Proc. Nati. Acad. Sci. USAVol. 91, pp. 7398-7403, August 1994

Review

Attention: The mechanisms of consciousness(psychology/awareness/seection/vual search)

Michael I. PosnerInstite of Cognitive and Decision Scencess, University of Oregon, Eugene, OR 97403

ABSTRACT A number of recent pa-pers and books discuss theoretical effortstoward a scientific understanding of con-sciousness. Progress in imaging networksof brain areas active when people performsimple tasks may provide a useful empir-ical background for distinguishing con-scious and unconscious information pro-cessing. Attentional networks includethose involved in orienting to sensory stim-uli, activating ideas from memory, andmaintaining the alert state. This paperreviews recent findings in relation to clas-sical issues in the study of attention andanatomical and physical theories of thenature of consciousness.

What is it to be conscious? This hasbecome a central question in many seri-ous scientific circles (1-6). Proposalsrange from the anatomical, for example,locating consciousness in the thalamus(6) or in thalamic-cortical interactions (2,6), to the physical, for example, the pro-posal that consciousness must rest onquantum principles (1, 4). Some propos-als combine physical and anatomical rea-soning. For example, Beck and Eccles (1)argue that conscious processing acts toincrease the probability of quantal dis-charge at the synapse.

In this paper I propose to discuss theissue of consciousness in light of recentfindings about attentional networks ofthe human brain that lead to selection ofsensory information, activate ideasstored in memory, and maintain the alertstate. I don't believe that any of thesemechanisms are "consciousness" itself,just asDNA is not "life," but I do believethat an understanding of consciousnessmust rest on an appreciation of the brainnetworks that subserve attention, inmuch the same way as a scientific anal-ysis of life without consideration of thestructure of DNA would seem vacuous.

Attention

The study of attention has a long historywithin psychology. William James (7)wrote at the turn of the century, "Every-one knows what attention is. It is thetaking possession by the mind in clearand vivid form of one out of what seem

several simultaneous objects or trains ofthought."The dominance of behavioral psychol-

ogy postponed research into the internalmechanisms of selective attention in thefirst half of this century. The finding thatthe integrity of the brain stem reticularformation was a necessity to maintain thealert state provided some anatomical re-ality to the study ofan aspect of attention(8). The quest for information-processingmechanisms to support the more selec-tive aspect of attention began after WorldWar II with studies of selective listening.A filter was proposed that was limited forinformation (in the formal sense of infor-mation theory) and located betweenhighly parallel sensory systems and alimited-capacity perceptual system (9).

Selective listening experiments sup-ported a view of attention that suggestedearly selection of relevant message, withnonselective information being lost toconscious processing. However, onsome occasions it was clear that unat-tended information was processed to ahigh level because there was evidencethat an important message on the unat-tended channel might interfere with theselected channel.

In the 1970s psychologists began todistinguish between automatic and con-trolled processes. It was found thatwords could activate other words similarin meaning (their semantic associates),even when the person had no awarenessof the words' presence. These studiesindicated that the parallel organizationfound for sensory information extendedto semantic processing. Thus, selecting aword meaning for active attention ap-peared to suppress the availability ofother word meanings. Attention wasviewed less as an early sensory bottle-neck and more as a system for providingpriority for motor acts, consciousness,and memory (10).Another approach to problems of se-

lectivity arose in work on the orientingreflex (11). The use of slow autonomicsystems (e.g., skin conductance as mea-sures of orienting) made it difficult toanalyze the cognitive components andneural systems underlying orienting.During the last 15 years there has been asteady advancement in our understand-ing ofthe neural systems related to visual

orienting from studies using single-cellrecording in alert monkeys (12). Thiswork showed a relatively restricted num-ber of areas in which the firing rates ofneurons were enhanced selectively whenmonkeys were trained to attend to alocation, at the level of the superior col-liculus (i.e., the midbrain), selective en-hancement could only be obtained wheneye movement was involved, but in theposterior parietal lobe of the cerebralcortex, selective enhancement occurredeven when the animal maintained fixa-tion. An area of the thalamus, the lateralpulvinar, was similar to the parietal lobein containing cells with the property ofselective enhancement.

Until recently, there has been a sepa-ration between human information pro-cessing and neuroscience approaches toattention using nonhuman animals. Theformer tended to describe attention, ei-ther in terms of a bottleneck that pre-vented limited-capacity central systemsfrom overload or as a resource that couldbe allocated to various processing sys-tems in a way analogous to the use of theterm in economics. On the other hand,neuroscience views emphasized severalseparate neural mechanisms that mightbe involved in orienting and maintainingalertness. Currently, there is an attemptto integrate these two within a cognitiveneuroscience of attention. An impressiveaspect of current developments in thisfield is the convergence ofevidence fromvarious methods of study. These includeperformance studies using reaction time,dual-task performance studies, recordingfrom scalp electrodes, and lesions in hu-mans and animals, as well as variousmethods for imaging and recording fromrestricted brain areas, including individ-ual cells (13).

Current progress in the anatomy of theattention system rests most heavily ontwo important methodological develop-ments. (i) The use of microelectrodeswith alert animals allowed evidence forthe increased activity of cell populationswith attention (12). (ii) Anatomical (e.g.,computerized tomography or magneticresonance imaging) and physiological[e.g., positron emission tomography

Abbreviation: PET, positron emission tomog-raphy.

7398

Proc. Natl. Acad. Sci. USA 91 (1994) 7399

(PET) and functional magnetic resonancimagery] methods of studying parts of tibrain allowed more meaningful investgations of localization of cognitive funtion in normal people (13). The futumshould see the use of localizing methodtogether with methods of tracing the timcourse of brain activity in the humasubject. This combination should provida convenient way to trace the rapid timedynamic changes that occur in the coursof human information processing.Three fundamental working hypothe

ses characterize the current state of elforts to develop a combined cognitiveneuroscience of attention. (i) There exists an attentional system ofthe brain thais anatomically separate from varioudata-processing systems that can be activated passively by visual and auditorinput. (ii) Attention is carried out by;network of anatomical areas. It is neitherthe property of a single brain area nor isit a collective function of the brain working as a whole. (iii) The brain areainvolved in attention do not carry out thesame function, but specific computationsare assigned to different areas (13).

It is not possible to specify the com-plete attentional system of the brain, butsomething is known about the areas thatcarry on three major attentional func-tions: orienting to sensory stimuli, par-ticularly locations in visual space; detect-ing target events, including ideas storedin memory; and maintaining the alertstate. Each of these areas of researchprovides information that relate to thetheories of consciousness discussed inmy first paragraph, but as will be appar-ent from the discussion below, each re-lates only partially to common definitionsof conscious processing.

Visual Orienting

How can one study attention? Crick ar-gues for the selection of a model systemthat involves consciousness in a limiteddomain. His choice is awareness duringvisual search. Visual search has alsobeen a traditional vehicle for the study ofattention. For example, if you are askedto report a vertical rectangle in Fig. 1, thetime to do so is linearly related to thenumber of distractors. This situation isthought to occur because one has toorient attention to each location. An in-dividual may do so by making an eyemovement, but it is also clear that theeyes can remain fixed and each positionexamined covertly. We now know quite alot about the mechanism that performsthis covert operation from studies of nor-mal subjects, brain-lesioned patients, andmonkeys (13).When subjects switch attention from

location to location, they activate pro-cesses in the parietal lobe of the oppositehemisphere (14). These cortical areas are

I E_ Ok

I_

-0l_

~

| _ @ _~

-o0

..0±0

_ w _~

|~_ _FIG. 1. Searching this display for a vertical rectangle requires attention to successive

locations because the nontargets include both vertical (ellipses) and rectangles (horizontal).Reaction times in this task are linearly related to the number of distractors. [This figure wasreproduced with permission from ref. 37 (copyright Elsevier, Cambridge, U.K.).]

involved in programming the switch ofattention in the opposite visual field. The

t mechanisms of the parietal lobe are notsymmetric. PET evidence shows that theright parietal lobe is involved with atten-tion shifts in both visual fields, whereasthe left parietal lobe seems restricted torightward shifts ofattention (14). There isa further hemispheric specialization aswell. The left hemisphere seems to be

[ more involved when attending to localinformation that might be important inrecognizing objects, whereas the righthemisphere seems more involved whenglobal features are involved, such as ingeneral navigation in an environment(15).These hemispheric differences are es-

pecially important in understanding howbrain damage influences our awarenessof the visual world. In normal people thetwo hemispheres operate as a unit invisual search, so that it does not matter ifall of the distractors are located in onevisual field or if they are distributedacross the visual fields. However, if thetwo hemispheres are separated surgicallyby cutting the fiber tract that runs be-tween them, this unity no longer applies,and each hemisphere's system operatesseparately, yielding a search rate that istwice as fast (16). Yet the patient is notreally aware that a change has takenplace due to the operation. If there isdamage to the right parietal lobe, subjectswill be very likely to neglect targets insearch tasks like those of Fig. 1. In otherwords, they can be unconscious of infor-mation on the side of space opposite thelesion. These same mechanisms are usedwhen these neglect patients recall infor-

f mation from memory, and they neglectthe left side of visual images (17).We know something of the route by

which the parietal mechanism influencesinformation about the target identity. Themost prominent hypothesis about thisroute is that it involves the pulvinar nu-cleus of the thalamus. There is goodevidence of activity in this general regionwhen subjects must pull out a target fromsurrounding clutter (18). Thus the inter-action of thalamus and the cortex playsan important role in our consciousness ofthe target. This finding fits well with thegeneral idea outlined by Crick.What are the consequences of attend-

ing to a visual object? We know fromcellular recording studies in monkeys(12) and from neuroimaging studies inpeople (13) that attention provides a rel-ative increase of neural activity whencompared with comparable unattendedinformation. For example, an instructionto attend to the color, form, or motion ofvisual input increases neural activity inthe regions that process this information(19). This finding is certainly consonantwith the proposal that mental eventswork on the rate oftransmission ofneuralimpulses that has been suggested by sev-eral theories of consciousness (1, 3).However, the principle of relative am-

plification by attention is a very generalone (13). It is found in all areas of thebrain that have been studied. Attention tosensory information amplifies brain areasused to processes that modality; simi-larly, attention to motor output activatesbrain areas used to generate the move-ment. This principle also appears to ap-ply to higher-level cognitive processes(13). For example, creating a visual im-

Review: Posner

Proc. Natl. Acad. Sci. USA 91 (1994)

age from a verbal instruction is nowknown to produce activation in visuallyspecific areas of the cortex that would beused to process visual input of the sametype.What is still not known is exactly how

this amplification is accomplished. Thereare theories at the neural level (20) andcellular observations that suggest a rolefor suppression of the unattended pro-cesses (21, 22), but the detailed cellularmechanisms remain to be clarified.Are the thalamic-visual cortex inter-

actions we have been describing "con-sciousness" as suggested by Crick? Onedefinition of consciousness involvesawareness of the outside world, and theinteractions of thalamic areas with thevisual cortex are certainly important forachieving focal awareness. By focalawareness I mean the type ofrecognitionthat one has of the target in Fig. 1. Tolocate the target one must effortfully en-gage different locations to be certain thatthey contain the constellation of features(form and orientation) required. How-ever, it would not be appropriate to sayyou are unaware of other objects in thefield that are not the focus of attention.One is roughly aware of the density ofnonattended objects-their extent andbasic format. It is possible to distinguishthe type of focal awareness needed tolocate the target from a more generalawareness of the background (23). Dam-age to the attention network involving theparietal lobe and associated thalamicareas produces a kind of loss of focalawareness on one side of the world. Ne-glect induced by parietal lesion may leavethe patient unconscious of this lack ofawareness, just as the split brain personis unaware that search ofthe visual worldhas lost integration. However, patientswith parietal lesions usually recoverawareness of information opposite thelesion but show a permanent loss of theability of stimuli arising there to produceorienting away from already attendedevents. This may lead to a failure to seeobjects on the left side of Fig. 1, but ifcued to that location, they become con-scious of them. In addition, normal sub-jects can attend to aspects of visual stim-uli, such as their color or form, withoutactivation of the parietal mechanisms wehave been describing. These facts seemto argue that the parietal-thalamic sys-tem represents an important pathway bywhich conscious processing is achieved,rather than consciousness itself. None-theless, the clarification of neglect thathas arisen from recent studies has helpedus grasp some important elements ofachieving focal awareness.

search all the target locations or can youconfine your search to the vertical ob-jects only? There is considerable evi-dence that search can be guided by in-formation about the color or orientationor other nonlocational features of thetarget (24). How is this implemented inthe brain? It seems that the recruitmentand control of posterior brain areas, inthis case, is supervised by an anatomi-cally distinct system that involves moreanterior structures that have sometimesbeen called an executive network (ref.13, pp. 168-174) (Fig. 2). In the study ofattentional amplification of color, form,or motion mentioned above there wasevidence for activation of a frontal atten-tional system, but no parietal activationwas found (19). It thus appears that twodifferent attentional systems serve assources of activation for color or form(frontal areas) and for location (parietal),although both may enter via the thalamusto amplify activity within the visual sys-tem at the same site (e.g., V4).

In guided search, selection by locationand selection by color or form occursimultaneously with relatively little inter-ference (24), unlike the situation for lo-cation when the corpus collosum is in-tact, in which attention cannot be sharedbetween the two fields (16). One specu-lative possibility would be that time shar-ing is possible when two anatomicallydistinct attentional sources are involved.The frontal areas that serve to guide

search appear to involve a network thatincludes at least portions of the basalganglia and of the anterior cingulate gy-rus (ref. 13, pp. 168-174). The anteriorportion of the cingulate gyrus appears tobe involved in a wide range of activitiesthat have been termed collectively "ex-ecutive function" (25). In PET languagestudies, when subjects were required toname the use of familiar nouns (e.g.,pound to hammer) activation of the an-terior cingulate along with left lateralcortex language areas was most promi-

nent. When subjects were required torespond to the ink color in which a con-flicting color name was presented(Stroop effect), there was strong activa-tion of the anterior cingulate along withprestriate color areas (13). The detectionof multiple color form or motion targetsin comparison with passive viewing ofthe same stimuli also activated the ante-rior cingulate (19). All of these situationsinvolve selection of targets from compet-ing inputs, which is considered a tradi-tional role of attention. In the case ofthisexecutive attentional network, the natureofthe target does not seem to matter verymuch.The term "executive" suggests two

important overall functions. (i) An exec-utive is informed about the processestaking place within the organization. Asystem that would be related to our sub-jective experience of focal attentionwould clearly play this function for asubset of current (sensory) and stored(memory) information. There are reasonsfor relating anterior cingulate function tofocal awareness of the target (13). Forexample, the intensity of cingulate activ-ity tends to increase with number oftargets in a set of stimuli and decreaseswith practice on any single stimulus set.These findings correspond to cognitivetheories linking focal attention to numberand difficulty of target detection.

(ii) A second function of an executiveis to exercise some control over the sys-tem. The anatomy of the anterior cingu-late provides pathways for connecting itto both the posterior parietal area and toanterior areas active during languagetasks (26). Working memory is generallythought to involve both a representationof past events and an executive systeminvolved in sustaining and transformingthis representation (27). Recent PET (28,29) and neurophysiological (30, 31) stud-ies show that lateral areas of the prefron-tal cortex play a key role in holdingon-line a representation of past events.-s~ p Ta "t rx14

Ir\-k

rArI...I

Attending to Ideas

Suppose you are asked to locate thevertical rectangle in Fig. 1. Must you

FIG. 2. The executive attentional system involves frontal structures, including the anteriorcingulate, and acts upon many different brain areas [reproduced with permission from ref. 13(copyright Freeman, New York)].

7400 Review: Posner

--- [IL II 7 t. 1 . T

_- '. 7,1 :,. --I. ;,''.-1 't '.

Proc. Natl. Acad. Sci. USA 91 (1994) 7401

Cellular recordings in the awake monkeyindicate that cells within the dorsolateralprefrontal cortex maintain a representa-tion of the spatial environment whenmonkeys have to hold in mind a locationto which to move their eyes after thestimulus disappears (30). Other cellswithin the inferior convexity hold a rep-resentation of the identity of past stimuli(31). Lateral areas of the left frontal andposterior cortex are also active in studieswhen people must obtain a quick associ-ation to word stimuli (13). While special-ized areas of the lateral prefrontal cortexappear to hold the relevant informationon-line, the anterior cingulate would beplaying a role in the executive functionsof awareness and control discussed incognitive studies and often found im-paired in subjects with frontal damage.Previously, the empirical results thatsupport the important role of the anteriorcingulate as a part of this network weresummarized as follows (32):

The anterior attention networkseems to be much more directly re-lated to awareness than the posteriornetwork, as has been indicated bythe PET studies cited previously.The use of subjective experience asevidence for a brain process relatedto consciousness has been criticizedby many authors. However, we notethat the evidence for the activationof the anterior cingulate is entirelyobjective; it does not rest upon anysubjective report. Nevertheless, ifone defines consciousness in termsofawareness, it is necessary to showevidence that the anterior attentionnetwork is related to phenomenalreports in a systematic way. In thissection, we notefive points, each of

which appears to relate subjectiveexperience to activation of the ante-rior attention system. First, the de-gree of activation of this networkincreases with the number of targetspresented in a semantic monitoringtask and decreases with the amountof practice in the task. At first onemight suppose that target detectionis confounded with task difficulty. si

But in our semantic monitoring task c

the same semantic decision must be c

made irrespective of the number of t(actual targets. In our tasks no stor- tiage or counting of targets was idneeded. Thus, we effectively disso- o

ciated target detection from task dif- tificulty. Nonetheless, anterior cingu- s(late activation was related to number clof targets present. The increase in vi

activation with number of targets juand reduction in such activation with dipractice corresponds to the common vi

finding in cognitive studies that con- v(scious attention is involved in target atdetection and is required to a greater ccdegree early in practice. As practice di

proceeds, feelings of effort and con-tinuous attention diminish, and de-tails ofperformance drop out of sub-jective experience.

Second, the anterior system ap-pears to be active during tasks re-quiring the subject to detect visualstimuli, when the targets involvecolor form, motion, or word seman-tics.

Third, the anterior attention sys-tem is activated when listening pas-sively to words, but not when watch-ing those words. This finding ap-pears to correspond subjectively tothe intrusive nature of auditorywords to consciousness when theyare presented in a quiet background.They seem to capture awareness.Reading does not have this intrusivecharacter. For a visual word to dom-inate awareness, an act of visualorienting is needed to boost its signalstrength.

Fourth, the anterior attention sys-tem is more active during conflictblocks of the Stroop test than duringnonconflict blocks. This is consis-tent with the commonly held ideathat conflict between word name andink color produces a strong con-scious effort to inhibit saying thewritten word. Finally, there is a re-lation between the vigilance systemand awareness. When one attends toa source of sensory input in order todetect an infrequent target, the sub-jective feeling is of emptying thehead of thoughts or feelings. Thissubjective "clearing of conscious-ness" appears to be accompanied byan increase in activation of the rightfrontal lobe vigilance network and areduction in the anterior cingulate.Just as feelings of effort associatedwith target detection or inhibitingprepotent responses are accompa-nied by evidence of cingulate activa-tion, so the clearing of thought isaccompanied by evidence of cingu-late inhibition.Dennett (33) provided a strong philo-

;ophical critique of those who implicitly,ling to a view that there is an arena ofConsciousness or what he calls the Car-esian Theater of the Mind. Nonetheless,he specific points made above seemed todentify cingulate activation with aspectsof awareness in so much tighter a wayhan previous efforts that it seems rea-onable to set them down with as muchlarity as possible. A somewhat similariew of the role of the cingulate in con-inction with the hippocampus has beeniscussed by Edelman (3). Edelmaniews these two brain structures as in-olved in the integration of interoceptivend exteroceptive information needed foronscious processing. In his work heistinguishes what he calls attention

(which is the increase in neural activitywithin brain areas currently of the orga-nism's concern) from consciousness,which is the source ofthese activations inthe cingulate-hippocampal system.There is a semantic difference in that Ihave called the former, the site at whichattention works, and the latter, atten-tional networks, but the distinction ap-pears minor.One new development is an increased

understanding of how the brain actuallyexecutes a voluntary instruction to at-tend to something (ref. 13, pp. 141-148).Studies employing PET have revealedimportant anatomical aspects of wordreading. Two major areas of activationappear within the visual system. A rightposterior temporal parietal area is acti-vated passively by visual features,whether in consonant strings or words. InPET studies, areas of the left frontal areaare active when subjects deal with themeaning of a word. When the process isextended by requiring association of sev-eral words to a given input or by slowingthe rate of presentation, this left frontalarea is joined by a posterior activation inWernecke's area.To study semantic and feature activa-

tion together, we used tasks that clearlyinvolved both. We measured electricalactivity from scalp electrodes. One taskwas looking for a visual feature; the se-mantic task required the subject to de-termine if the word referred to a naturalor a manufactured item. Our reasoningwas that in the visual task, subjectswould be attending to the feature level,and in the semantic task attention wouldbe directed to the meaning of the word.If, as has been described in the PETwork, attention serves to amplify com-putations, it should be possible to seeamplifications of the voltages in thewaveforms in the right posterior area infeature analysis and in the left frontalarea in semantics, depending upon thetask used. We used exactly the samestimuli in the two tasks.

Results showed that electrical activityfrom the left frontal area was more pos-itive at -200-300 msec when the taskwas semantic, whereas the right poste-rior area showed more positivity whenthe task was feature search. These effectswere not confined to single electrodesites. For the posterior area, it was pos-sible to compare the electrode sites firstshowing the greater right hemisphere ac-tivation at 100 msec associated with theprocessing of visual attributes automati-cally with those showing the amplifica-tion due to voluntary attribute search at=250 msec. Our comparison supportedthe idea that roughly the same areas thatfirst carried out the visual-attribute com-putations on the letter string were reac-tivated 100 msec later when subjectswere looking for the thick letter. This fit

ReVview: Posner

Proc. Natl. Acad. Sci. USA 91 (1994)

with the idea that subjects can voluntarilyreactivate areas of the brain that per-formed the task automatically when theyare instructed to deal with that computa-tion voluntarily. The semantic effect wasfound at several frontal sites bilaterally.This result differs from the PET data,which were strictly left lateralized, al-though there was some evidence that theleft frontal area showed the effect morestrongly than the right.A popular idea in modern physiology is

called reentrant processing. Basically,this is the idea that higher-level associa-tions are made by fibers that reenter thebrain areas which processed the initialinput (3, 34). Reentrant processing maybe contrasted with more traditional no-tions that higher functions are confined tohigher-associational areas ofthe brain. Inour studies, the visual computation oc-curs at 100 msec, followed by a semanticcomputation that begins at -200 msec.When the instruction is to search thestring for a visual feature, the electrodesaround the area originally performing thevisual computation are reactivated. Sim-ilarly, when asked to make a semanticcomputation, the area thought to performsuch computations was amplified in elec-trical activity.

If the brain operates in this way, wemight then be able to instruct the subjectto compute the same functions in differ-ent orders and thus reprogram the orderof the underlying computations. To in-vestigate this, we defined what we call aconjunction task by asking subjects ei-ther to respond to targets that had thickletters and were animate or were animateand had thick letters. Both tasks wereidentical except for the order of the com-putations.We did not expect the subjects to ac-

tually compute the functions in a serialfashion. The reaction times for targetssuggested that the subjects were able toreorder the priority of the underlyingcomputations in the conjunction task.The electrical activity also reflected theinstructed priority. The results provideus with a basis for understanding how thebrain can carry out so many differenttasks upon visual input. Some aspects ofthe underlying computations are not af-fected by instructions. The visual at-tribute area of the right posterior brainseems to carry out the computation onthe input string at 100 msec, irrespectiveof whether the person is concerned withvisual features as a part ofthe task or not.However, when the task is identified aslooking for a thick letter, these samebrain areas are reactivated and presum-ably carry out the additional computa-tions necessary to make sure that one ofthe letters has just enough thickening to

constitute a target. Attention thus canamplify computations within particularareas but often does so by reentering the

area, not by amplifying its initial activa-tion. The order of these underlying com-putations can be reprogrammed by at-tending at different times, just as theorder of attended locations in visualsearch can be changed at will.The data on computing conjunctions

implies two important ideas related to theattentional control of information pro-cessing. (i) The results ofattentional con-trol are widely distributed, resulting inamplification of activity in the anatomicalareas that originally computed that infor-mation. (ii) The source of this attentionalcontrol need not involve a system thathas access to the information being am-plified but can be a system that has con-nections to places where the computa-tions occur.As the result of activity within the

attention network, the relevant brain ar-eas will be amplified and/or irrelevantones will be inhibited, leaving the brain tobe dominated by the selected computa-tions. If this were the correct theory ofattentional control, one would expect tofind the source of attention to lie in sys-tems widely connected to other brainareas, but not otherwise unique in struc-ture. As pointed out by Goldman-Rakic(26), this appears to be the basic organi-zation of frontal midline networks. An-terior cingulate connections to limbic,thalamic, and basal ganglia pathwayswould distribute its activity to the widelydispersed connections we have seen to beinvolved in cognitive computations.

Maintaining the Alert State

The earliest studies of the anatomy ofattention involved maintenance of thealert state. Cognitive psychologists havestudied changes in alerting, by warningsignals to prepare for the task. There isevidence that an increase in alertnessimproves the speed ofprocessing events.The trade-off between improved speedand reduced accuracy with warning sig-nals has led to a view that alerting doesnot act to improve the buildup of infor-mation concerning the nature of the tar-get but, instead, acts on the attentionalsystem to enhance the speed of actionstaken toward the target (ref. 13, pp. 174-176).There has been some improvement in

our understanding of the neural systemsrelated to alerting over the last few years.Patients with lesions of the right frontalarea have difficulty in maintaining thealert state. In addition, experimentalstudies of blood flow in normal peopleduring tasks that require maintainingalertness show right frontal activation.The neurotransmitter norepinephrine

appears to be involved in maintaining the

alert state. This norepinephrine pathwayarises in the midbrain, but the right fron-tal area appears to have a special role in

its cortical distribution. Among posteriorvisual areas in the monkey, norepineph-rine pathways are selective for areas in-volved in visual spatial attention. Recentstudies (35) using drugs to reduce norepi-nephrine in alert monkeys show that thisreduction blocks the increased process-ing speed usually found with warningsignals, thus supporting the role of nor-epinephrine in achieving alertness.

Applications and Future Diections

We have a start on understanding thecircuitry that underlies orienting of atten-tion. However, more detailed cellularstudies in monkeys are necessary to testthese hypotheses and to understand morecompletely the time course and the con-trol structures involved in covert orient-ing of attention. Even more fascinating isthe possibility that the microstructure ofareas involved in attention will be differ-ent somehow in organization from thoseareas carrying out passive data process-ing. Such differences could give us a clueas to the way in which brain tissue mightrelate to subjective experience.Much remains unknown concerning

the more anterior executive system.Studies of blood flow and metabolism innormal people should be adequate toprovide candidate areas involved in as-pects of attention. It will then be possibleto test further the general proposal thatthese constitute a unified system and thatconstituent computations are localized.The idea of attention as a network of

anatomical areas makes relevant study ofboth the comparative anatomy of theseareas and their development in infancy(ref. 13, pp. 181-192). In the first fewmonths of life, infants develop nearlyadult abilities to orient to external events,but the cognitive control produced by theexecutive attention network requiresmany months or years of development.Studies of orienting and motor controlare beginning to lead to an understandingof this developmental process. As moreabout the maturational processes ofbrainand transmitter system is understood, itcould be possible to match developingattentional abilities with changing biolog-ical mechanisms. The neural mechanismsof attention must support not only com-mon development among infants in theirregulatory abilities, but also the obviousdifferences among infants in their ratesand success of attentional control.There are many disorders that are of-

ten supposed to involve attention-in-cluding neglect, depression, schizophre-nia, and attention-deficit disorder. Thespecification ofattention in terms ofanat-omy and function has already provenuseful in clarifying some of the underly-ing bases for these disorders (ref. 13, pp.217-222). The development oftheories ofdeficits might also foster the integration

7402 Review: Posner

Proc. Natl. Acad. Sci. USA 91 (1994) 7403

of psychiatric and higher-level neurolog-ical disorders, both of which might affectthe brain's attentional system.

Consciousness

The issues of selection and control cen-tral to the study of attention are alsoimportant aspects of most theories ofconsciousness. Sperry (36), in particular,has argued that cognitive control overearlier evolving neural systems repre-sents an emergent property of complexnetworks.The study of visual attention has im-

plicated a cortical-thalamic networkused for orienting that has many similar-ities to the visual awareness system fa-vored by Crick (2) as a model for the studyof consciousness. Although there may bedoubts about whether it is a completemodel, the importance of studies of thissystem as a way of integrating human andmonkey research is clear. Study ofhigherforms of attentional selectivity has fo-cused on frontal structures that havesomething in common with the analysis ofconsciousness proposed by Edelman (3)and others. The role of the cingulate andother frontal areas in the development ofvolition remains of critical importance.Neuroimagng studies have tended to con-firm the role of attention in providing arelative amplification of neural activitywithin specific sensory or motor sites atwhich computations take place. This is aview favored by Beck and Eccles (1), aswell as many others, as a critical aspect ofconsciousness. Whether the progress instudies ofattention and brain mechanismswill provide a complete analysis of con-sciousness or whether fundamentally newmechanisms such as those that mightcome from quantum physics (1, 4) will beneeded is, of course, a matter of opinion,but what seems clear is that there has beenvery considerable recent progress towardunderstanding brain networks relevant tothe production of conscious experience,

and the tools are present for considerablefuture development.

This research was supported by grants fromthe Office of Naval Research and by the J. S.McDonnell Foundation and Pew MemorialTrusts through a grant to the University ofOregon.

1. Beck, F. & Eccles, J. C. (1992) Proc.NatI. Acad. Sci. USA 89, 11357-11361.

2. Crick, F. (1994) The Astonishing Hy-pothesis (Scribner's, New York).

3. Edelman, G. M. (1989) The Remem-bered Present (Basic, New York).

4. Penrose, R. (1989) The Emperor's NewMind (Oxford Univ. Press, Oxford).

5. Zeki, S. (1993) A Vision of the Brain(Blackwell, Oxford).

6. Baars, B. (1988) A Cognitive Theory ofConsciousness (Cambridge Univ. Press,Cambridge, U.K.).

7. James, W. (1907) Psychology (Holt,Rinehart & Winston, New York).

8. Moruzzi, G. & Magoun, H. V. (1949)Electroencephalogr. Clin. Neurophysiol.1, 445-473.

9. Broadbent, D. E. (1958) Perception andCommunication (Pergamon, London).

10. Allport, D. A. (1989) in Foundations ofCognitive Science, ed. Posner, M. I.(Bradford Books/MIT Press, Cam-bridge, MA), pp. 636-82.

11. Sokolov, Y. N. (1963) Perception andthe Conditioned Reflex (Pergamon, Lon-don).

12. Wurtz, R. H., Goldberg, M. E. & Rob-inson, D. L. (1980) Prog. Psychobiol.Physiol. Psychol. 9, 43-83.

13. Posner, M. I. & Raichle, M. E. (1994)Images ofMind (Sci. Am. Library, NewYork).

14. Corbetta, M., Miezin, F. M., Shulman,G. L. & Petersen, S. E. (1993) J. Neu-rosci. 13, 1202-1226.

15. Robertson, L. C., Lamb, M. R. &Knight, R. T. (1988) J. Neurosci. 8,3757-3769.

16. Luck, S. J., Hillyard, S. A., Mangun,G. R. & Gazzag9a, M. S. (1989) Nature(London) 342, 543-545.

17. Bisiach, E. (1992) in The Neuropsychol-ogy of Conscious, eds. Milner, A. D. &

Rugg, M. (Academic, London), pp. 113-137.

18. LaBerge, D. & Buchsbaum, M. S. (1990)J. Neurosci. 10, 613-619.

19. Corbetta, M., Miezin, F. M., Dobmeyer,S., Shulman, G. L. & Petersen, S. E.(1991) J. Neurosci. 11, 2388-2402.

20. Van Essen, D. C., Anderson, C. H. &Olshausen, B. A. (1994) in Large ScaleNeuronal Theories of the Brain, eds.Koch, C. & Davis, J. (MIT Press, Cam-bridge, MA), pp. 271-300.

21. Moran, J. & Desimone, R. (1985) Science229, 782-784.

22. Chelazzi, L., Miller, E. K., Duncan, J. &Desimone, R. (1993) Nature (London)363, 345-347.

23. Iwasaki, S. (1993) Cognition 49, 211-233.24. Wolfe, J. K. M., Cave, K. R. & Franzel,

S. L. (1989) J. Exp. Psychol. Hum. Per-ception Performance 15, 419-433.

25. Vogt, B. A., Finch, D. M. & Olson,C. R. (1992) Cereb. Cortex 2/6, 435-443.

26. Goldman-Rakic, P. S. (1988) Annu. Rev.Neurosci. 11, 137-156.

27. Baddeley, A. D. (1990) Working Mem-ory (Oxford Univ. Press, Oxford).

28. Jonides, J., Smith, E. E., Koeppe,R. A., Awh, E., Minoshima, S. & Min-tun, M. A. (1993) Nature (London) 363,623-635.

29. Paulesu, E., Frith, C. D. & Frackowiak,R. S. (1993) Nature (London) 363, 342-345.

30. Funahashi, S., Chafee, M. V. & Gold-man-Rakic, P. (1993) Nature (London)365, 753-756.

31. Wilson, F. A. W., Scalaidhe, S. P. O. &Goldman-Rakic, P. (1993) Science 260,1955-1958.

32. Posner, M. I. & Rothbart, M. K. (1992)in The Neuropsychology of Conscious,eds. Milner, A. D. & Rugg, M. (Academ-ic, London), pp. 91-112.

33. Dennett, D. (1991) Consciousness Ex-plained (Little Brown, Boston).

34. Edelman, G. & Mountcastle, V. (1978)The Mindful Brain (MIT Press, Cam-bridge, MA).

35. Witte, E. A. (1993) Doctoral dissertation(Univ. of Oregon, Eugene).

36. Sperry, R. W. (1988) Am. Psychol. 43,607-613.

37. Posner, M. I. & Dehaene, S. (1994)Trends Neurosci. 17, 75-79.

Review: Posner