The claim of the multiple realizability of mental states by brain states has been a major feature of the dominant philosophy of mind of the late 20th century. The claim is usually motivated by evidence that mental states are multiply realized, both within humans and between humans and other species. We challenge this contention by focusing on how neuroscientists differentiate brain areas. The fact that they rely centrally on psychological measures in mapping the brain and do so in a comparative fashion undercuts the likelihood that, at least within organic life forms, we are likely to find cases of multiply realized psychological functions.
Hilary Putnam's original multiple realizability claim, that the same mental state can be realized by different brain states, and/or that the same brain state can realize different mental states, has become orthodoxy in the philosophy of mind. He employed it to argue against a version of the identity theory popular in the 1960s, which sought to identify mental states with specific brain states. In sum, if the relation between mental states and brain states is many to many--the main thrust of multiple realizability--then no identity relation holds between mental states and brain states. One common corollary of this rejection of the identity thesis is the contention that information about the brain is of little or no relevance to understanding mental processes.
When Putnam (1967) characterizes brain states he treats them as physical-chemical states of the brain although he never tells us just what these are. Moreover, while it may be an intuitively plausible way of characterizing brain states, the notion of a brain state is actually something of a philosopher's fiction. A notion closer to what neuroscientists would use is activity in the same brain part or conglomerate of parts, thus making the identification of these separate parts a task critical to the practice of neuroscientists. Below, we discuss how the scientifically operative notion of a "brain state" differs from the sort of fine-grained conception of brain states employed in philosophy. In section 1, we focus on neuroanatomical brain mapping research, emphasizing two noteworthy points: (1) even in neuroanatomy, the appeal to function, especially psychological function, is an essential part of the both the project and its tools, and (2) the cartographic project itself is frequently carried out comparatively--across species. Yet, to heed the call of multiple realizability, not only would brain taxonomy have to be carried out independently of mental function, but it would also have to proceed without comparative evaluation. We extend this line of thinking into section 2, where we emphasize the reciprocity of the heuristic relationship between psychological function and brain mapping research.
Since the multiple realizability argument has been so influential in philosophical circles, in part 3, we undertake to explain why. While we believe there are several reasons, we concentrate on two, in particular: (1) different grain sizes have been used in the taxonomies of mental and brain states when establishing the one-to-many and the many-to-one mappings between them, and (2) the multiple realizability argument has been presented in a contextual vacuum; i.e., its advocates have typically failed to provide us with a statement of the identity conditions of mental states and brain states to which they were implicitly appealing in asserting cases of one-to-many and many-to-one relationships.
1. Neuroanatomical Approaches to Mapping the Brain
The task of mapping the brain has been a challenging and controversial one. The brain does not come pre-delineated into parts, no more than does the world's geography. Rather, investigators have to decide what kinds of criteria to use to mark borders. In this section, we discuss the success of some of the more straightforwardly neuroanatomical approaches, setting aside functional criteria until the next section.
One of the most celebrated maps of the brain was produced at the beginning of the 20th century by Korbinian Brodmann. At roughly the same time, the neuron doctrine, or the view that nerve fibers are separable, individual cells, was just gaining acceptance. This doctrine rested largely on the development of selective stains, such as the Golgi stain, which also made it possible to recognize different types of neurons in the cortex, and to discover that the cortex generally consisted of six different layers of neurons (which stained differentially). Also, it soon became apparent that there were significant differences across the cortex according to the types of cells found and the thickness of different layers. This provided the basis for differentiating brain areas and developing maps of the cortical surface, such as Brodmann's well-known map of 47 areas in the human brain. Significantly, Brodmann's goal in identifying different regions of the brain was, ultimately, to understand function; he writes: "Although my studies of localisation are based on purely anatomical considerations and were initially conceived to resolve only anatomical problems, from the outset my ultimate goal was the advancement of a theory of function and its pathological deviations." (1909/1994, p. 243). It is because they were likely to be functionally important that Brodmann thought cytoarchitectonic differences should matter in differentiating areas. Also, his work was distinctly comparative, drawing from 55 species ranging over 11 different orders of mammals, including other primates. Throughout, Brodmann used the same numbering systems to identify what he took to be homologous areas in the different species, arguing that there is similarity in the overall patterns of parcellation, constancy in broader regions across species, and persistence of individual areas. While his map was later criticized (see Lashley and Clark, 1946), and other researchers developed somewhat different cortical maps, Brodmann's map was widely accepted, and is still used today as a common reference.
Many of the advances beyond Brodmann in the current day involve developing maps of the brain at a much finer resolution than Brodmann's. For example, working just in the visual cortex of the macaque (principally Brodmann's areas 17, 18, and 19), David van Essen and his collaborators have differentiated 32 different processing areas (Felleman and van Essen, 1991). Just as with Brodmann, there remains considerable uncertainty as to exactly how many brain regions there are. Moreover, there are a variety of tools used to identify brain regions. Felleman and van Essen single out three different sets of criteria that figure prominently in their study: architectonics, connectivity, and topographical organization. Architectonics, Brodmann's tool, they note, has been useful in identifying only a minority of the areas of visual cortex. Topographical organization, which refers to the orderly projection of the visual field over each area, was useful in distinguishing about half of the areas, while common connectivity patterns between cells in one area and those in another, were useful for identifying most all of the areas. We should emphasize that topographical organization and connectivity are both clearly features of functional import. Connectivity is important insofar as it provides the vehicle for information to be moved from one processing area to another, and topography preserves the orderly arrangement of the visual scene, as projected onto the retina, so as to allow spatial relations in the processing area to stand in for spatial relations in the visual scene. Lastly, we should emphasize that while the goal is to understand human visual processing, this work has been carried out mainly in the macaque, with the clear assumption that organizational findings in them will generalize to humans.
2. Bets against Multiple Realizability in Empirical Research: Using Psychological Function in Brain Mapping and Neural Processing Information in Decomposing Cognition
Next, we will summarize a body of research which typifies the sort of productive interactions to be had between neuroscientific research and cognitive research when multiple consilience is assumed, and multiple realizability eschewed. We begin with the central role of function in brain mapping research.
Franz Josef Gall's familiar phrenological maps constitute one of the earliest functionally-based maps of the brain. While a superb neuroanatomist, his work suffered from several bad assumptions, such as that the cranium provided a reliable measure of the size of the underlying brain area, and that the size of a brain region was directly proportional to its functional effectiveness. Although phrenology ultimately failed, two important aspect of Gall's legacy are: (1) he demarcated areas in light of psychological functions, thus giving each region a functional characterization and (2) he based his claims on a comparative study of skulls, thus assuming that his mappings would hold across species.
Despite Gall's failure, other localizing approaches soon followed, most notably, deficit studies. The locus classicus for this approach was Paul Broca's famous research in the last century on a subject named "Tan" (so dubbed because of the one syllable he could utter) who had suffered damage to the third convolution of the left frontal lobe with the result of loss of articulate speech. Another famous deficit subject was Phineas Gage, who, after suffering a lesion in ventromedial pre-frontal cortex, underwent severe changes in personality and behavior. In interpreting such deficits, researchers then and now have implicitly rejected multiple realizability between human brains; they assume that the same brain areas are subserving the same cognitive function in all human brains, so that damage to a brain area results in a deficit to a particular cognitive function that is performed by that area in undamaged brains. It is also noteworthy that many of the lesion studies that are employed to understand the operation of the human brain are actually done on other organisms, with the results being extrapolated to the human case. While certain obvious problems exist (e.g., that functional deficits following brain damage do not necessitate that the damaged area actually performed the now deficient function), many weakness in deficit studies can be mitigated by combining them with other approaches, such as the stimulation techniques pioneered by Gustave Fritsch and Eduard Hitzig in 1870, by David Ferrier in 1886, and in this century, by Wilder Penfield and other investigators. Stimulation procedures seek to identify a psychological function with particular brain areas, and in the hands of Ferrier, especially, was pursued comparatively.
Recently, neuroimaging (especially PET and fMRI), have provided an additional approach to identifying functionally significant areas in the brain (see, for example, Petersen et al., 1989). From our vantage point, four aspects of neuroimaging research are particularly relevant to the question of multiple realizability. First, the analysis of mental processes plays a central role in the interpretation of activation patterns; both in the subtractive techniques employed to hone in on the brain activity which is specific to a given function, and in the task analysis itself, determining how best to exemplify and tax certain mental capacities in test subjects. Second, there are differences between individual brains; far from deterring development of neuroimaging, this has led researchers to pursue ways to map activations onto a common coordinate system. Third, the relatively low signal to noise ratio, especially in PET, makes it necessary to average data across subjects, thus canceling out individual differences and highlighting commonalities. The fact that any results at all survive averaging as well as transformation onto a common brain indicates a great deal of commonality--much more than the multiple realizability arguments would have us believe. Finally, most neuroimaging to date is performed on humans, but given the lack of detailed neuroanatomy for humans, the results are mapped onto the more detailed neuroanatomical analyses developed for other primates.
We now turn to examine the converse side of the interdisciplinary relationship we have been discussing and summarize a case study drawn from visual processing research of using differentiations discovered in neural processing to guide cognitive decomposition. Much of the work in decomposing visual processing has taken place at the micro-level, and can be traced back at least to the turn of the century in the work of Brodmann, on the one hand, and Salomon Eberhard Henschen, on the other, both of whom advanced our awareness of specialized visual processing regions in the cortex. Today, using advanced single-cell recording techniques, researchers such as Semir Zeki (1993) further confirm the functional specialization of cells from different visual regions.
In a series of landmark studies in the early 1980s Leslie Ungerleider and Mortimer Mishkin advanced a decomposition at the level of neural pathways. Relying largely on lesion studies conducted in monkeys by Pohl, and others, in the early 1970s, they differentiated two main routes for processing visual information. Given the deficits produced by the lesions, they assigned to one of the pathways the task of analyzing "the physical properties of a visual object (such as its size, color, texture and shape)" (Mishkin et al., 1993, p. 20); this is the so-called "what" pathway. The functional significance of the other route, the "where" pathway, was confirmed when damage to a particular part of it in monkeys resulted in their inability to select a response location on the basis of a visual landmark.
The proposed distinction between these two pathways motivated a synthesizing proposal by Livingstone and Hubel (1988) to relate further visual processing details, such as the distinction between magnocellular and parvocellular processing streams through the lateral geniculate nucleus, with the what and where systems of Mishkin and Ungerleider. For our purposes, what is important about this research is that the idea of decomposing visual processing into two separate processing systems was suggested by neuroanatomy and would not likely have been proposed drawing purely on behavioral data. Moreover, although behavioral perceptual studies with humans were used to support the decomposition, the original work was done with various species of monkeys - a leap made on the assumption that their cognitive processing is likely to be substantially similar to humans.
The account of two processing system we have presented, following Mishkin and Ungerleider and Livingstone and Hubel, while offering a novel decomposition of the visual processing system that accounts for a great deal of the data, turns out to be too simple. We have already mentioned above the research of Felleman and van Essen (1991) identifying 32 different processing areas in the macaque. Their work also shows that the different processing areas are highly interconnected, with approximately 1/3 of the possible interconnections realized, most of them reciprocally. Moreover, there is a substantial amount of cross-talk between components in the different processing streams. The evidence seems to suggest that one can identify different primary processing streams along the general lines suggested by Ungerleider and Mishkin, but that the distinctions are not as absolute nor as simple as first proposed. Instead of two, largely segregated routes, it is a more accurate to say that the processing components early in the visual system take on responsibility for processing different kinds of information about visual scenes, and that later areas in the system are dedicated to solving specific sorts of problems (e.g., coordinating limb movements). This is just the sort of decomposition that can guide further cognitive analysis of visual cognition.
3. A Diagnosis of Why Multiple Realizability Looked Plausible
Our efforts have been aimed at discrediting the assumption that multiple realizability can be established by showing that psychological states are in fact multiply realized across individuals or species. Given the use of psychological function and comparative analysis in identifying brain areas, there is little room left for identifying different realizations of mental states. Yet, a further interesting question arises as to why the claim of multiple realizability of mental states has been so compelling in the first place. One reason is almost certainly that those advancing the claim were not attending to the actual procedures by which neuroscientists identify brain areas but rather to an intuitive view of what would constitute sameness or difference. It seems obvious, for example, that a rat brain is sufficiently different from a human brain so that one could never treat their states as identical. It seems plausible to say that while there are differences in brain states between organisms, there are still many circumstances in which we would attribute the same cognitive state to them, thus making them appear to be multiply realized. Despite the intuitiveness of the claim of neural differences, our endeavor has been to show that such differences do not prevent neuroscientists from identifying common brain areas in different species. Accordingly, neuroscientists attempt to identify the same brain areas and same brain processing in different organisms despite whatever differences there are.
But there is another way we can look at this same issue. When comparing psychological states across different individuals, psychologists also tend to ignore differences and focus on commonalities. Likewise, philosophers such as Putnam, who proposed comparing mental states such as hunger across species as remote as humans and octopi have abstracted away from differences. At anything less than a very abstract level, hunger is different in the octopi than in humans; nonetheless, just as neuroscientists abstract away from differences between brains in identifying brain areas and brain processes, so do psychologists and philosophers in identifying mental states. Thus, one diagnosis of what has made the multiple realizability claim as plausible as it has been is that researchers have employed different grains of analysis in identifying mental states and brain states, using a coarse grain to identify mental states as the same across individuals and species and a fine grain to differentiate brain states. Having invoked different grains, it is relatively easy to make a case for multiple realizability. But if the grain size is kept constant, then multiple realizability looks far less plausible. One can adopt either a coarse or a fine grain, as long as one uses the same grain on both the brain and mind side. For example, one can adopt a relatively coarse grain, equating mental states over different individuals or across species. If one employs the same grain, though, one will equate activity in brain areas across species, and a one-to-one mapping is preserved. Conversely, one can adopt a very fine grain, and differentiate mental states between individuals, or even in the same individual over time. If one adopts a similarly fine grain in analyzing the brain, then one is likely to map the mental differences onto brain differences, and brain differences unto mental differences. It is usually the case that in apparent instances of different brain activity failing to produce apparent differences in mental states, one has usually just not used a fine enough grain to analyze mental states. Even when taking the functionalist's tack of identifying mental states in terms of their input and output relations to other states and behaviors, these inputs and outputs themselves are open to variations in grain. The belief that it's about to rain, for instance, might be construed broadly, as wetness-avoiding-behavior, or narrowly, as umbrella seeking behavior.
The appropriateness of grain size, of course, depends on the reason why one is making the comparison in the first place. This raises the next reason why we believe that multiple realizability has been so readily accepted: the lack of context. Whenever one asks whether two items are the same or different, the question makes little sense unless one asks about sameness or difference with respect to some other consideration. A human's mental state and that of an octopus might well be counted as the same in so far as they are associated with some general feature (such as food-seeking behavior, in the case of hunger). But with respect to other considerations, such as how one seeks the food, what foods are sought, and under what conditions, etc., the food-seeking behavior is different; correspondingly, there would then be a difference in mental state. This much seems simple and apparent, but the assertion that what we broadly call "hunger" is the same mental state when instanced in humans and octopi has apparently been widely and easily accepted without considering what contexts would make such equation of mental states reasonable. If we consider the context, and keep it fixed in doing comparative analysis both psychologically and neurally, then it is far less likely that we will come up with genuine cases of multiple realization.
One possible context which could inform the discussion of sameness and difference within mental states and brain states is the one which we have focused upon here, that invoked in neuroscientific research into brain taxonomy. It is clear that neuroscientists define the context broadly, and as a result have found relevant similarities in brains across species, as well as in different brains within the same species. Two consequences of this choice of wide grain is that it has enabled researchers to make powerful predictions about the cognitive effects activity (or lesions) in a given area will have across individuals and species, and has begun to foster enhanced understanding of the information processing components that underlie behavior.
We urge that while multiple realizability sounds intuitively plausible, it nonetheless constitutes a bad wager. Evidence that psychological functions are in fact multiply realized is not likely to be forthcoming given the practices of neuroscientists. Betting on multiple consilience instead allows for the fruitful use of neuroscience in guiding our understanding of cognitive systems. For some theorists, multiple realizability is an essential part of functionalism, thus suggesting that the fall of multiple realizability would take functionalism down with it. We do not see this as inevitable; even if one can identify mental states with activity in brain areas, that does not render the functional characterization of mental states any less important. As we have emphasized, the identification of brain areas cannot even proceed without the guidance of functional characterization.
Brodmann, K. (1909/1994). Localisation in the cerebral cortex. Translated by L. J. Garey. London: Smith Gordon.
Felleman, D. J. & van Essen, D. C. (1991). Distributed hierarchical processing in the primate cerebral cortex. Cerebral Cortex, 1, 1-47.
Lashley, K. S. and Clark, G. (1946). The cytoarchitecture of the cerebral cortex of Ateles: A critical examination of architectonic studies. Journal of Comparative Neurology, 85, 223-305.
Livingstone, M. and Hubel, D. (1988). Segregation of form, color, movement, and depth: Anatomy, physiology, and perception. Science, 240, 740-49.
Mishkin, M., Ungerleider, L. G., and Macko, K. A. (1983). Object vision and spatial vision: Two cortical pathways. Trends in Neurosciences, 6, 414-417.
Petersen, S. E., Fox, P. T., Posner, M. I., Mintun, M., and Raichle, M. E. (1989). Positron emission tomographic studies of the processing of single words. Journal of Cognitive Neuroscience, 1, 153-170.
Putnam, H. (1967). Psychological predicates. In W. H. Capitan and D. D. Merrill (Eds.) Art, mind, and religion (pp. 37-48). Pittsburgh, University of Pittsburgh Press.
Zeki, S. M. (1993). A vision of the brain. Oxford: Blackwell Scientific.