Neuroeconomics and the metastable brain.

been defined in various ways, including the level of embedding in a goal–subgoal hierarchy during problem-solving [6], the number of relations being simultaneously processed during reasoning [7], or the number of items held in working memory [8]. Reviews of RLPFC recruitment across multiple domains [2,9] show that activations are more frequent when the complexity of cognitive processing is high than when it is relatively low. This selectivity would permit reverse neuroimaging inference to a much greater extent, especially when specific cognitive processes such as relational integration [7], the evaluation of self-generated information [4], or subgoal processing [5,10] are theorized at the highest levels of cognitive complexity. Implications and questions for future research In summary, brain regions differ not only in their overall selectivity of response, but also in terms of the specific task characteristics they are selective to. This suggests that reverse inference can be improved by incorporating information about the relevant task characteristics into neu-roimaging databases and meta-analyses. At present, databases generally classify tasks according to their cognitive domain and contain virtually no information about the level of task complexity, possibly because complexity of processing can be difficult to quantify and compare across tasks. In addition, a number of other questions emerge: Are there other brain regions that show selectivity to the level of task complexity but lack selectivity to task domain? How can we compare levels of cognitive complexity across different cognitive domains? What other task characteristics, in addition to complexity and domain, might be relevant in determining the selectivity of brain regions? Clearly, much remains to be resolved. In the meantime, the framework presented by Poldrack places clear constraints on the inferences that can be drawn with the limited information that is currently available.