UCLA

Functional magnetic resonance imaging (fMRI) has built the prevailing picture of the social brain. However, fMRI’s hemodynamic basis collapses two dimensions of neural activity that electrophysiology preserves: the millisecond timing of activity and its distribution across the frequency spectrum. This dissertation uses human intracranial electrophysiology (iEEG; N=34) to recover those dimensions for mentalizing – reasoning about mental states – and to reconcile the resulting picture with the fMRI literature it appears at points to contradict. Paper I, using high-frequency broadband (HFB) activity as a proxy for population spiking, shows that the default mode network's apparent simultaneity during mentalizing is in fact a posterior-to-anterior sequence, from visual cortex through temporoparietal DMN to medial prefrontal cortex. Self- and other-mentalizing engage near-identical populations distinguished by the timing – not the magnitude – of their activity, recasting the "other-selective" dmPFC of fMRI as a difference in duration mistaken for a difference in intensity. Paper II decomposes this activity across the full spectrum, revealing a dual-phase architecture shared near-identically by mentalizing and autobiographical memory and jointly distinct from arithmetic. Arithmetic is the spectral outlier: it suppresses HFB in dmPFC while driving sustained theta in the same sites – a signature consistent with frontal-midline theta, the canonical correlate of cognitive control. Because BOLD couples to HFB and is largely blind to low-frequency power, the medial-prefrontal "deactivation" reported during cognitive tasks is the HFB-suppression half of a dissociation whose low-frequency half moves oppositely; the deactivation is faithfully measured but does not amount to disengagement. Paper III uses spectral cross-classification to show that mentalizing recruits autobiographical-like computation along the same hierarchy, graded (Autobiographical > Self ≈ Other > Cognitive) and short of full autobiographical engagement, with self and other differing in timing rather than content. Together, the papers show that what fMRI represents as one activation is a sequence, a spectrum, and a computation – dimensions that, as paper IV argues, source-localized EEG and MEG can recover non-invasively and at scale. Restoring them does not overturn the fMRI view of the social brain; it recovers what the map leaves out and, in one case, reverses the sign of the inference the map supports.