On the Limits of Localization of Cognitive Processes in the Brain.

The latest editorial contributed by Sue Hespos on August 25, 2000 in Hot Science

In the latest Hot Science piece, William Uttal takes on one of the most fundamental issues in cognitive neuroscience: Can cognitive functions be localized in the brain? Using the past as a guide, Uttal argues that it is critical for modern neuroscientists to re-open the entire localization debate. Uttal's goal is not to derail cognitive neuroscience, but rather alert us to the dangers implicit in asking questions at the wrong level...

Psychology has always been in search of metaphors and explanatory theories. Earlier we had to do with hydraulic, mechanical, electrical, and eventually computer models to serve as heuristics to help guide our thinking about the nature of cognition. In this century a new science- neurophysiology -- and a remarkable collection of new physiological recording tools have become available as an alternative to these older metaphors. We have gone through a series of physiological measures including, the galvanic skin response, the electroencephalograph, and the evoked brain potential, each of which promised to provide a material key to understanding mental activity. All of these methods were especially exciting for psychologists because they promised to provide a noninvasive means of correlating brain activity with mental actions. In the main, however, none of these methods has been successful in answering even the most basic questions of how the brain produces or encodes mental activity. The main reason for this failure has been the fact that these measures are asking questions as the wrong level. The ultimate basis of mental activity must be the informational state of a huge collection of neurons interacting, not en masse, but as an intricate web, a network in which the details of the intercommunicated information are salient. Measures of integrated activity such as the EEG or the EVBP simply do not assay the essence of the relationship between mind and brain.

The latest "new" methodology

Now there is another entry in the search for a metaphorical model. The availability of the PET and fMRI scanning procedures in the last decade has once again excited psychologists. Indeed, it has more than just excited them. Entire sections of experimental psychology in some of our most prestigious university departments have abandoned purely cognitive studies in favor of correlative studies of these images and behavioral tests. Furthermore, some departments have frighteningly over committed their resources to this single line of research. I believe this to be a programmatic error that is based upon an inadequate consideration of the basic assumptions and logic of the research that is emerging willy-nilly from this breathless attack on one of the most fundamental questions of psychobiology - the issue of whether or not mental processes can be localized in particular regions of the brain. It seems to me that there should be a cooling off period before we charge ahead into a research paradigm that has many unanswered questions and faces many conceptual, technical, and logical problems.

In the following paragraphs, my goal is to raise some cautions and to stimulate a bit of reflection about what is currently going on in many neuroscience laboratories. Some of the cautions are age-old ones, but some are associated with the most modern technical matters.

Six suggestions

First, perhaps the most difficult challenge that has to be faced by those who are comparing brain images and cognitive processes is the uncertainty involved in precisely defining the components of mental activity. Throughout the history of psychology, we have tried to define mental activity in an enormous number of different ways. Other than the antique and persisting trichotomy of "input-central-output", efforts to develop sharp definitions of mental modules have been notoriously unsuccessful. Every century defines their own mental components and few of these definitions are perpetuated into the next. A few very general terms persist - memory, emotions, percepts, etc. - but even these are fraught with lexicographic difficulties. Arguably, the mental modules that psychology currently uses are either a priori or ad hoc hypothetical constructs or are operationally defined by the experiments we use to study mental activity. At least one survey (Grafman, Partiot, and Hollnagel, 1995) goes on for seven pages listing the variety of cognitive processes that have been associated with the frontal cortex in particular! Clearly, an adequate classification of mental processes is not yet at hand.

Second, the findings that have emerged from the scanning-cognitive laboratories are not yet stable. Pulverm|ller (1999) has pointed out that the cognitive processing of word meanings has been "located" in all of the major lobes of the brain! Few studies are replicated under the same conditions, and often those that are do not support each other.

Third, there is ample evidence, especially that emerging from some of the newer event-related scanning procedures that the cognitive processes are not localized but the result of widely distributed action in the brain.

Fourth, there is a host of technical uncertainties and a highly fragile logical chain between neural activity and the scanned outputs from fMRI and PET systems and even more concern about what these signals mean. Experts in the field are well aware of these difficulties, but often we psychologists take at face value some highly dubious steps in the logic. At the very least, it must be appreciated that it is a mathematical truism that any bounded field will exhibit a maximum. This means that there will always be a peak of activity someplace in, for example, a fMRI image. Correlations between behavior and cognitive activity are, therefore, guaranteed regardless of the actual biology of the situation. The emphasis on "hot spots" incorrectly directs attention away from critical changes of activity in other regions - both increases and decreases.

Fifth, The statistical and experimental design aspects of the scanning procedures are also matters of deep concern. Small shifts in criterion levels can force drastically different interpretations of data. Normalization and averaging procedures may produce spurious conclusions concerning localization. The frailty of the subtraction and double dissociation methods, and the elaborate processing necessary to see anything at all raise serious concerns about whether this new approach will fail in the same way that the older methods did to answer the most basic questions faced by cognitive neuroscience.

Finally, despite its implicit acceptance by many researchers in this field, the localization versus distribution issue remains unresolved. There is a theoretical bias toward "localization" abroad in cognitive neuroscience these days that may be totally unjustified. The entire scanning-cognition effort is based upon the assumption that mental processes or modules are actually localized in particular regions of the brain. However, there is abundant evidence that this may be a misreading of the data. The brain is a highly interconnected, redundant, and nonlinear system that is more likely to use a distributed representation scheme than a highly localized one. Localization is an easy way out for experimental design, but it may be fundamentally incorrect in principle. Not in the sense of any obsolescent idea of "mass action" but, rather, in terms of a complex network of interacting parts. There is, in this regard, a great confusion in this field over such a simple matter as the necessity versus the sufficiency of a brain region's role in a cognitive process. Experiments may quite properly show that one region of the brain is necessary to carry out some mental task, but that does not rule out the possibility that many other regions are also required for the process to occur. The "necessary" region may not be "sufficient" to encode the cognitive act. The emphasis on associating one or a few regions with some cognitive task may thus produce an illusion of localization where none, in fact, exists.

Conclusion: The Challenge

I hope that my readers will not do the field of cognitive neuroscience the disservice of dismissing this essay as just a "pessimistic" view. Given the state of the science, it may be more realistic than pessimistic. At the very least, it seems to me that we should be considering these issues rather than plunging ahead into what may be an enormous waste of resources and time. Whether, my point of view is correct or not, there is an obligation to at least consider the questions that are raised here.

In this brief opinion piece, it is not possible for me to provide the scientific citations to support the assertions that I make. A much more complete rendition of the argument against an assumption of brain localization, and, thus, the importance of a considered evaluation of what psychologists are doing in scanning laboratories is presented in my forthcoming book - The New Phrenology: The Limits of Localizing Cognitive Processes in the Brain. (MIT Press.Spring 2001)


Grafman, J., Partiot, A., & Hollnagel, C. (1995). Fables in the prefrontal cortex. In Behavioral and Brain Sciences, 18, 349-358.
Pulvermüller, F. (1999). Words in the brain's language. In Behavioral and Brain Sciences, 22, 253-336.

by William R. Uttal
Professor Emeritus (Psychology) University of Michigan
Professor Emeritus (Engineering) Arizona State University