Consistent with this suggestion, a recent study found that although the majority of voxels in the fusiform face area (FFA, Kanwisher et al., 1997 and Kanwisher, 2010) was suppressed for a repeated face, a subset of voxels reliably showed the reverse pattern (de Gardelle et al., 2013), termed repetition enhancement (see also Turk-Browne et al., 2006 and Müller et al., 2013). Intriguingly, these two populations of voxels also showed different patterns of functional connectivity. It will be intriguing Torin 1 mw to test whether the STS, TPJ, PC, or MPFC similarly contain subsets of voxels with enhanced responses to predicted actions or beliefs, and whether these voxels have distinctive patterns of functional connectivity

with other regions, especially because unlike face processing, the direction of information flow among regions involved in theory of mind is largely

unknown. Second, because both predictor neurons and error neurons may have preferred stimuli (or stimulus features), it may be possible to identify the content of the prediction independent from the response to the subsequent stimulus. For example, the response of the FFA seems to increase Venetoclax when a face stimulus is predicted, as well as (and partially independent from) when a face stimulus is observed (den Ouden et al., 2010 and Egner et al., 2010). Note though that neither of the existing studies could fully independently identify the response to predicting a face, because in both cases, the probability of a face was exactly reciprocal to the probability of the only other possible stimulus, a house. By including a third category of stimulus, or a third possible cue, or by independently varying the predictive value of the two cues, it should be possible

to independently measure category-specific responses to the prediction of a category, versus the response all to that category when observed. Third, and relatedly, predictor neurons can signal the expectation of a stimulus that never occurs. In some cases, the absence of an expected stimulus should generate error activity (den Ouden et al., 2010, Todorovic et al., 2011 and Wacongne et al., 2012). For example, the activity pattern in IT generated by the surprising absence of an object contains information about the identity of the absent stimulus (Peelen and Kastner, 2011). Unlike the “signed” (i.e., below baseline) error response in reward systems, sensory neurons thus seem to show an increased response to an unexpectedly absent stimulus (though note that there is some disagreement as to whether this activity is driven only by the prediction signal before the stimulus is expected to appear, or by a combination of the prediction signal with a subsequent error signal when the stimulus fails to appear, e.g., den Ouden et al., 2010). Fourth, the prediction and the error signals could be separable in time.