Uncertainty, Not Mental Content, Drives Dorsomedial Prefrontal Engagement during Inferences about Others

Abstract

To navigate social life, humans make inferences about the intentions, beliefs, emotions, and personalities of other people, i.e., they mentalize. A network of brain regions consistently engages more during mentalizing than during carefully controlled comparison tasks, sometimes cited as evidence of domain-specific mentalizing processes. Here we investigated the possibility that engagement of these regions during mentalizing may be due to uncertainty. We scanned 46 participants (33 female, 13 male) using fMRI as they made mental and non-mental inferences (about human minds, human bodies, and physical objects) under varying levels of uncertainty. Uncertainty explained activation in a key region of the mentalizing network: the dorsal medial prefrontal cortex (DMPFC). Higher uncertainty was associated with greater DMPFC engagement across conditions, and, when controlling for uncertainty, DMPFC engagement did not differentiate mental from non-mental inferences. Results suggest that the apparently selective DMPFC engagement during social inference may be better understood as a response to uncertainty, which is often elevated in social contexts, with implications for the cognitive architecture of the social brain and disorders of social function.

Keywords: default mode network; domain generality; domain specificity; mentalizing; social cognition; social neuroscience.

Figure 1.

Stimuli and experimental paradigm. a, Stimulus examples and sample trial. Participants made 216 total inferences across three categories in a pseudorandomized, interleaved fashion. Participants first viewed a characteristic of a person or piece of furniture (the given characteristic); then, they reported how likely it was that a second characteristic (the inferred characteristic) described the same person or piece of furniture. b, Uncertainty ratings collected from independent raters were strongly correlated with ratings from the participants in the fMRI study across personality (r = 0.80), physical (r = 0.87), and furniture (r = 0.72) conditions (all ps < 0.001). Points indicate the uncertainty scores for each pair of characteristics (e.g., smart, sincere), averaged across participants.

Figure 2.

Whole-brain analysis: parametric response to uncertainty as measured by independent ratings across conditions in DMPFC. a, DMPFC cluster resulting from a whole-brain analysis of the parametric effect of uncertainty across conditions (whole-brain FWE-corrected p < 0.05). b, Parameter estimates for the parametric effect of uncertainty in each condition individually extracted from the DMPFC cluster in panel a. Error bars represent the standard error of the mean. Points indicate beta estimates for each participant in each condition.

Figure 3.

Parametric response to two additional operationalizations of uncertainty across conditions in DMPFC. DMPFC clusters resulting from whole-brain analyses of the parametric effect of uncertainty across conditions as measured by a. Judgment extremity of the participants, where we contrasted trials on which participants chose more uncertain options (i.e., “somewhat likely” and “somewhat unlikely”) with those on which they chose more certain options (i.e., “extremely likely” and “extremely unlikely”). b, Semantic distance between characteristics in each pair where semantically distant pairs were considered to have higher uncertainty (cluster-level FWE correction, p < 0.05).

Figure 4.

A priori region-of-interest (ROI) responses to uncertainty in the mentalizing network. a, Independent ROIs associated with mentalizing (Dufour et al., 2013) for DMPFC (blue), PCC (green), left TPJ (orange), right TPJ (purple), and VMPFC (pink). b, Beta estimates for the parametric effect of uncertainty across conditions extracted from the independent ROIs illustrated in panel a. Only DMPFC activation increased with uncertainty. Error bars represent the standard error of the mean; asterisks mark significant differences from zero after Bonferroni’s correction (***corrected p < 0.001, **corrected p < 0.01).

Figure 5.

A priori region-of-interest (ROI) responses to uncertainty in the DMPFC, dACC, and pre-SMA. a, Independent ROIs associated with the DMPFC (blue; Dufour et al., 2013), dACC (purple; Gartner et al., 2024), and pre-SMA (green; Mayka et al., 2006). b, Beta estimates for the parametric effect of uncertainty across conditions extracted from the independent ROIs illustrated in panel a. In addition to the DMPFC, dACC showed a positive, but weaker response to uncertainty, whereas pre-SMA activation did not track uncertainty. Error bars represent the standard error of the mean; asterisks mark significant differences from zero after Bonferroni’s correction (***corrected p < 0.001, **corrected p < 0.01).

Figure 6.

Response during human personality versus human physical inferences when uncertainty is equated. a, Clusters from the whole-brain analysis that showed more activation during the human personality than the human physical condition without uncertainty confounds. We see no activations in the MPFC for this comparison, whereas we see higher activation in both left TPJ (x = −52) and right TPJ (x = 52; cluster-level FWE correction, p < 0.05). b, Beta estimates for the contrast between uncertainty-equated human personality and human physical conditions extracted from the independent mentalizing ROIs (Dufour et al., 2013; *corrected p < 0.05, **corrected p < 0.01). Left and right TPJ continue to show more activation during mental than physical inferences when uncertainty is not a confound. When uncertainty is equated, MPFC no longer distinguishes mental from physical inferences.