Developing Intelligence

The organization of the human prefrontal cortex (PFC) is a lasting mystery in cognitive neuroscience, but not for lack of answers – the issue is deciding among them, since all seem to characterize prefrontal function in very different but apparently equally-valid ways. If this mystery were resolved, it could revolutionize cognitive neuroscience and neuropsychology as well as education and artificial intelligence – prefrontal cortex is crucial for the most high-level cognitive abilities we know of, including the executive functions, but no existing theory of intelligence specifies the exact neuronal computations which give rise to them.

Adding to the number of potential organizational schemes is a new update to a theory by Corbetta, Shulman and colleagues which puts the concept of attention first, contrasting with dominant theories of PFC organization which posit timespan-, abstraction-, hierarchical-, and/or content-based maps or gradients.

Following the exposition of Corbetta, Patel, and Shulman, here are the basic points:


1) To attend to rapidly changing goal-relevant information and yet not preclude orienting to novel or unexpected (e.g., threatening) information, the brain makes use of two separable attention systems: a “dorsal frontoparietal attention network” and a “ventral frontoparietal attention network.”

2) The dorsal attention network generates endogenous, top-down, goal-driven attention to select stimuli and link them to appropriate responses. This is particularly apparent following instructive cues, which selectively activate this network. Consists of intraparietal sulcus (IPS), superior parietal lobe (SPL), and dorsal PFC (dlPFC).

3) The ventral attention network is responsible for exogenous shifts of attention, or stimulus-driven attentional reorienting, accomplished by “circuit breaking” the dorsal network. At a high level, this can be desrcibed as “monitoring whether the sensory input departs from a current model of ongoing behavior.” The ventral network “detects salient and behaviorally relevant” stimuli by accumulating evidence for them in a way reminiscent of diffusion models; one evidence surpasses a decision threshold, a phasic norepinephrine response from the locus coeruleus yields a P300, which “interrupts” (see here), “resets” (see here) or updates (see here) the dorsal attention network. The network consists of right-lateralized inferior and middle frontal gyri (rIFG & rMFG), insula, frontal operculum, and temporal parietal junction (TPJ).

4) Activity in the ventral network is enhanced by low-probability stimuli, behaviorally-relevant items: controversy exists over how much highly “distinctive but entirely irrelevant can attract our attention,” and the authors imply that the non-target related P3a may occur only when subjects have no ongoing task. This goal-relevant selectivity is due to a top-down bias from the dorsal network, via MFG, which may ensure that task-relevant items alone can trigger a ventral reorienting response.

5) The temporal dynamics of these processes are unclear, with some evidence suggesting that ventral activity is too slow to be a trigger for dorsal activity. On the other hand, granger causality estimates suggest that ventral activity does preceede dorsal activity, and TMS of ventral areas disrupts orienting. The authors also (somewhat cryptically) suggest that synchrony-based effects may allow even a transient, weak interrupt/reset/updating signal might quickly disrupt attention if it takes the form of an attractor in a dynamical system, with reference to this paper.

In summary, this theory of PFC organization addresses frontal & parietal interactions, the ventral/dorsal distinction, and the interactions of the “default” and “task-networks” in terms of these two parieto-frontal systems. The theory implies they could be distinguished in terms of temporal dynamics (since the ventral reorienting response should be transient, whereas the dorsal system acts through sustained top-down control), in terms of response eligibility (since instructive cues activate only the dorsal system, whereas only behaviorally-relevant targets activate the ventral system), and in terms of monitoring/selection (since the dorsal system selects, whereas the ventral system monitors candidate information for possible selection).