3
INTRODUCTION Critics of ‘general intelligence’ will sometimes point out that no single measurement can do justice to the complexity of people’s mental abilities (Sternberg, 2000). Obviously this is true. People bring a very large number of characteristics – skills, proficiencies, kinds of knowledge – to bear in their behaviour, and as all work from Spearman (1904) onwards has shown, many separate sources of variance contribute to individual differences in task performance. Furthermore, and as Spearman (1927) also made clear, it is only a matter of terminology whether or not we choose to call all these separate things aspects of ‘intelligence’. The fact remains that standard tests of general intelligence, though of course they cannot measure everything, do seem to measure something important. Most interesting are problem-solving tasks like Raven’s Progressive Matrices (Raven et al., 1988), which on their own measure something closely similar to the ‘IQ’ obtained by averaging scores across a variety of sub-tests with very different content (Wechsler, 1987). Broad positive correlations, not only with all manner of laboratory tasks but with many achievements of everyday life, suggest the possibility that these problem-solving tests capture some common processing ingredient in many, perhaps most, cognitive activities (Spearman, 1904). One hypothesis is that this common processing ingredient is closely concerned with the functions of prefrontal cortex (Duncan, 1990; Duncan et al., 1995, 1996, 2000). Certainly, prefrontal lesions can impair the structure of many different kinds of behaviour (Duncan, 1986; Luria, 1966). Though not yet firmly established, this hypothesis bears importantly both on the conception of general intelligence and the understanding of frontal lobe function. WHY IT MATTERS FOR THE STUDY OF INTELLIGENCE For the psychometrician, neuroscience matters because it can inform not just the where but also the how of cognitive function. In this respect, the most detailed information is likely to come from single cell recording in the behaving monkey, and indeed, such studies suggest a number of striking conclusions (Duncan, 2001; Miller and Cohen, 2001). In much of the lateral prefrontal cortex, a substantial proportion of cells will typically respond to the events of whatever, arbitrary task the monkey has been trained to carry out (e.g. Asaad et al., 2000; Wallis et al., 2001). Responses are of many different kinds, reflecting stimulus categorization, response planning or execution, maintenance of task-relevant information across memory delays, abstract decision rules, reward expectancy or receipt and so on. Typically, cells of these different kinds will be closely intermingled, with at best statistical differences between one prefrontal recording area and another (e.g. Rao et al., 1997; Wallis et al., 2001). Even single cells, furthermore, do not have fixed response properties - instead they adapt to code the particular information of relevance to current behaviour (e.g. Everling et al., 2002; Freedman et al., 2000; Sakagami and Niki, 1994). These physiological results are reminiscent of a central concept in symbolic artificial intelligence. Though ‘working memory’often refers just to preservation of stimulus information across a brief delay, working memory in a much more powerful sense is central to problem-solving programs like SOAR (Newell, 1990) or ACT* (Anderson, 1983). Whatever the domain worked upon, from spatial navigation to trigonometry, working memory is used to build up a temporary representation or model of some aspect of the world. This model is based on a combination of perceptual input and long-term knowledge. It shows how a part of the world is structured and what goals are set. The model controls behaviour in that it provides the context for action choice; SOAR, for example, is a production system (Newell and Simon, 1972) in which a production is only executed (an internal or external action taken) when its conditions are matched in the working memory model. Obviously, task requirements lose control of the program’s behaviour when the working memory representation is impaired, in a way reminiscent of deficits after frontal lobe lesions (Kimberg and Farah, 1993). A temporary, flexible representation of task-relevant knowledge in prefrontal cortex is exactly what single cell data suggest (Duncan, 2001). Returning to general intelligence, the proposal would be that a modelling function of this sort contributes to all kinds of cognitive activity. In any given task, of course, it would be only one source of variance, combining with more specific processing functions, knowledge sources etc to Cortex, (2005) 41, 215-217 FORUM ON INTELLIGENCE FRONTAL LOBE FUNCTION AND GENERAL INTELLIGENCE: WHY IT MATTERS John Duncan (MRC Cognition and Brain Sciences Unit, Cambridge, UK)

Frontal Lobe Function and General Intelligence: Why it Matters

Embed Size (px)

Citation preview

Page 1: Frontal Lobe Function and General Intelligence: Why it Matters

INTRODUCTION

Critics of ‘general intelligence’ will sometimespoint out that no single measurement can do justiceto the complexity of people’s mental abilities(Sternberg, 2000). Obviously this is true. Peoplebring a very large number of characteristics –skills, proficiencies, kinds of knowledge – to bearin their behaviour, and as all work from Spearman(1904) onwards has shown, many separate sourcesof variance contribute to individual differences intask performance. Furthermore, and as Spearman(1927) also made clear, it is only a matter ofterminology whether or not we choose to call allthese separate things aspects of ‘intelligence’.

The fact remains that standard tests of generalintelligence, though of course they cannot measureeverything, do seem to measure somethingimportant. Most interesting are problem-solvingtasks like Raven’s Progressive Matrices (Raven etal., 1988), which on their own measure somethingclosely similar to the ‘IQ’ obtained by averagingscores across a variety of sub-tests with verydifferent content (Wechsler, 1987). Broad positivecorrelations, not only with all manner of laboratorytasks but with many achievements of everyday life,suggest the possibility that these problem-solvingtests capture some common processing ingredientin many, perhaps most, cognitive activities(Spearman, 1904).

One hypothesis is that this common processingingredient is closely concerned with the functionsof prefrontal cortex (Duncan, 1990; Duncan et al.,1995, 1996, 2000). Certainly, prefrontal lesions canimpair the structure of many different kinds ofbehaviour (Duncan, 1986; Luria, 1966). Thoughnot yet firmly established, this hypothesis bearsimportantly both on the conception of generalintelligence and the understanding of frontal lobefunction.

WHY IT MATTERS FOR THE STUDY

OF INTELLIGENCE

For the psychometrician, neuroscience mattersbecause it can inform not just the where but alsothe how of cognitive function. In this respect, themost detailed information is likely to come fromsingle cell recording in the behaving monkey, andindeed, such studies suggest a number of strikingconclusions (Duncan, 2001; Miller and Cohen,

2001). In much of the lateral prefrontal cortex, asubstantial proportion of cells will typicallyrespond to the events of whatever, arbitrary taskthe monkey has been trained to carry out (e.g.Asaad et al., 2000; Wallis et al., 2001). Responsesare of many different kinds, reflecting stimuluscategorization, response planning or execution,maintenance of task-relevant information acrossmemory delays, abstract decision rules, rewardexpectancy or receipt and so on. Typically, cells ofthese different kinds will be closely intermingled,with at best statistical differences between oneprefrontal recording area and another (e.g. Rao etal., 1997; Wallis et al., 2001). Even single cells,furthermore, do not have fixed response properties- instead they adapt to code the particularinformation of relevance to current behaviour (e.g.Everling et al., 2002; Freedman et al., 2000;Sakagami and Niki, 1994).

These physiological results are reminiscent of acentral concept in symbolic artificial intelligence.Though ‘working memory’often refers just topreservation of stimulus information across a briefdelay, working memory in a much more powerfulsense is central to problem-solving programs likeSOAR (Newell, 1990) or ACT* (Anderson, 1983).Whatever the domain worked upon, from spatialnavigation to trigonometry, working memory isused to build up a temporary representation ormodel of some aspect of the world. This model isbased on a combination of perceptual input andlong-term knowledge. It shows how a part of theworld is structured and what goals are set. Themodel controls behaviour in that it provides thecontext for action choice; SOAR, for example, is aproduction system (Newell and Simon, 1972) inwhich a production is only executed (an internal orexternal action taken) when its conditions arematched in the working memory model. Obviously,task requirements lose control of the program’sbehaviour when the working memory representationis impaired, in a way reminiscent of deficits afterfrontal lobe lesions (Kimberg and Farah, 1993). Atemporary, flexible representation of task-relevantknowledge in prefrontal cortex is exactly whatsingle cell data suggest (Duncan, 2001).

Returning to general intelligence, the proposalwould be that a modelling function of this sortcontributes to all kinds of cognitive activity. In anygiven task, of course, it would be only one sourceof variance, combining with more specificprocessing functions, knowledge sources etc to

Cortex, (2005) 41, 215-217

FORUM ON INTELLIGENCEFRONTAL LOBE FUNCTION AND GENERAL INTELLIGENCE:

WHY IT MATTERS

John Duncan

(MRC Cognition and Brain Sciences Unit, Cambridge, UK)

Page 2: Frontal Lobe Function and General Intelligence: Why it Matters

determine overall performance. Importantly, thedata would imply that such modelling is especiallyimportant in novel problem-solving tasks of thesort conventionally used to measure generalintelligence.

How might individual differences in such afunction be implemented? When an animal istrained to search for a particular target stimulus(Everling et al., 2002), many prefrontal neuronsappear to act as ‘detectors’ specifically tuned torespond to that target. Such tuning, however, is notperfect – at the same time some cells respond toirrelevant nontargets that the animal should ignore.Conceivably, variations in general intelligencecould reflect the precision of such tuning – theextent to which only relevant facts are representedin the current task model.

Certainly this is only one hypothesis, and onethat may be refined or overturned in future work.What it illustrates, however, is that physiologicaldata do not simply bear on the where of cognition,or localization in the brain (Sternberg, 2000). Theyare important for their potential to bear on how –on functional models cast simultaneously atcognitive and neural levels.

WHY IT MATTERS FOR THE STUDY OF FRONTAL

LOBE FUNCTION

Equally, a model of general intelligence like theone sketched above bears on much that is done inthe cognitive neuropsychology and neuroscience offrontal lobe function. Commonly, prefrontal cortexis supposed to underlie abstract ‘executivefunctions’ such as mental set switching orinhibition of prepotent responses. Conventional‘frontal lobe’ tests, such as Wisconsin card sorting(Milner, 1963) or nonstandard sentencecompletions (Burgess and Shallice, 1996), arepresumed to reflect such functions. A psychometricperspective suggests serious questions about thisapproach. Certainly, these ‘frontal lobe’ tasks – likeany others – will share variance with aconventional intelligence test like Raven’sProgressive Matrices. Of course, these tests willalso have more specific variance associated withtheir particular content. Which source is importantfor the test’s association with frontal lobe lesions?In terms of the account sketched above, is the keything dependence on a general task-modellingfunction, or more specific requirements e.g. toswitch between mental sets?

Each account makes fairly clear predictions. Ifthe more general variance is crucial, thenpartialling out scores on a general intelligence testshould remove differences between frontal andcontrol groups. If more specific functions such asset switching are crucial, then frontal deficitsshould remain even when general intelligence ispartialled, and ideally, several tests of a proposed

function should converge to show similar results.In practice, however, neuropsychological studiesare rarely designed for such questions to beaddressed.

Functional imaging raises similar issues. Theimaging literature shows frontal lobe activation inconnection with the widest possible variety ofcognitive demands. Commonly, activation in anyone study will be interpreted in terms of thespecific content of the task used. Comparisonacross cognitive demands, however, suggests twointeresting conclusions (Duncan and Owen, 2000).Anatomically, frontal lobe recruitment by demandslike working memory, response inhibition orperceptual difficulty is quite specific. These simpledemands produce a characteristic pattern of activityin and around the posterior part of the inferiorfrontal sulcus, in the frontal operculum/anteriorinsula, and in the dorsal part of the anteriorcingulate. In cognitive terms, however, there isvery little specificity. Instead much the samegeneral activation pattern is seen for very differenttask requirements. Again, such results suggest thepossibility that much observed frontal activityreflects adaptation of something quite general –e.g. a working memory/modelling function – to thespecific demands of any particular experiment.Certainly, imaging data should be interpreted inlight of the monkey data, showing that even singlecells change their response properties from onecontext to another, and that any region of‘activation’ in an imaging study will reflectcombined activity in intermingled cells whoseindividual properties are very different.

QUALIFICATIONS

The attempt to link general intelligence tofrontal lobe function is undoubtedly preliminary.Even if the hypothesis is broadly correct, it ishighly unlikely that prefrontal cortex works alonein supporting a function like the working memoryof programs like SOAR. Regions of parietal cortex,for example, are often active together withprefrontal cortex, and in single cell studies,prefrontal and parietal neurons can show verysimilar properties (Chafee and Goldman-Rakic,1998). Perhaps the most plausible view is thatprefrontal cortex enhances working memoryfunction in various ways – for example, byimproving focus or flexibility.

Though adaptable function restricts functionalspecializations within prefrontal cortex,furthermore, it is obvious that some specializationsexist, at least relative if not absolute. Semanticjudgments, for example, will often producestronger activity in left frontal cortex than right(Thompson-Schill et al., 1997), while the wholeorbitofrontal surface is rarely activated by simplecognitive demands (Duncan and Owen, 2000).

216 John Duncan

Page 3: Frontal Lobe Function and General Intelligence: Why it Matters

For these reasons it is highly unlikely, eitherthat an account of general intelligence can be givenexclusively in terms of prefrontal function, or thatan account of prefrontal function can be givenexclusively in terms of a single generalintelligence. This said, the hypothesis of a closelink is important. If it is correct, it has strongimplications for both fields – for the psychometricsof general intelligence and the neuroscience ofprefrontal function.

REFERENCES

ANDERSON JR. The Architecture of Cognition. Cambridge, MA:Harvard University Press, 1983.

ASAAD WF, RAINER G and MILLER EK. Task-specific neuralactivity in the primate prefrontal cortex. Journal ofNeurophysiology, 84: 451-459, 2000.

BURGESS PW and SHALLICE T. Response suppression, initiationand strategy use following frontal lobe lesions.Neuropsychologia, 34: 263-273, 1996.

CHAFEE MW and GOLDMAN-RAKIC PS. Matching patterns ofactivity in primate prefrontal area 8a and parietal area 7ipneurons during a spatial working memory task. Journal ofNeurophysiology, 79: 2919-2940, 1998.

DUNCAN J. Disorganization of behaviour after frontal lobe damage.Cognitive Neuropsychology, 3: 271-290, 1986.

DUNCAN J. Goal weighting and the choice of behaviour in acomplex world. Ergonomics, 33: 1265-1279, 1990.

DUNCAN J. An adaptive coding model of neural function inprefrontal cortex. Nature Reviews Neuroscience, 2: 820-829,2001.

DUNCAN J, BURGESS P and EMSLIE H. Fluid intelligence afterfrontal lobe lesions. Neuropsychologia, 33: 261-268, 1995.

DUNCAN J, EMSLIE H, WILLIAMS P, JOHNSON R and FREER C.Intelligence and the frontal lobe: The organization of goal-directed behavior. Cognitive Psychology, 30: 257-303,1996.

DUNCAN J and OWEN AM. Common regions of the human frontallobe recruited by diverse cognitive demands. Trends inNeurosciences, 23: 475-483, 2000.

DUNCAN J, SEITZ RJ, KOLODNY J, BOR D, HERZOG H, AHMED A,NEWELL FN and EMSLIE H. A neural basis for generalintelligence. Science, 289: 457-460, 2000.

EVERLING S, TINSLEY CJ, GAFFAN D and DUNCAN J. Filtering ofneural signals by focused attention in the monkey prefrontalcortex. Nature Neuroscience, 5: 671-676, 2002.

FREEDMAN DJ, RIESENHUBER M, POGGIO T and MILLER EK.Categorical representation of visual stimuli in the primateprefrontal cortex. Science, 291: 312-316, 2001.

KIMBERG DY and FARAH MJ. A unified account of cognitiveimpairments following frontal lobe damage: The role ofworking memory in complex, organized behavior. Journal ofExperimental Psychology: General, 122: 411-428, 1993.

LURIA AR. Higher Cortical Functions in Man. London: Tavistock,1966.

MILLER EK and COHEN JD. An integrative theory of prefrontalfunction. Annual Review of Neuroscience, 24: 167-202, 2001.

MILNER B. Effects of different brain lesions on card sorting.Archives of Neurology, 9: 90-100, 1963.

NEWELL A. Unified Theories of Cognition. Cambridge, MA:Harvard University Press, 1990.

NEWELL A and SIMON HA. Human Problem Solving. EnglewoodCliffs: Prentice-Hall, 1972.

RAO SC, RAINER G and MILLER EK. Integration of what and wherein the primate prefrontal cortex. Science, 276: 821-824, 1997.

RAVEN JC, COURT JH and RAVEN J. Manual for Raven’sProgressive Matrices and Vocabulary Scales. London: HKLewis, 1988.

SAKAGAMI M and NIKI H. Encoding of behavioral significance ofvisual stimuli by primate prefrontal neurons: Relation torelevant task conditions. Experimental Brain Research, 97:423-436, 1994.

SPEARMAN C. General intelligence, objectively determined andmeasured. American Journal of Psychology, 15: 201-293,1904.

SPEARMAN C. The Abilities of Man. New York: Macmillan, 1927.STERNBERG RJ. The holey grail of general intelligence. Science,

289: 399-401, 2000.THOMPSON-SCHILL SL, D’ESPOSITO MD, AGUIRRE GK and FARAH

MJ. Role of left inferior prefrontal cortex in retrieval ofsemantic knowledge: A reevaluation. Proceedings of theNational Academy of Sciences of the United States ofAmerica, 94: 14792-14797, 1997.

WALLIS JD, ANDERSON KC and MILLER EK. Single neurons inprefrontal cortex encode abstract rules. Nature, 411: 953-956,2001.

WECHSLER D. Wechsler Memory Scale - Revised. San Antonio:Psychological Corporation, 1987.

John Duncan, MRC Cognition and Brain Sciences Unit, 15 Chaucer Rd, CambridgeCB2 2EF, UK. e-mail: [email protected]

Cortex Forum 217