Attention

Abstract

Attention is the set of processes by which the mind allocates its limited resources to some information at the expense of the rest. It is not a single faculty but an umbrella over several partly separable functions: selecting among competing inputs, dividing effort across concurrent tasks, sustaining focus over time, and orienting to locations and objects. A century of research has moved from William James's introspective definition through the information-processing bottleneck models of the mid-twentieth century to a contemporary account in which attention is a biased competition among neural representations, governed by distinct alerting, orienting, and executive-control networks. This article surveys the field as a whole and links down to the detailed treatments of its parts. Three interactive demonstrations model the dual-task bottleneck, the vigilance decrement, and the search-slope signature that separates parallel from serial visual search.

Keywords: attention, selective attention, divided attention, vigilance, attention networks

Attention is the process that decides, moment to moment, which fraction of the available information receives the brain's limited processing capacity. William James gave the field its enduring definition more than a century ago: attention is the taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneous objects or trains of thought (James, 1890). The definition still frames the problem, because it identifies both the selectivity that makes attention useful and the capacity limit that makes it necessary. The sensory surfaces register far more than the nervous system can identify, remember, and act upon, so some principled allocation is unavoidable, and the study of attention is the study of how that allocation is made, what its limits are, and how it is implemented in the brain. What follows treats attention as the umbrella construct it is, sketching each of its major varieties and the theories built to explain them, and pointing to the fuller accounts of the subtopics where they exist.

What Attention Is

Attention names a functional relationship, not an organ or a place: it is the prioritizing of some information for fuller processing than the rest receives. The guiding assumption, formalized at the start of the information-processing era, is that the nervous system is a channel of limited capacity, so that increasing the resources devoted to one stream necessarily reduces what remains for others (Broadbent, 1958). A complementary framing treats attention not as a fixed structural gate but as a pool of effort or *capacity* that can be graded and divided, its total size enlarged by arousal and its allocation governed by an executive that weighs the momentary demands of each task (Kahneman, 1973). The two framings — a bottleneck to be positioned and a resource to be shared — recur throughout the field and are less rivals than descriptions of different limits.

Because attention is an umbrella term, any survey needs a taxonomy. One influential scheme divides attention by its target: *external* attention selects among perceptual inputs and their locations, features, and objects, whereas *internal* attention selects among the contents of memory, task sets, and lines of thought (Chun et al., 2011). A cross-cutting scheme divides attention by its function, distinguishing the selection of one input from among many, the division of effort across several tasks, the sustaining of focus across time, and the orienting of processing toward a location or object. Figure 1 lays out this functional taxonomy, which organizes the sections that follow.

Varieties of Attention

The functional divisions are not merely descriptive; they dissociate. Selective and divided attention trade against each other, since resources spent guarding a single channel are unavailable for a second task, and both differ from the tonic *alertness* that sustained attention maintains over minutes or hours. Orienting, the movement of processing through space or across objects, can proceed covertly, without any movement of the eyes, and follows a partly different logic again. That these functions can be impaired independently, load onto separable brain networks, and show different developmental and pharmacological profiles is the empirical warrant for treating attention as a family rather than a single mechanism (Petersen & Posner, 2012). The sections below take each in turn, beginning with the variety that has generated the most theory: selective attention.

Selective Attention and the Bottleneck

Selective attention is the choosing of one source of information over competitors, and its experimental study began with the cocktail party problem that Colin Cherry named: how a listener follows one conversation while many compete in the same room (Cherry, 1953). Using dichotic listening, in which a different message is played to each ear and the listener repeats one aloud, Cherry found that the unattended message was processed remarkably little beyond its physical properties. Donald Broadbent gathered such results into filter theory, the first information-processing model of attention, in which an early filter admits one physically defined channel into a limited-capacity system and rejects the rest before meaning is analyzed (Broadbent, 1958). The strict early filter did not survive intact: Anne Treisman showed that the unattended channel is *attenuated* rather than blocked, so that salient or primed words still break through (Treisman, 1960), while John and Diana Deutsch argued that selection occurs late, after all input is analyzed for meaning, at the point of response (Deutsch & Deutsch, 1963).

The early-versus-late debate ran for two decades because both positions could accommodate the evidence by moving the bottleneck. Nilli Lavie reframed it: selection is neither fixed early nor fixed late but occurs at a locus set by the perceptual load of the attended task (Lavie, 1995). Under high load, capacity is exhausted on the relevant task and distractors go unprocessed, so selection is effectively early; under low load, spare capacity spills automatically onto distractors, so selection is effectively late. The account subsumes the older theories as limiting cases and yields the paradoxical, well-supported prediction that a more demanding task can reduce distraction (Lavie, 2005). The detailed history, the leakage findings, and the neural evidence are treated in the companion article on selective attention; here it is one variety among several.

Divided Attention and the Central Bottleneck

Where selective attention concerns ignoring, divided attention concerns doing two things at once, and the ability is sharply bounded. The clearest evidence comes from the *psychological refractory period*: when two tasks requiring separate responses follow each other closely, the response to the second is delayed, and the delay grows as the interval between the tasks shrinks (Pashler, 1994). The pattern implicates a central bottleneck at the stage of response selection: perceptual analysis of the two stimuli can overlap in parallel, but the decision about how to respond to the second cannot begin until the first is resolved, so the second response waits. The bottleneck is remarkably general, appearing even when each task alone is trivial, which suggests it is a structural limit on the act of choosing a response rather than a shortage of perceptual resources.

A capacity account complements the structural one. Daniel Kahneman treated attention as a single pool of effort supplied by arousal and allocated by an executive, so that two tasks can be combined only if their summed demand does not exceed the pool, and performance on either degrades gracefully as the other's demand rises (Kahneman, 1973). Dual-task performance thus reflects both a hard limit on concurrent response selection and a softer competition for shared capacity, and which one dominates depends on the tasks; the companion article on divided attention develops the bottleneck and capacity accounts, multiple-resource theory, and task switching in full. The demonstration below models the refractory-period effect, letting the interval between two tasks vary and reading off the delay it imposes on the second response.

Automaticity: When Attention Is Not Needed

Not all processing demands attention. Walter Schneider and Richard Shiffrin distinguished *controlled* processing, which is slow, serial, effortful, and capacity-limited, from *automatic* processing, which is fast, parallel, and effortless once established (Schneider & Shiffrin, 1977). Their visual-search experiments showed that consistent practice, in which a stimulus is always a target and never a distractor, gradually converts a slow controlled search into an automatic one whose speed no longer depends on the number of items; inconsistent mapping, where the same item is sometimes target and sometimes distractor, keeps search controlled indefinitely. Automaticity is therefore learned, and it frees attention for other work, but at the cost of control: an automatic process runs whether or not it is wanted.

The Stroop task is the enduring demonstration of that cost. Naming the ink color of a word is markedly slowed when the word spells a conflicting color, because reading has become so automatic that the word's meaning is extracted involuntarily and competes with the color response (Stroop, 1935). Half a century of research established the effect as the benchmark measure of automaticity and of the cognitive control that must override it, robust across hundreds of variations (MacLeod, 1991). Automaticity thus marks the boundary of attention's domain: it is the set of processes that have escaped the need for it.

Sustained Attention and Vigilance

Attention must not only be directed but maintained, and maintaining it over time is itself effortful and imperfect. Norman Mackworth, working on why radar operators missed targets late in a watch, built the *Clock Test*, in which observers monitored a pointer for occasional double jumps over a long session, and documented the *vigilance decrement*: detection accuracy falls measurably within the first half hour and continues to decline thereafter (Mackworth, 1948). The decrement is one of the most reliable findings in applied attention research and matters wherever a person must watch for rare, unpredictable signals — screening, inspection, driving, and monitoring automated systems. It reflects the difficulty of holding a high level of tonic alertness against a task that provides little to engage it, and it is worsened by low signal rate, low signal salience, and time on task; the companion article on sustained attention treats the decrement, the sensitivity-criterion distinction, mind-wandering, and the neural basis in full. The demonstration below models the decrement, letting watch length and signal salience vary and showing their effect on the detection rate over time.

Orienting the Spotlight

Attention can be aimed through space independently of the eyes, a covert orienting that Michael Posner measured with the *spatial cueing* paradigm (Posner, 1980). A cue marks a likely location; detection is faster when the target appears there and slower when it appears elsewhere, relative to a neutral baseline, yielding a benefit of valid cueing and a cost of invalid cueing. The pattern shows that attention behaves like a spotlight or zoom lens that can be moved ahead of any eye movement, enhancing processing at its focus. Posner separated *exogenous* orienting, captured automatically by a peripheral event, from *endogenous* orienting, directed voluntarily by a symbolic cue, a behavioral division that maps onto partly separable neural systems. Orienting also has measurable perceptual consequences beyond speed: directing covert attention to a location increases apparent contrast and spatial resolution, so that attention changes what is seen and not only how fast it is reported (Carrasco, 2011).

Visual Search and Feature Integration

Everyday orienting usually means searching a cluttered scene for a target, and the time it takes reveals how attention is deployed. Anne Treisman's feature-integration theory holds that simple features such as color and orientation are registered in parallel across the whole field, without attention, but that combining features into an object requires focal attention applied to one location at a time (Treisman & Gelade, 1980). The prediction is a signature in the data: when a target is defined by a single feature it *pops out*, and search time barely rises as more distractors are added, but when a target is defined by a *conjunction* of features, search time climbs steeply with set size because attention must inspect items roughly one by one. Without attention to bind them, features can miscombine into *illusory conjunctions*, the report of a red X where a red O and a green X were shown. The slope of search time against set size thus indexes whether search is parallel or serial, and the demonstration below computes it for feature and conjunction search.

Note. The four varieties are partly separable functions unified by a shared capacity limit and, at the neural level, by overlapping control networks (Petersen & Posner, 2012).

Attention in the Brain

Michael Posner and Steven Petersen proposed that attention is not one system but three, anatomically and functionally distinct: an *alerting* network that establishes and maintains tonic readiness, an *orienting* network that selects locations and shifts the spotlight, and an *executive* network that detects and resolves conflict among responses (Posner & Petersen, 1990). Two decades of imaging refined rather than overturned the scheme, tying each network to identifiable regions and neuromodulators — noradrenaline for alerting, acetylcholine for orienting, dopamine for executive control (Petersen & Posner, 2012). Maurizio Corbetta and Gordon Shulman drew a complementary distinction within orienting, between a dorsal frontoparietal network that directs attention according to goals and a ventral, right-lateralized network that detects salient events and interrupts the dorsal system to reorient toward them (Corbetta & Shulman, 2002).

Underneath these control systems, the mechanism attention exerts on sensory cortex is a *biased competition*. Robert Desimone and John Duncan proposed that multiple stimuli compete for representation in the visual system and that attention resolves the competition in favor of the behaviorally relevant object, enhancing the neurons that code it and suppressing those that code its rivals (Desimone & Duncan, 1995). This reframes selection as the outcome of a graded contest rather than the flipping of a discrete filter, and because the bias can be applied wherever the competition is strongest, it accommodates both the early and the late behavioral effects that the bottleneck models fought over. The control networks supply the bias; the competition in sensory cortex resolves it. The fuller treatment of biased competition and the dorsal–ventral distinction lives in the selective attention article.

Attention and Awareness

If attention governs what is fully processed, it should also govern what reaches awareness, and the most striking evidence is the failure to notice the unattended. In *inattentional blindness*, observers absorbed in a demanding task fail to see a salient but unexpected event in plain view; in the best-known study, about half of viewers counting basketball passes never notice a person in a gorilla suit walking through the display (Simons & Chabris, 1999). The result shows that attention is close to necessary for the conscious perception of even conspicuous objects, and that the richness observers believe they see is partly an illusion. The dependence is not total — some rapid, overlearned categorizations survive near-total withdrawal of attention — but for detailed, reportable perception, attention and awareness travel closely together. The phenomenon also marks the practical stakes of the whole field: a driver attending to a phone conversation can look directly at a hazard and fail to see it.

Open Questions

The largest open question is whether *attention* names a natural kind at all or merely a loose collection of control operations that happen to share a word. The functional varieties dissociate, load onto different networks, and obey different limits, which has led some to argue that the unqualified term does more harm than good and should be replaced by names for the specific mechanisms. Against that, the varieties interact so pervasively — orienting serves selection, selection presupposes a capacity that division exhausts, sustained alertness gates them all — that treating them wholly separately loses the coordination that made a single construct attractive in the first place. A second open question concerns the currency of the limit: whether the fundamental constraint is a structural bottleneck at response selection, a graded pool of shared capacity, or, at the neural level, the dynamics of competition among representations, and whether these are three descriptions of one limit or genuinely distinct ones. A third concerns internal attention, the selection among memories and thoughts, which is harder to manipulate and measure than its external counterpart and remains comparatively underspecified (Chun et al., 2011). What is not in doubt is the centrality of the problem: perception, memory, and action are all organized around the same scarce resource, and no account of any of them is complete without it.

Worked Example

Consider a visual-search experiment that measures reaction time as the number of items on the display, the *set size*, is varied. Search time is well described by a straight line, RT equals an intercept plus a slope times the set size, where the intercept absorbs the fixed costs of perceiving the display and making a response, and the slope estimates the time added per item. In a feature-search condition, where the target is defined by a single feature and pops out, suppose mean reaction time is 408 milliseconds at set size 4 and 432 milliseconds at set size 16. The slope is the change in time divided by the change in set size: 432 minus 408 is 24 milliseconds, divided by 16 minus 4, which is 12 items, giving 2 milliseconds per item. A slope near zero is the signature of parallel search: adding distractors costs almost nothing because all items are inspected at once.

In a conjunction-search condition, where the target is defined by a combination of two features, suppose mean reaction time is 500 milliseconds at set size 4 and 800 milliseconds at set size 16. The slope is 800 minus 500, which is 300 milliseconds, divided by the same 12 items, giving 25 milliseconds per item — more than ten times the feature-search slope. This steep slope is the signature of serial search: attention inspects the items roughly one at a time, so each added item costs a fixed increment. A further diagnostic is the ratio of the target-absent slope to the target-present slope, which for serial self-terminating search is about two to one, because confirming that no target is present requires inspecting every item whereas finding one requires inspecting on average half of them. The VisualSearchDemo above computes reaction time from the fixed intercept, the per-item slope of the search type the reader chooses, and a set size the reader sets, and reproduces both the slope arithmetic and the two-to-one absent-to-present ratio carried out here.

Discussion

The century of attention research traces a consistent movement from structure toward dynamics. James described attention introspectively; Broadbent and his successors modeled it as a fixed place in an information-processing pipeline; load theory turned that fixed place into a variable set by task demands; and the neuroscience of attention replaced the metaphor of a gate with that of a competition biased by goals and salience (Broadbent, 1958; Lavie, 1995; Desimone & Duncan, 1995). At each step the earlier account survived as a special case rather than being discarded, which is why the field can still speak of filters and bottlenecks while believing that neither is a literal structure. The parallel movement in the study of control, from a single attentional resource to a set of interacting alerting, orienting, and executive networks, tells the same story: what looked like one thing resolved, on closer inspection, into a coordinated several (Posner & Petersen, 1990; Petersen & Posner, 2012).

That resolution is the source of both the field's power and its central tension. Decomposing attention into separable functions has made it tractable, letting each variety be measured with its own paradigm and tied to its own circuitry. But the functions were grouped under one word because they share a capacity limit and constantly trade against one another, and a decomposition thorough enough to lose that shared limit would explain the parts at the cost of the whole (Kahneman, 1973). The enduring demonstrations — the audibility of one's own name across a crowded room, the invisibility of an unexpected gorilla, the steep climb of search time when features must be conjoined — are vivid precisely because they expose the single scarce resource that all the varieties draw upon (Cherry, 1953; Simons & Chabris, 1999; Treisman & Gelade, 1980). Attention endures as the organizing problem of cognitive psychology because it sits at the junction of perception, memory, and action, and rations the capacity on which all three depend.

Glossary

Alerting network.

The attention system that establishes and maintains a tonic state of readiness to respond, associated with noradrenaline.

Attenuation.

Treisman's proposal that unattended input is turned down rather than blocked, so it is recognized only when the relevant threshold is low.

Automatic processing.

Fast, parallel, effortless processing that runs without attention once established by consistent practice, at the cost of voluntary control.

Biased competition.

The account in which stimuli compete for cortical representation and attention resolves the contest in favor of the behaviorally relevant item.

Bottleneck.

A stage of limited capacity at which parallel processing gives way to the serial handling of a single item or response.

Capacity.

The limited pool of processing effort, enlarged by arousal, that an executive allocates across concurrent tasks.

Conjunction search.

Search for a target defined by a combination of features, which requires focal attention and yields search time that rises steeply with set size.

Covert orienting.

The shifting of attention to a location without moving the eyes, measurable through the costs and benefits of spatial cueing.

Divided attention.

The attempt to process two or more sources or tasks at once, bounded by a central bottleneck and by shared capacity.

Executive network.

The attention system that detects and resolves conflict among competing responses, associated with dopamine and the anterior cingulate.

Feature-integration theory.

Treisman's account in which features are registered in parallel but focal attention is required to bind them into an object.

Filter theory.

Broadbent's model in which an early filter admits one physically defined channel into a limited-capacity system and rejects the rest.

Illusory conjunction.

A miscombination of features from different objects that occurs when focal attention is unavailable to bind them correctly.

Inattentional blindness.

The failure to notice a salient but unexpected object when attention is engaged elsewhere.

Internal attention.

The selection among the contents of memory, task sets, and thought, as distinct from the selection among perceptual inputs.

Perceptual load.

The processing demand of the attended task, which in load theory determines whether spare capacity spills onto distractors.

Psychological refractory period.

The delay in responding to the second of two closely spaced tasks, evidence of a central bottleneck at response selection.

Search slope.

The increase in search time per added display item, near zero for parallel feature search and steep for serial conjunction search.

Selective attention.

The prioritizing of one source of information over competitors, the most studied variety of attention.

Sustained attention.

The maintenance of focus on a task over an extended period, subject to the vigilance decrement.

Vigilance decrement.

The decline in the detection of rare signals as time on a monitoring task increases.

Key Researchers

William James (1842-1910). Psychologist and philosopher at Harvard University; gave the field its enduring definition of attention in The Principles of Psychology (1890). Wikipedia

Donald E. Broadbent (1926-1993). Psychologist at the MRC Applied Psychology Unit in Cambridge and later Oxford; proposed filter theory, the first information-processing model of attention. Wikipedia

E. Colin Cherry (1914-1979). Scientist at Imperial College London; defined the cocktail party problem and pioneered the dichotic-listening method for studying auditory attention. Wikipedia

Neville Moray (1935-2017). British engineering psychologist at the University of Toronto and later Surrey; demonstrated the own-name breakthrough effect that constrains theories of filtering. Wikipedia

Anne M. Treisman (1935-2018). Cognitive psychologist at Princeton University; developed the attenuation model of selection and feature-integration theory of how attention binds features. Wikipedia

Diana Deutsch. Professor Emeritus of Psychology at the University of California, San Diego; co-authored the late-selection theory and is a leading researcher on the psychology of music. Faculty Page - Google Scholar - Wikipedia

Richard M. Shiffrin. Distinguished Professor of Psychological and Brain Sciences at Indiana University Bloomington; with Walter Schneider, distinguished controlled from automatic processing. Faculty Page - Google Scholar - ORCID - Wikipedia

Michael I. Posner. Professor Emeritus of Psychology at the University of Oregon; devised the spatial cueing paradigm and, with Steven Petersen, mapped attention into alerting, orienting, and executive networks. Faculty Page - Google Scholar - ORCID - Wikipedia)

Steven E. Petersen. James S. McDonnell Professor of Cognitive Neuroscience at Washington University in St. Louis; with Michael Posner, defined the human attention system in neuroimaging terms. Faculty Page - Google Scholar - ORCID

Nilli Lavie. Professor at the UCL Institute of Cognitive Neuroscience; formulated load theory, which makes the locus of selection depend on the perceptual load of the current task. Faculty Page - Google Scholar - ORCID - Wikipedia

Harold Pashler. Distinguished Professor of Psychology at the University of California, San Diego; characterized the central bottleneck of dual-task performance and the psychological refractory period. Faculty Page - Google Scholar - ORCID - Wikipedia

Marvin M. Chun. Richard M. Colgate Professor of Psychology at Yale University; proposed the taxonomy dividing attention into external and internal varieties. Faculty Page - Google Scholar - ORCID

Robert Desimone. Director of the McGovern Institute for Brain Research at MIT; co-proposed the biased-competition account of selective attention in visual cortex. Faculty Page - Google Scholar - ORCID - Wikipedia

Maurizio Corbetta. Professor of Neuroscience at the University of Padua; distinguished the dorsal goal-directed and ventral stimulus-driven attention networks in the human brain. Faculty Page - Google Scholar - ORCID

Marisa Carrasco. Professor of Psychology and Neural Science at New York University; showed that covert attention alters the appearance of stimuli, not merely the speed of responding. Faculty Page - Google Scholar - ORCID - Wikipedia

Daniel J. Simons. Professor of Psychology at the University of Illinois Urbana-Champaign; demonstrated inattentional and change blindness, including the invisible-gorilla study. Faculty Page - Google Scholar - ORCID - Wikipedia

Frequently Asked Questions

What is attention in psychology?
Attention is the set of processes that allocate the mind's limited capacity to some information at the expense of the rest, made necessary because sensory input far exceeds what can be fully processed (James, 1890).

Is attention a single thing?
No. It is an umbrella for partly separable functions such as selective, divided, sustained, and orienting attention, which dissociate behaviorally and load onto different brain networks (Petersen & Posner, 2012).

What is the difference between selective and divided attention?
Selective attention chooses one input and ignores others, whereas divided attention tries to handle several tasks at once and is bounded by a central bottleneck at response selection (Pashler, 1994).

Why does performance drop during a long monitoring task?
Because sustained attention is subject to the vigilance decrement: the detection of rare signals declines within the first half hour and continues to fall with time on task (Mackworth, 1948).

What is automatic attention?
Automatic processing is fast, parallel, and effortless and runs without attention once consistent practice establishes it, though it can no longer be easily suppressed (Schneider & Shiffrin, 1977).

How is attention directed without moving the eyes?
Through covert orienting, measured by spatial cueing: a valid cue speeds detection at its location and an invalid cue slows it, showing attention shifts like a spotlight ahead of eye movements (Posner, 1980).

How does the brain control attention?
Through separable alerting, orienting, and executive networks that bias a competition among stimuli for representation in sensory cortex (Posner & Petersen, 1990).

Can people miss obvious things when attending elsewhere?
Yes. In inattentional blindness, observers absorbed in a task fail to notice a salient unexpected event, such as a person in a gorilla suit walking through a scene they are watching (Simons & Chabris, 1999).

References