs than when they make inferences about mechanical processes (Saxe & Kanwisher, 2003), when they infer others’ emotions than when they make physical judgments (Ochsner et al., 2004), and when they play a game against a human counterpart than when they play against a computer (Rilling et al., 2004). The activation of these regions in social compared with nonsocial tasks has been so consistent that they are now known collectively as the mentalizing network (Frith & Frith, 2006).

The Mentalizing Network and Domain Specificity

Evidence that a particular set of brain regions is consistently more engaged during mentalizing than other cognitive tasks is broadly consistent with two main interpretations. One possibility is that there are cognitive processes specialized for aspects of social-information processing (i.e., that cannot be reduced to the cognitive processes associated with nonsocial-information processing; Adolphs, 2003; Stone et al., 1998). According to this domain-specific view, regions of the mentalizing network may have evolved to subserve and are preferentially activated during inferences with social content (Baron-Cohen, 1995; Leslie, 1987; Tomasello, 1999). If true, this could either be because social inferences impose unique processing demands on cognition that are not present in nonsocial contexts (e.g., the need to construct representations of mental states) or because information content in the social domain (e.g., personality traits, beliefs, and/or emotions) is represented separately from nonsocial content even if it is similar in structure (Adolphs, 2009; Spunt & Adolphs, 2017).

An alternative possibility is that the activation of the mentalizing network during social inferences primarily reflects the engagement of domain-general processes that are disproportionately applicable, but not unique, to the social domain because of basic properties that tend to be more associated with, but not strictly limited to, social thought (Adolphs, 2009; Baetens et al., 2013; Heyes, 2014; Karmiloff-Smith, 2015). Of course, the entirety of the mentalizing network need not necessarily be characterized in the same way; it could be the case that some activations are best understood as reflecting domain-general and others as reflecting domain-specific operations.

Can Activation in the Mentalizing Network Be Explained by Uncertainty?

One basic property that might generally differentiate social from nonsocial settings is uncertainty. Whereas phenomena in the nonsocial domain are often fairly objective and deterministic, phenomena in the social domain are often more subjective and probabilistic (Dennett, 1987; FeldmanHall & Shenhav, 2019; Jenkins & Mitchell, 2010; Karmarkar & Jenkins, 2020; Mitchell, 2009). In particular, uncertainty in the social domain may stem from a need to make predictions about others on the basis of indirect information about their thoughts, beliefs, and emotions (i.e., mental states) and personality characteristics. It is this ability to mentalize about others, or engage theory of mind, that enables perceivers to reason about others’ underlying mental states and traits even as they diverge from overt behavior. Yet in part because underlying states and overt behavior can indeed diverge, such reasoning is necessarily shrouded in at least some uncertainty. In turn, this raises the possibility that at least some findings previously interpreted as evidence for the domain specificity of social-cognitive processes might instead be explained by the engagement of domain-general processes associated with uncertainty and/or its reduction.

In this perspective, we propose uncertainty as a lens through which to reconsider past comparisons of social and nonsocial cognition. First, using independent ratings of the uncertainty associated with social and nonsocial tasks used in past fMRI experiments, we find evidence that social tasks in the fMRI literature have higher levels of uncertainty than their nonsocial counterparts. Next, we consider to what extent elevated levels of uncertainty might plausibly account for the observed activation of any of the brain regions of the mentalizing network during tasks with social content. We find evidence of a domain-general role for the DMPFC in uncertainty processing and/or reduction during social inferences. Finally, we examine the possibility that the type of uncertainty that characterizes the social domain tends to be more epistemic in nature (i.e., that it arises more from limited knowledge of an existing reality than from randomness) and suggest a role for DMPFC in epistemic uncertainty processing across both social and nonsocial inferences.

Levels of Uncertainty in the Social Versus Nonsocial World

We first reviewed past fMRI investigations that compared mentalizing tasks (“social” tasks) to tasks that do not require mentalizing (“nonsocial” tasks) to assess the degree to which they differed in uncertainty. Although tasks in social and nonsocial conditions are often well matched in terms of their basic structural properties, it is possible that inferences about other minds are generally less constrained by external information than those involving mechanical events, corresponding to higher uncertainty in the social domain, on average (Jenkins & Mitchell, 2010). For instance, one study typical of this literature (Gallagher et al., 2000), in which participants were asked to make inferences about the reasons for social and nonsocial events, illustrates how the social events may have lent themselves to a larger number of explanations (e.g., a burglar might have turn himself in upon seeing a passing policeman for any of a variety of reasons) than the nonsocial events, which were arguably more constrained by known information (e.g., there is probably one main reason why, on a foggy day, an army with better foot soldiers would have won over the army with better air power). Similar differences in uncertainty could have plausibly arisen in other comparisons typical of this literature, including judgments of how pleased photographed people were to have their picture taken compared with judgments about how symmetrical the people’s faces are (Mitchell et al., 2005), inferences about photographed people’s emotional states compared with inferences about whether they are indoors or outdoors (Ochsner et al., 2004), judgments about whether certain traits describe a given person compared with semantic judgments regarding these traits (Schmitz et al., 2004), and playing a game interactively with others compared with playing alone (Elliott et al., 2006).

DMPFC activation tracks with independent ratings of uncertainty in social tasks

To gain empirical traction on the possibility that social tasks in past fMRI studies have had higher levels of uncertainty than their nonsocial counterparts, we recruited a set of 110 participants (62 females, 48 males; age: M = 19.8 years, range = 18–23) to read descriptions of tasks previously used in social cognition and uncertainty literatures and to evaluate the uncertainty therein (see Supplementary Information for methodological details and Table S1 for the full list of included studies in the Supplemental Material available online). Specifically, we included three types of fMRI studies: (a) those in which a social versus nonsocial comparison was made, (b) those in which uncertainty may have been manipulated incidentally between two conditions in the social domain (e.g., familiar vs. unfamiliar target), and (c) those in which uncertainty was explicitly manipulated in the nonsocial domain. In selecting these studies, our aim was to acquire an illustrative sample of each type of comparison rather than to compile an exhaustive catalogue of all such studies. Accordingly, we aimed to identify approximately 10 studies that exemplified the manipulations in each of the three categories above. Each participant read 14 task descriptions that corresponded to two conditions each from a random subset of seven out of 29 studies. Participants rated the level of uncertainty in each condition separately on a scale of 0 to 100, anchored by 0 = no uncertainty and 100 = high uncertainty (see Fig. S1 in the Supplemental Material).

To examine whether participants perceived the social conditions to have higher levels of uncertainty, we ran a linear mixed-effects model in which task condition was entered as a fixed effect and participant and study were entered as random effects. Results indicated that social conditions were associated with greater uncertainty, on average, than their nonsocial comparison conditions. Specifically, participants rated the level of uncertainty to be higher in social (M = 52.07) than nonsocial conditions (M = 40.65), t(422) = 4.83, p < .001 (Fig. 1).

OPEN IN VIEWER

Fig. 1.

(a) Peak medial prefrontal cortex, precuneus, and right temporal-parietal junction activations reported in the original studies contrasting social (mentalizing) and nonsocial (nonmentalizing) task conditions in the social cognition literature. (b) The average uncertainty ratings by the independent sample of participants for these conditions. Participants rated social tasks to have higher levels of uncertainty compared with the nonsocial comparison tasks (***p < .001).

Uncertainty in the Nonsocial World

To what extent can the different degree of uncertainty observed between social and nonsocial cognition account for the differences in brain activations during social versus nonsocial tasks? To address this question, we examined to what degree modulations of uncertainty outside the social domain are associated with activations in brain regions typically implicated in social cognition. That is, if the reason for the engagement of a brain region in social compared with nonsocial tasks is that social tasks are associated with higher uncertainty, then that brain region’s response should also be associated with uncertainty more generally, including outside the social domain.

Past studies outside the social domain have observed uncertainty-related activations in a variety of brain regions, including the insula, dorsal anterior cingulate cortex, caudate, precuneus, dorsolateral prefrontal cortex, medial prefrontal cortex (MPFC), orbitofrontal cortex, posterior parietal cortices, and amygdala (Demanuele et al., 2015; Farrar et al., 2018; Krain et al., 2006; McGuire et al., 2014; Monosov, 2020; Morriss et al., 2019; Nassar et al., 2019; White et al., 2014). These findings derive from a wide range of tasks including but not limited to anticipation of stimuli with uncertain valence (e.g., Klumpers et al., 2015), anticipation of uncertain rewards (e.g., Bjork & Hommer, 2007), probabilistic learning (Bhanji et al., 2010; Volz et al., 2003, 2004), learning dynamic environmental contingencies (McGuire et al., 2014; Nassar et al., 2019), prediction under incomplete information (e.g., Elliott et al., 1999), and gambling (Cohen et al., 2005).

To better examine potential overlaps between brain regions associated with uncertainty and those associated with social cognition, one should take into account whether the type of uncertainty is the same in both literatures. Although both social and nonsocial inferences may be associated with both epistemic and aleatory uncertainty to some degree, we suggest that tasks typical of the social-cognition literature primarily invoke greater epistemic uncertainty than their nonsocial counterparts (Kruglanski, 1990). In typical social-inference tasks, there is generally a “true” state of the world (e.g., another person’s belief, preference, or emotion) about which participants’ uncertainty arises largely because of their lacking direct access to the hidden contents of another person’s mind. Accordingly, we were especially interested in whether any brain regions associated with social cognition are also associated with epistemic uncertainty outside the social domain.

DMPFC activation tracks with independent ratings of uncertainty in nonsocial tasks

To what degree does DMPFC activation in past studies of nonsocial cognition track with epistemic uncertainty? To explore this question, we identified for each study in the nonsocial-uncertainty literature the condition in which higher and lower DMPFC activation had been observed. We then compared our participants’ uncertainty ratings across these conditions. In these nonsocial tasks, conditions under which higher DMPFC activation was observed were rated as having higher levels of uncertainty (M = 64.47) than those under which lower DMPFC activation was observed (M = 36.03), t(376) = 11.61, p < .001 (red circles in Fig. 2b).

OPEN IN VIEWER

Fig. 2.

(a) Peak middle frontal activation coordinates for studies from the social-cognition literature in which there was a social versus nonsocial contrast (green), social cognition literature in which uncertainty was incidentally manipulated (blue), and nonsocial-cognition literature in which uncertainty was directly manipulated (red). (b) Uncertainty ratings for conditions associated with lower and higher dorsomedial prefrontal cortex (DMPFC) activation showing that participants perceived conditions under which higher DMPFC activation was observed to have higher levels of uncertainty (***p < .001).

Together, this evidence suggests that DMPFC, a brain region that has been associated with social cognition, also responds more to nonsocial-task conditions with higher compared with lower uncertainty. Although the type of uncertainty that is modulated was not always explicitly labeled in these nonsocial studies, an informal, retrospective categorization suggests that the uncertainty modulated across task conditions associated with differences in DMPFC may have tended to be more epistemic in nature. Given the argument that social conditions tend to have higher levels of epistemic uncertainty, this observation points to the intriguing possibility that activations in DMPFC under social conditions do not reflect social processing per se but, instead, reflect processing of epistemic uncertainty. Accordingly, we next investigated to what extent this region emerged from past comparisons in the social-cognition literature that may have incidentally manipulated uncertainty.

Uncertainty in the Social World

Although few studies to date have manipulated uncertainty deliberately within the social domain, at least one study observed that more uncertain inferences about others’ mental states were associated with greater activation in DMPFC. In this study (Jenkins & Mitchell, 2010), participants answered questions about scenarios that supported either a more or less certain inference about the character’s beliefs or preferences. For both belief and preference inferences, higher activation was observed in DMPFC (and VMPFC) in response to scenarios with higher (vs. lower) uncertainty. This pattern was not observed in other regions of the mentalizing network, including the TPJ, which suggests that the elevated response in DMPFC did not arise simply from overall mentalizing demand.

The possibility that uncertainty can account for some of the differences observed between social and nonsocial cognition draws further support from studies that, in retrospect, could be interpreted to have manipulated the level of uncertainty incidentally in the social domain even though manipulating uncertainty was not the explicit goal of those studies. For example, DMPFC has been found to respond more during inferences about the preferences of dissimilar others (whose preferences may be more uncertain) than similar others (Mitchell et al., 2006; Tamir & Mitchell, 2010). More generally, activation in this brain region has been shown to increase with the social distance of the target (Amodio & Frith, 2006; Gallagher et al., 2002; Gallagher & Frith, 2003; Jenkins & Mitchell, 2011; Mitchell et al., 2006; Molenberghs et al., 2016; Van Overwalle, 2009).

DMPFC Responds to Uncertainty Across Domains

If DMPFC plays a domain-general role in uncertainty processing and/or reduction, its response should be related to uncertainty not only in each domain but also when collapsing across domains. To investigate this, we next examined whether conditions associated with higher DMPFC activation, regardless of domain, were perceived to involve more uncertainty by our independent rater participants. To do this, we compared the uncertainty ratings between conditions across all studies in which a differential DMPFC response was observed. In line with the idea that DMPFC is involved in uncertainty processing in both social and nonsocial domains, conditions in which higher DMPFC activation had been observed were rated as significantly more uncertain (M = 59.24) than those in which lower DMPFC activation had been observed (M = 37.82), collapsing across all tasks, t(1316) = 15.82, p < .001 (Fig. 2). Together, these results suggest at least tentatively that DMPFC activation tracks with the uncertainty around an inference regardless of the domain in which that inference is made.

Discussion

Previous research on social cognition has revealed neural circuitry that is more engaged during mentalizing than during a variety of nonsocial comparison tasks (Adolphs, 2003, 2006, 2009; Amodio & Frith, 2006; Jenkins & Mitchell, 2011; Molenberghs et al., 2016; Saxe, 2006; Van Overwalle, 2009). This evidence has been used to support the hypothesis that mentalizing is subserved by specialized cognitive processes that may have evolved specifically to support the kinds of inferences and predictions about others’ hidden mental states and personality traits that enable humans to navigate their social environments (Spunt & Adolphs, 2017). However, an alternative explanation for these findings is that social cognition disproportionately recruits a set of domain-general cognitive processes that are not dedicated to but happen to be disproportionately useful for inferences about social compared with nonsocial phenomena. With this in mind, we hypothesized that if inferences in the social world are characterized by greater uncertainty, a domain-general mechanism that processes and reduces this uncertainty could account, at least partly, for previously observed differences in activation between social and nonsocial cognition. Independent ratings of stimuli from past studies suggest that the kinds of stimuli typically used to investigate social cognition are associated with more uncertainty, on average, than their nonsocial counterparts. This is consistent with the possibility that the social domain may be more uncertain in general, perhaps because of the inherent unobservability of other people’s mental states.

Although it is generally appreciated that inferences about other minds may be relatively indirect or imprecise (Dennett, 1987; FeldmanHall & Shenhav, 2019; Jenkins & Mitchell, 2010; Mitchell, 2009), to our knowledge, no research has investigated empirically the possibility that uncertainty could explain the patterns of brain activation that consistently distinguish social from nonsocial thought. To evaluate whether uncertainty could explain any of these differences, we compared neural responses associated with social (vs. nonsocial) cognition with those associated with higher (vs. lower) uncertainty across past fMRI experiments. This approach revealed that one region consistently associated with social cognition, DMPFC, also emerges consistently from the uncertainty literature (see Fig. 2). More specifically, the activation of DMPFC across these literatures appears to be most consistent with a role in the processing and/or reduction of epistemic uncertainty (i.e., uncertainty that is internal and theoretically reducible). This would be consistent with a hypothesized role for the DMPFC in activities related to imagination or mental simulation (Addis et al., 2009; Andrews-Hanna, 2012; Buckner & Carroll, 2007; Hassabis & Maguire, 2009; Jenkins & Mitchell, 2010; Spreng et al., 2009).

At the same time, this exploratory effort provides only a fairly indirect test of the broader hypothesis because it relies on the nature and degrees of uncertainty that happen to have accompanied the tasks used in past research. To the extent that epistemic uncertainty explains the apparently selective activation of DMPFC in social compared with nonsocial tasks, differences in this region’s activation between social and nonsocial domains should be attenuated when epistemic uncertainty is held constant—a possibility open to direct empirical test in future research.

This analysis is consistent with the possibility that uncertainty might explain the engagement of DMPFC in mentalizing tasks, but uncertainty does not appear to account for the activation of other regions associated with mentalizing (e.g., TPJ, STS, precuneus). This could indicate that activation in regions other than DMPFC is indeed best characterized as specific to the social domain, a possibility that draws some support from developmental findings that the right TPJ, but not the medial prefrontal cortex, becomes highly selective for certain types of mental-state content across childhood (Gweon et al., 2012). Another possibility, somewhat less plausible, is that activity in these brain regions also relates to uncertainty, but in a way not captured by the studies discussed here. A third possibility is that the response of these brain regions may be reducible to domain-general properties other than uncertainty, such as salience, information sparsity, abstractness, or others, that might also differentiate social from nonsocial conditions in past experiments (Baetens et al., 2013). Future research is needed to better understand the domain specificity or domain generality of other brain regions associated with social thought.

Relatedly, although we observe that activations from past social cognition and uncertainty studies converge in the DMPFC, it is still possible that different DMPFC subregions may show different response profiles. In particular, an inspection of the peak-activation coordinates observed across these studies tentatively suggests that there may be a spatial gradient in DMPFC whereby more posterior subregions may be associated more with nonsocial uncertainty and more anterior subregions may be associated more with social processing and/or social uncertainty. Relatedly, aspects of DMPFC have been associated with response, decision, and strategy conflict, which may signal a greater need to reduce uncertainty (Venkatraman et al., 2009). Although intriguing, it is not possible to draw strong conclusions about this possibility with the available evidence given that (a) previous studies did not directly manipulate and compare social and nonsocial uncertainty and (b) peak-coordinate information necessarily obscures the full extent of the activations, which may indeed overlap (or may not). Future studies that can compare across conditions in the same set of participants are needed to address the degree to which social and nonsocial uncertainty are associated with activations in overlapping versus distinct DMPFC subregions.

To the extent that DMPFC plays a domain-general role in cognition associated with uncertainty, what might that role be? One possibility is that DMPFC activation simply reflects participants’ subjective feelings of uncertainty. If that were the case, we would expect this region to respond more under higher than lower uncertainty conditions independent of whether the uncertainty is aleatory or epistemic in nature. This prediction is inconsistent with our assessment that DMPFC responds more under conditions with higher epistemic uncertainty but not aleatory uncertainty. A second, related possibility is that DMPFC tracks with the difficulty of holding in mind multiple possible alternatives, which is correlated with uncertainty. This, too, seems unlikely, however, given that higher working memory load outside the social domain is typically associated not with activations but, rather, with deactivations of cortical midline structures, including DMPFC (Arsalidou et al., 2013; Gusnard & Raichle, 2001; Jenkins, 2019; Raichle, 2015), and that working memory demands inside and outside the social domain have been found to have opposite, rather than similar, effects on DMPFC activity (Meyer et al., 2012).

With this in mind, we believe that perhaps the most likely explanation is that DMPFC activation reflects a process that contributes to the reduction of uncertainty across social and nonsocial contexts. A variety of possible strategies can be used to reduce uncertainty, ranging from applying heuristics or stereotypes to imagining or simulating possible scenarios. The tendency for the social world to be characterized by more uncertainty may make some of these processes disproportionately applicable to, even if not specialized or unique to, social compared with nonsocial thought. In particular, the possibility that DMPFC activation reflects domain-general cognitive processes related to uncertainty reduction dovetails with the possibility that activation in DMPFC may track with epistemic uncertainty, which stems from incomplete knowledge and is reducible through inference. This possibility is especially plausible given the observation that activation in MPFC is associated with imagination and memory (Addis et al., 2009; Andrews-Hanna, 2012; Buckner & Carroll, 2007; Hassabis & Maguire, 2009; Spreng et al., 2009) as well as the integration of contextual and self-related information (Wittmann et al., 2016, 2021), both of which could contribute to the reduction of uncertainty. Likewise, a recent meta-analysis observed convergence in the DMPFC between metacognition (thinking about one’s own mind) and mentalizing (thinking about the minds of others; Vaccaro & Fleming, 2018). Further work is needed to better understand the nature of the core cognitive processes that support uncertainty reduction, but this existing evidence is consistent with a possible role for DMPFC in integrating information to generate novel predictions (Jenkins & Mitchell, 2010).

By suggesting a reconceptualization of mentalizing processes and DMPFC, this perspective makes contact with current evidence related to disorders of social function. For example, although ASD is often associated with deficits in mentalizing, or theory of mind (Baron-Cohen, 1995), some accounts suggest more domain-general cognitive deficits in ASD (Rajendran & Mitchell, 2007). Individuals on the autism spectrum have been found to exhibit higher than average aversion to uncertainty (Vasa et al., 2018), which is intriguingly consistent with the possibility that atypical performance on social tasks, which tend to be more uncertain, could stem from atypicalities in the cognitive processes used to reduce uncertainty in both social and nonsocial contexts. This possibility makes further contact with reports that individuals with ASD may adopt a “systemizing” approach, cataloguing the rules and regularities that characterize different systems, more than an “empathizing” approach, using imagination and/or perspective-taking, to understand the social world (Baron-Cohen, 2002, 2009). Although systematizing might be a successful approach in low-uncertainty environments, it may fall short when environmental uncertainty is high. Whether a tendency to systematize ultimately stems from a preference for structure, an aversion to uncertainty, or some combination, this tendency could help to explain what appear to be specifically social deficits. Finally, research also suggests a link between intolerance to uncertainty and social anxiety disorder (SAD; Boelen & Reijntjes, 2009). Intriguingly, functional abnormalities have been observed in MPFC in both ASD (Kennedy et al., 2006) and SAD (Sripada et al., 2009). Although abnormalities in social cognition and uncertainty processing have been treated separately, the present analysis suggests that they could share a common root, which may be a fruitful direction for future research.

Conclusion

Here, we put forth the possibility that different levels and/or types of uncertainty might differentiate social from nonsocial inferences, on average, which suggests a possible reinterpretation of the engagement of brain regions during social compared with nonsocial thought. Our analysis suggests that social conditions tend to be more uncertain than nonsocial conditions in past research and further suggests that the DMPFC, one of the brain regions most consistently associated with social cognition, might play a domain-general role in uncertainty processing that spans both social and nonsocial domains.

Supplemental Material