Holding the Line

Working Memory, Control, and the Dorsolateral Prefrontal Cortex

A different problem from a different cortex

The last section gave the animal a currency. It can now hold up unlike options — food, water, a mate, unexplored ground — and rank them on a common scale of worth, and the largest wins. Suppose it has done so. Suppose the contest is settled: the animal has chosen to cross the clearing for food. The valuing is done. And yet the hard part, in a sense, is only beginning — because between the choice and the food lies a stretch of time and a world full of distractions, and the goal must somehow stay in charge across all of it.

This is a genuinely different problem from valuation, and it is solved by a genuinely different part of the frontal lobe. Recall the two architectural trends we drew in the overview. The valuation work of the last section belonged to the ventral and medial cortex — the older, agranular trend that grew up with the brain’s visceral and emotional core, and evaluates options in the currency of bodily worth. The problem we turn to now belongs to the dorsal and lateral trend — the granular cortex that grew up with the sensory, spatial, and motor-planning world, and whose business is not what an option is worth but holding a chosen one in command of behavior over time. The ventral cortex answers which future is worth pursuing. The dorsolateral cortex answers how to keep pursuing it when the present keeps offering reasons not to.

Keep that division firmly in mind, because the two systems are so easily blurred under the single word “executive,” and because the whole point of separating them is that they fail in opposite-feeling ways. Damage the ventral system and the animal chooses badly — the values go wrong. Damage the dorsolateral system and the animal often chooses perfectly well and then never arrives, pulled off course by whatever is brightest or most habitual. The first is a failure of judgment; the second is a failure of holding the line. This section is about the line, about the cortex that holds it, and about the several ways behavior comes apart when it cannot be held.

The mechanism of holding: persistent activity

Begin with the most basic version of the problem, stripped of any distraction. An animal sees something it will need to act on — a location, say — and then the thing disappears. A moment later it must act on the now-absent information. Between the vanishing and the acting, the information has to be kept, held in some active form across a gap during which the world is no longer providing it. This is working memory: not the long-term storage of facts, but the temporary holding-online of information so that it can guide behavior in the absence of the thing itself. It is the simplest case of acting on the not-present, and it is the foundation on which everything else in this section is built.

The classic demonstration that the dorsolateral cortex does this — and the experiment that defined a career and much of a field — is Patricia Goldman-Rakic’s oculomotor delayed-response task. A monkey fixates a central point. A target flashes briefly somewhere in the periphery and then goes dark. The monkey must keep its eyes on the center through a delay of several seconds during which nothing marks the target’s former location, and then, when released, flick its gaze to where the target had been. Correct saccades earn juice. The task is trivial if the target stays visible and impossible if the memory of it is lost; what it isolates is precisely the holding of a vanished location across an empty delay.

What Goldman-Rakic found in the dorsolateral cortex is the cellular face of working memory, and it is worth seeing plainly because it is the section’s central object. Neurons fired during the delay — in the gap, when the target was no longer visible and the monkey was simply holding its location in mind. And different neurons fired for different locations: a given cell would sustain its activity through the delay when the remembered target had been in its preferred part of space, and stay quiet for targets elsewhere. These cells had memory fields — each one holding a particular location online by remaining active through the delay that represented it. When the activity faltered, the monkey’s memory faltered with it; the persistent firing was the held information.

Sit with what this is, because it is the anti-homunculus argument in its most concrete form anywhere in the unit. There is no executive in this picture who decides to remember the location and consults a stored note. There is a population of neurons that stays lit — a representation kept active across the gap, holding the information by the simple fact of continuing to fire. “Holding something in mind” is not the act of an inner agent attending to an inner screen. It is persistent activity: a pattern that, once kindled, sustains itself long enough to bridge the absence of the thing it represents. The delay cell is what working memory is made of, and it is a mechanism, not a homunculus’s deed.

The same machinery is visible, less directly, in humans. Dorsolateral cortex — particularly in the right middle frontal gyrus for spatial material — activates during the delay of a working-memory task, and its activity grows as more items are held, though it plateaus after only a few, a hint of the famously sharp limit on how much can be kept online at once. And there is a revealing detail: during the delay, activity also rises in the posterior cortical regions specialized for the kind of material being held — face-selective cortex when faces are remembered, place-selective cortex when places are. This suggests that the dorsolateral cortex is not a storehouse where the content itself sits, but something more like a spotlight that keeps the relevant posterior representations active — holding the content online by sustaining the very sensory regions that first represented it.

That reframing matters, and several researchers have pressed it: working memory may be best understood not as a separate buffer but as attention directed inward, to internal representations, using much the same frontal-parietal control machinery that directs attention outward to the world. On this view “holding something in mind” and “attending to something” are close to the same operation pointed in different directions — and the dorsolateral cortex’s job is the directing, not the storing. This is friendly to the anti-homunculus line rather than threatening to it: it removes even the storehouse, leaving only a mechanism that keeps the right representations active. There is no inner room where the remembered thing is kept. There is only a process that holds certain patterns lit against the dark.

Holding is also suppressing: the unity beneath the executive

It is traditional to list the dorsolateral cortex’s functions as a set of separate “executive” abilities — initiating behavior, shifting between rules, inhibiting responses, holding things in working memory — and then to admit, sheepishly, that these are almost impossible to pull apart in practice. To shift to a new rule you must inhibit the old one; to inhibit the old one you must be holding the new one in mind; the rule you are holding is in working memory. The list keeps collapsing into itself.

I want to suggest that this is not a nuisance but the central clue, and that the apparent list is really one function seen from several sides. The function is the one we have been describing: holding a chosen goal or rule in command of behavior over time. Everything on the executive list is a face of that single job.

Consider inhibition, usually treated as a capacity in its own right. To hold goal A in command is, necessarily, to suppress the responses that would serve some competing B — the prepotent action, the habitual reaction, the thing the last stimulus would have you do. Maintaining a goal and inhibiting its competitors are not two operations; they are the positive and negative faces of one operation, the way lifting one basal-ganglia program’s brake while clamping the others was a single act of selection in the last unit. There need be no dedicated “inhibitory module” that stops behavior, any more than there was a dedicated chooser in the striatum. “Inhibition” is what it looks like, from the outside, when a maintained goal holds its ground against a competitor. (Whether there is in fact a special circuit for outright stopping — a fast brake on action, perhaps in the ventrolateral cortex — is a real and open question, and we will keep it open. But much of what gets called inhibition needs no such dedicated stopper; it falls straight out of maintenance.)

Consider set-shifting, the ability to abandon one rule for another when circumstances demand. This is simply updating which goal is in command — releasing the maintained rule and installing a new one. The maintenance and the shifting are the same machinery operating in two modes: hold steady, or swap. Consider planning, the sequencing of actions toward a goal not yet reached. This is maintenance extended across a series — keeping the goal in command not for one step but through a whole chain of sub-steps, each of which must be held in turn without losing the overarching aim. And consider initiation itself, the starting of goal-directed behavior: it is the installation of a goal into command in the first place, the kindling of the maintained representation that the rest of the system will then defend.

Seen this way, the dorsolateral cortex does not house a committee of executives. It does one thing — keep a represented goal in charge of behavior against the pull of the present — and “working memory,” “inhibition,” “shifting,” and “planning” are that one thing described under different task demands. This is why the functions refuse to separate: they were never separate. And it is why the word “executive,” though a useful clinical label, is a trap if taken literally. There is no executive doing these things. There is a maintained representation, and a set of consequences that follow from maintaining it.

The several faces of capture

If the dorsolateral cortex’s job is to keep a chosen goal in command against the pull of the present, then its failure has a characteristic and unmistakable form: capture. When the line cannot be held, behavior is seized by whatever is most salient, most habitual, or most immediately available — the goal loses its grip, and the present takes over. The clinical literature on frontal damage is, read this way, a catalogue of captures, each revealing the goal losing its hold to a different captor.

Capture by the present object

The most startling form is the one the French neurologist Lhermitte described as utilization behavior, or environmental dependency. His patients, with frontal damage, could not keep their hands off the objects around them. Presented with a hammer and a nail and a picture leaning against the wall, a patient would hang the picture; shown a bed, he would undress and get into it; handed a syringe, she would inject the examiner. In a room arranged as a cocktail party, one patient simply began serving the others. The objects, Lhermitte said, compelled their use: “the presentation of objects implies the order to grasp and use them.” The patient had become dependent on the outside world, captured by the affordances in front of him.

We have seen this logic before, one level down. In the last unit, the sandwich afforded both eating and throwing, and the basal ganglia resolved the competition between those candidate actions. But what determined that the eating affordance, rather than the throwing one, governed behavior? Part of the answer, we can now say, is the frontal lobe: it holds behavior free of the merely-present affordance, so that what you do is governed by your goal rather than by whatever the environment happens to offer. Lhermitte’s patients show what is left when that freedom is removed. The affordance wins by default, because nothing is holding a goal in command against it. Utilization behavior is the dorsolateral function failing at its most basic: the present object, which a held goal would normally override, simply takes the wheel.

Capture by the prepotent rule

A second form is the perseveration revealed by the Wisconsin Card Sort. Cards can be matched by color, by shape, or by number; the patient sorts, the examiner says right or wrong, and after the patient has settled into the correct rule, the examiner silently changes it. Now the old rule earns only error. The healthy sorter, after a few wrong answers, abandons the old rule and finds the new one. The patient with frontal damage cannot let go: they keep sorting by the rule that used to work, perseverating against a steady stream of negative feedback, captured by the rule that is no longer true.

This is the same failure as utilization behavior, displaced from objects to rules. The patient’s behavior is governed by something present and prepotent — here, the established sorting rule — rather than by the updated goal the feedback demands. Holding a goal in command means, when the world changes, releasing the old goal and installing the new one; perseveration is the failure of that release. The old rule has captured behavior and will not give it back. (The healthy performance of this task, fittingly, draws on the dorsolateral cortex: the requirement to stop doing what you have been doing and do something else is one of its signature demands.)

Capture by the salient distractor

A third form is subtler and quantifiable. Hold three faces in working memory across a delay, and then, during the delay, flash a distracting image — a neutral picture, or an emotionally charged one. The dorsolateral activity that was sustaining the remembered faces drops when the distractor appears, and drops most when the distractor is emotional; task performance falls with it, again worst for emotional distractors. Meanwhile the ventral lateral cortex and the amygdala light up to exactly those emotional intrusions. The maintained goal is being knocked offline by a salient event, and one can watch the trade-off happen: as the emotional distractor captures the ventral, emotion-sensitive machinery, it pulls the dorsolateral cortex off the goal it was holding.

This is capture caught in the act, and it draws the two systems of this unit into direct contact. The emotional distractor is, in the terms of the last section, a sudden bid for value — a salient, affectively loaded event — and it competes with the held goal for control of behavior. The dorsolateral cortex’s task is to defend the goal against that bid; when it cannot, the salient present wins. (Whether the ventral lateral activity here is itself the “circuit-breaker” that interrupts the goal, or rather a protection against the interruption, is not settled — but the phenomenon is clear: a vivid present can seize behavior from a held intention, and holding the line means resisting exactly that seizure.)

When holding the line is the wrong thing to do

It would be easy, after this catalogue of failures, to conclude that more dorsolateral control is simply better — that holding the line is always the right thing and capture always the wrong one. This is false, and the place it goes wrong is instructive, because it shows that the frontal contribution is not a wiser agent but a bias, and biases can mislead.

Consider a problem that requires an unusual solution — the kind where the obvious approach is a trap and the answer lies “outside the box.” On certain such problems, patients with dorsolateral damage actually outperform healthy controls. The interpretation, due to Frith and others, is that the dorsolateral cortex sculpts the response space: it constrains the set of actions and interpretations you consider, biasing you toward the solutions that usually work. Most of the time this is exactly what you want — it keeps behavior on the well-worn path toward the goal. But when the usual path is a dead end and the answer is atypical, that very constraint becomes a handicap, and the patient who lacks it, unconstrained, stumbles onto the unusual solution the healthy brain has been holding itself away from.

This is the dorsolateral analogue of a result we met in the overview, where patients with ventral damage, missing the anticipatory sting of loss, made bets that happened to be advantageous. In both cases the frontal contribution is revealed as a prior — a bias toward what is usually right — rather than a faculty of correct judgment. The held goal, the sculpted response space, the maintained rule: these keep behavior pointed at what has worked before. That is overwhelmingly useful, which is why we have the machinery. But it is a bias, not an oracle, and when the situation is unusual the bias is wrong, and the person who lacks it does better. The frontal lobe is not a chief who knows the right answer. It is a mechanism that holds behavior to the ordinarily-right answer — and the difference shows precisely when the ordinary answer fails.

A note on frontal memory failures

Before leaving the dorsolateral cortex, a brief word on a class of deficits that sit alongside working memory and are easily confused with it. Patients with frontal damage often show problems not with remembering facts but with remembering the circumstances of facts: which of two things they saw more recently (recency memory), or where and when they learned something (source memory). They may know a fact perfectly while having lost track of when they encountered it, or falsely believe they had always known something they were just told.

These fit the section’s theme more than they first appear: keeping track of the temporal and contextual frame of one’s own experience is itself a kind of holding — a maintenance of the order and origin of things, against which a particular memory can be placed. But honesty requires flagging that the localization here is genuinely disputed. While source-memory deficits are classically attributed to frontal damage, careful work (Thaiss and Petrides among others) has found such deficits with temporal-lobe damage instead, suggesting that earlier studies may have implicated the frontal lobe only because their lesions were large enough to take temporal tissue too. We raise these deficits to acknowledge them and to note their fit with the holding theme, while marking the attribution as one of the less settled claims in this area — a reminder that the lesion method’s hazards, which we flagged at the unit’s start, are not hypothetical.

Where this leaves us, and what it sets up

The dorsolateral cortex, then, supplies what the valuing cortex of the last section could not: not a judgment of what an option is worth, but the holding of a chosen goal in command of behavior across the time and distraction that separate intention from arrival. Its mechanism is persistent activity — representations kept lit against the absence of the thing they represent, visible in the delay cells with their memory fields. Its several apparent functions — working memory, inhibition, shifting, planning, initiation — are one function under different demands: keep the goal in charge. And its failures are all captures, the goal losing its grip to the present object, the prepotent rule, or the salient distractor. Throughout, there is no executive — only a maintained representation and the consequences of maintaining it, which is exactly what “executive control” turns out to mean.

We now have two of the unit’s systems in hand, and the contrast between them is sharp. The ventral cortex values possible futures; the dorsolateral cortex holds the chosen one in command. But two large problems remain, and the next sections take them up in turn.

The first is the one the valuation section deferred and this section has not touched. We have shown how worth is represented and how a goal is held, but not how a merely possible future — one that has not happened, that the animal must imagine — comes to carry worth at all, in a way that can move behavior before the future is real. The held goal in this section was already chosen; the valued option in the last was already present or cued. Neither section explained how an imagined consequence gets to be felt. That is the next problem, and its answer runs back to the ventral cortex’s oldest inheritance — its origins in the body and the emotions — and reconnects this unit to the interoceptive machinery of the very first one.

The second problem is one this section has, in passing, sharpened. We have praised the dorsolateral cortex for holding a goal against capture by the salient and the novel — for keeping behavior on the well-worn path. But an animal that only held the line, that was never pulled off its current goal by something new, would never discover anything. It would pursue what it already valued and learn nothing of what else the world affords. Holding the line, taken alone, is a recipe for a competent animal that never explores. There must be a countervailing pull — a drive to be drawn off the goal by the novel, to sample the unknown precisely when no urgent goal is in command. That drive stands in direct tension with everything this section has praised, and it is the subject of a later one. For now, note the tension and hold it: the cortex that keeps you pointed at your goal must somehow coexist with a drive that pulls you away from it, and an animal that lives well needs both.

What we are sure of, and what is still open

As before, it is worth separating the settled core from the frontier.

What is well established. The dorsolateral prefrontal cortex is essential for working memory — the temporary holding-online of information to guide behavior in the absence of the eliciting stimulus. Neurons there show sustained, spatially specific delay-period activity (“memory fields”), demonstrated by Goldman-Rakic in the oculomotor delayed-response task, and damage to this cortex impairs delayed-response performance. In humans, dorsolateral activity accompanies the maintenance of information across a delay, scales with load up to a sharp limit, and is associated with sustained activity in posterior regions specialized for the held content. Frontal damage produces a recognizable family of “executive” deficits — perseveration and impaired set-shifting (the Wisconsin Card Sort), impaired planning (the Tower of London), and distractibility — and a vivid release phenomenon, utilization behavior (Lhermitte), in which behavior is captured by the affordances of present objects. Emotional distractors reduce dorsolateral maintenance activity and impair working-memory performance while engaging ventral lateral cortex and the amygdala.

What remains contested or unsettled. The mechanism of working-memory maintenance is not fully settled: whether information is held by literally persistent firing, or by transient and “activity-silent” mechanisms such as short-term synaptic changes that can be read out as needed, is an active debate, and the simple picture of a steadily firing cell is likely too clean. Whether the dorsolateral cortex stores content or instead controls and sustains representations housed elsewhere — the “attention to internal representations” view — bears on what working memory fundamentally is, and the storehouse picture is increasingly questioned. Whether there exists a dedicated mechanism for response inhibition (a specific stopping circuit, perhaps in the ventrolateral cortex) or whether most “inhibition” is simply the obverse of maintenance and selection is not resolved; the status of the ventral lateral “circuit-breaker” in particular is uncertain. And the localization of contextual-memory deficits — recency and especially source memory — to the frontal lobe is genuinely disputed, with evidence implicating the temporal lobe and the suspicion that large lesions confounded earlier work. The dorsolateral cortex’s role in holding goals online is among the best-established facts in this unit; the precise mechanism of the holding, and the proper inventory of “executive” sub-functions, remain open.