External and Internal Attention: One Selective Move in Two Directions

Pay attention to a single voice at a loud party and the other voices drop away. You have not turned them off; your ears still receive them. You have drawn a line around one stream and let the rest fall outside it. That line is what attention is: a boundary between what gets selected for full processing and what does not. The part worth noticing is that the same move runs in two directions, outward and inward, and they look like one operation.

External attention: how the brain selects in the world (Posner's spotlight and biased competition)

The well-mapped direction is outward, attention to what the senses deliver. Michael Posner (1980) gave the classic picture, an attentional spotlight that orients to a location and sharpens processing there; a cue that pulls attention to a spot speeds up whatever appears at that spot. Robert Desimone and John Duncan (1995) supplied the mechanism underneath, biased competition: when several things fall on the same patch of visual cortex they compete to drive the neurons, and attention is the bias that tips the competition toward one of them. Both describe the same act at two levels of zoom: Posner the behavior, Desimone and Duncan the neural competition underneath. Attention selects one stream from the perceptual field and draws the boundary that leaves the rest outside.

Internal attention: selecting among your own thoughts, memories, and rules

Attention also turns inward, and this is not a metaphor borrowed from the outward case. Marvin Chun, Julie Golomb, and Nicholas Turk-Browne (2011) organized the field around exactly this split. External attention selects among sensory inputs: locations, moments, features, objects. Internal attention selects among internally generated contents: a memory to hold, a rule to apply, a response to ready, a thought to keep in mind through the next one. The operation is the same, selection of some content for privileged processing and exclusion of the rest, while the material it works over is inside rather than outside.

So the two modes are one move in two directions. Outward, it draws a boundary in the sensory field. Inward, it draws a boundary in the field of one's own generated contents.

The inward edge, marked speculative

There is a far end of the inward mode that I want to set down lightly, as the speculative half of this picture and not as load-bearing. Internal attention can take as its content the system's own ongoing states, attention turned on attention, a short self-directed loop. I am not going to build anything on this here, and in particular I draw no conclusion about self-awareness from it. The cautious version is only that the inward mode admits a setting in which what is selected is the processing itself, and that this would be continuous with ordinary internal attention rather than a separate faculty. How far it goes is left open; the account below rests entirely on the two ordinary modes.

How to test it: do external and internal attention share one mechanism?

Saying the two modes are one move is really two claims, and it helps to keep them apart. One is that they are the same kind of operation. The stronger one, the one worth betting on, is that they draw on the same single pool of resources rather than two separate pools that merely look alike. The same operation could in principle run twice on two private budgets, so a shared budget is an extra commitment, and that is the version this prediction targets.

If outward and inward attention share one budget, they should share more than a name. They should draw on the same limited capacity, lean on the same control machinery, and charge the same toll to cross between them: switching attention from something out in the world to something held in your head should cost about as much as switching within either domain, which is what a single control set would do. The prediction, then, is that loading external attention and loading internal attention interfere through one shared bottleneck.

The honest part: this test refutes much harder than it confirms. If the two turn out fully separable, each able to run at full strength while the other is saturated, then "one move in two directions" is wrong and they are two faculties that happen to share a word. That is a clean kill. But the reverse is murkier. Finding that both lean on the same frontoparietal regions and both hit a capacity ceiling does not by itself pick out this idea, because a rival picture, two genuinely separate faculties that both draw on one general executive resource, predicts the very same overlap. To tell them apart you need a sharper signature than co-location: a single bottleneck where loading one side eats into the other in proportion, the symmetric trade-off you see in dual-task interference, plus the cross-domain switch cost matching within-domain switching. The reach of the claim is settled by whether those specific signatures hold, not by overlap alone.