The spectacular improvements in Positron Emission Tomography (PET) and Functional Magnetic Resonance Imaging (fMRI) over the last three decades have led researchers to a reconsideration of the brain and the mind.1 A form of neurophilia has made its way into contemporary culture.2 The brain is no longer envisaged as an organ a subject possesses, but as an entity with which he identifies, with which he is at one.3

Behaviorism has retreated into the shadows.4 Gone are sequences of stimuli and their responses. Adieu pigeons. So far as the mind goes, things go in and things come out, but there must be something, cognitivist scientists agree, between them.5 This is very much the view of control theorists, who depict a control system in terms of an input-output mapping and a set of internal states.6 Those interested in the brain itself typically think of such states as physical or chemical; those interested in abstractions, as states of a Turing machine or some other computational device.

Imaging techniques have given a new urgency to Paul Broca’s suggestive thesis that mental functions have specific cerebral locations.7 Broca’s ideas were first studied by examining cerebral injuries, hardly the best way in which to study normal cerebral function. It is now possible, thanks to functional imaging, to visualize the metabolic activity of different cerebral areas associated with either manifest behavior or specific mental states.

Both Nuclear Magnetic Resonance Imaging and Magnetic Resonance Imaging (MRI) are achievements of the early 1970s. Hydrogen atoms in living tissue are excited by means of an intense magnetic field. The excited atoms emit a measurable frequency signal. Variations in the magnetic field evoke signals of different strengths, and the contrast between tissues reflects the rate at which excited hydrogen atoms return to an equilibrium state. The frequency of the emitted signal depends both on the concentration of protons in tissue and on the nature of their neighborhood. Maximum proton concentrations are found, obviously enough, in water. Bones emit a limited signal; fat, which is very rich in water, a more powerful one. The dynamics of water molecules in the brain form the basis of diffusion MRI, a technique used to highlight the contrast between the brain’s white and gray matter.8

In 1890, Charles Roy and Charles Sherrington observed that blood flow increases in the vessels irrigating active brain regions; and, what is more, varies according to tissue metabolism and activity.9 Accounting for two percent of body mass, the brain consumes some twenty percent of the body’s total energy. Neurons have no energy reserves. Their metabolic activity depends on the uptake of blood. The smallest variation in neuronal activity results in a variation in metabolic demand, one causing a concomitant variation in blood flow. In the brain, some groups of neurons modify their dynamics as the brain fulfills a task. Such processes demand a rapid and localized modulation of blood flow. Vasodilatation takes place a few hundred milliseconds after the modifications of synaptic activity has started, but its maximum is reached only after a few seconds of sustained activity. Additional glucose, it would seem, is consumed anaerobically. For reasons that remain unclear, more oxygen than needed still reaches the activated neurons.10 Moreover, variations from which an image is constructed stem from venous vessels and not brain tissue. One consequence is reduced diagnostic specificity, inasmuch as veins drain blood from several cortical areas, as opposed to the arteries, which irrigate them specifically.

Does functional cerebral imaging take the magnetic properties of hemoglobin into account? It does. Magnetic properties vary to the extent that hemoglobin is linked with oxygen—hence the acronym BOLD, for blood-oxygen-level dependent. Images show a contrast between oxyhemoglobin-enriched regions and those where blood flow remains unchanged.11 The relationship between increased synaptic activity, on the one hand, and increased consumption of glucose and oxygen, on the other, is far from obvious. BOLD signals do not simply represent the magnetic properties of blood. They also depend on the poorly understood physiological properties responsible for hemodynamic responses.

Functional diffusion imagery, by way of contrast, is based on the fact that neuronal activation is accompanied by reduced diffusion of water molecules. The signal directly reflects neuronal activity. Changes are simultaneous, or nearly so. Arteries or veins support cerebral territories whose size is vastly greater than that of functional neuronal units. The resolution of these images is insufficient to account in detail for the sub-units activated by a given mental operation. The link between the signal collected and effective neuronal activity is not immediate.

Whatever the technique, the assertion that an image reflects or represents a real cognitive function rests on a series of assumptions:

  • The recorded signal corresponds to the variation of local concentrations in oxygenated blood.
  • This variation results from changes in blood flow.
  • The flow is correlated to metabolic and neuronal activity.
  • Neuronal activity is the mechanism responsible for cognitive performance.

These assumptions hardly suggest a simple image of the brain.

There are several types of protocols for associating cerebral images and cognitive activities. The study of the brain at rest provides a map of basic synaptic activity. Under the so-called activation protocol, the same subject is studied in several cognitive states. The small amplitude of the response generally requires averaging several measures. Blockwise protocols expose the subject to different tasks; and in event-related protocols, the tasks are presented unpredictably. The subtraction protocol consists in comparing a task B with a reference task A. The difference B – A is then used to identify specific sites in the brain. The discrepancy between verbal and performance-based intelligence quotients has long been used to identify asymmetries in hemispheric functions in right-handed patients with epilepsy;12 and with limited clinical success, I might add. If this approach is appropriate in the study of sensory and motor functions, it is rarely useful in the study of cognitive tasks, which mobilize more complex and less differentiated networks. The method rests, after all, on the assumption that processes isolated by the protocol are dedicated; but no known cerebral module is dedicated entirely to a particular task. Conversely, several cerebral networks may take part in identical cognitive operations.

The great diversity of methods and materials used in imaging studies, as well as differences among subjects, make any comparison between different studies difficult. Cerebral and cranial anatomy sometimes makes it impossible to determine whether activation is taking place. The distortions caused by the cavities of the sinuses render the activations of the tips of the frontal and temporal lobes invisible. A region is deemed specific with respect to some mental or behavioral state, not so much because it responds to a given stimulus, but because it responds more to that stimulus than another region. It is immensely difficult to assign a given mental state to a specific substrate. In group studies, image normalization techniques serve to smooth out individual singularities.

Which brain is which?

The results are partially determined by the choice between a functional model and an anatomical model of the average brain.

No matter the techniques involved, the image that is finally observed is constructed from physical data. Filtering is at work. Only some of the signals are retained; and they are subject to statistical analysis. Principal component analysis is an example. An MRI image is constructed; it is not an image of the brain in action. Results vary accordingly. This raises an obvious question about the reliability of published work. Joshua Carp, from the University of Michigan, has shown that 4,608 combinations of different analyses could be obtained from the same original signal, according to the pre-treatment of the signals and the nature of the statistical treatment selected.13

Mismatches are not rare between the conventional data of clinical neuropathology and normal observations. A simple task involves far more distributed activity than what might be expected from cerebral injuries. It is the lack of standardization with respect to the reference state that explains why certain results do not match. In some instances, the subject is asked to think about nothing, or to daydream, but in other cases, he is asked to think of, or is exposed to, circumstances varying markedly from those recorded in the reference state.

Neuroimaging reveals that certain structures may be involved in the realization of very different operations. A meta-analysis of the relationship between language and Broca’s area considered 3,222 experimental comparisons derived from 749 articles; the results indicated that 166 studies report an activation of Broca’s area when there is language processing, but that 199 studies report activation in the absence of language processing.14 Figures for the non-activation of Broca’s area under comparable circumstances are similar. Bayes’ theorem shows that the selectivity of Broca’s area is positive but weak, which means that if Broca’s area is, indeed, associated with language processing, it is associated neither exclusively nor systematically.

Neuroimages provide only an indirect measure of brain activity: The images result from a statistical treatment; a cerebral area may be activated for multiple reasons; the recording conditions are not natural.

How could such images adequately reflect the mind’s operations?

There is no simple and obvious correlation between the activity of the nervous system and any functional units of behavior. The sequences involved in psychological or cognitive functions do not necessarily correspond to the sequence of operations carried out by the brain. There are no good grounds to assume that the brain operates according to categories matching our vocabulary or our psychological concepts. The modularity of the mind is unlike the modularity of the brain. Cerebral activities are both distributed and redundant. Cognitive functions emerge from the interaction and permanent reconfiguration of a network’s elements. But identifying the elements of the network is not enough to ascribe a functional role to each of these elements. The same region can be involved in several functions. For example, the temporoparietal junction (or angular gyrus) is involved both in the attribution of intentions to others and in the spatial integration of bodily sensations.

The acts of thinking, or evoking a mental image, or imagining a movement, suffice to cause a slight increase in blood flow and in blood oxygenation in a region, or a set of regions, of the brain. There is no evidence that these are the regions that have generated these thoughts. What MRI does reveal is the action of the mind as it causes changes to the cerebral networks affected by thought. The thoughts themselves cannot be observed directly. They can only be inferred. What, after all, can then be seen in the brain beyond the brain’s normal activity? Imagery enables us to see the mind’s instrument at work, but not its content.

The psychiatrist Édouard Zarifian has argued that the mind in thinking must contain a mechanism in the brain, a point never in dispute; it must undertake an act or series of acts in cognition; and it must assign these acts a meaning.15 Thought can not be reduced to only one of these dimensions. Each is the subject of a particular field in neuropsychology or neuropsychiatry. The brain is an organ like any other, but its cognitive processes fall outside the scope of physiology. Although auditory hallucinations involve the temporal lobe in the epileptic and the schizophrenic, the meaning they assign to this experience is generally quite different. What can we learn from neuroimaging in this case? According to Zarifian, nothing or almost nothing. The brain does not think, it conditions thought, and what is actually in play in thought and action does not originate solely from nervous physiology. The hippocampus, which plays a role in spatial memory, does not contain the knowledge of any place in particular. Only an agent with the power to move from one place to another possesses knowledge of this sort.16

How and for whom does a configuration of the brain become meaningful? A map of cortical activation is meaningless as such. A symbol represents something for someone, but a physical system cannot interpret the symbols that it generates. The progression from a particular configuration of the brain to a meaningful state of mind remains entirely unexplained. Visual perception is an example. Different characteristics of a perceived object are treated in parallel by distinct channels dealing with color, shape, movement, position, and contrast. The perceiving subject’s perceptions cannot be reduced to any particular area of the brain. “There are, in the nervous system,” Pierre Lévy has remarked quite sensibly, “only distributions of electrochemical flows and variations in pulse frequency and, properly speaking, no sounds, no colors, and no odors.”17 It is a human subject who sees colors in the world and not neuronal configurations in the brain. “It is a man,” Erwin Straus observed somewhat vaguely, but not inaccurately, “who perceives, and who thinks, and not his brain.”18

Perception, memory, cognition, or action, require an acting and conscious subject. Cognitive science cannot exist without “a black market supplied with contraband merchandise originating from existential psychology.”19

All that is very well, but what are we, if not our brains?

The common answer to this question assigns the characteristic properties of the mind to the brain. This is a method as unassailable as it is irrelevant. To say that the brain is focused, that it decides, or that it is searching, or making an effort, is incoherent. The activation of a cerebral area associated with a given function of the mind does not mean that the cerebral area determines that function: it is only an instrument. A hammer does not give rise to a house. The brain is no more conscious than the heart. I need both to be conscious, but neither is conscious of itself. Or of me.

Some regions of the brain are, of course, more important than others for the exercise of reason, memory, and self-consciousness, but that does not mean that those regions are provided with reason, with memory, or with self-consciousness. Allocating consciousness to neurons is a pretense. “We understand either a psychic discourse,” Paul Ricoeur pointed out to Jean-Pierre Changeux, “or a neuronal discourse [but] their relationship raises an issue because we cannot define how they relate to each other.”20

This is to reprise old arguments in favor of Cartesian dualism. The mind cannot be reduced to the brain. Old. But neither can the brain be reduced to the mind. New.

  1. Michael Posner and Marcus Raichle, Images of Mind (New York: Freeman & Co., 1994); Denis Le Bihan, Le cerveau de cristal : Ce que nous révèle la neuro-imagerie (The Brain Crystal: What Neuroimaging Reveals) (Paris: Odile Jacob, 2012). 
  2. Paolo Legrenzi and Carlo Umiltà, Neuromania: On the Limits of Brain Science (Oxford: University Press, 2011). 
  3. Fabrice Guillaume, Guy Tiberghien, and Jean-Yves Baudouin, Le cerveau n’est pas ce que vous pensez : Images et mirages du cerveau (The Brain is Not What You Think: Images and Mirages of the Brain) (Coll. Points de vue et débats scientifiques, Grenoble: PUG, 2013); Denis Forest, Neuroscepticisme : Les sciences du cerveau sous le scalpel de l’épistémologue (Neuroscepticism: Brain Sciences under the Epistemologist’s Scalpel) (Montreuil-sous-Bois: Ithaque, 2014). 
  4. A doctrine according to which the science of psychology is limited to the observation of behavior (motor, verbal, or neurovegetative) and that this can be explained without reference to mental events or internal psychological processes. Furthermore, behaviors can be measured, trained, and changed. See John Watson, “Psychology as the Behaviorist Views It,” Psychological Review 20 (1913): 158–77. 
  5. The autopoiesis paradigm (or operational fence) developed by Francisco Varela and Humberto Maturana, seeks to free cognitive sciences from computational dead ends. See Francisco Varela, Principles of Biological Autonomy (New York: Elsevier/North-Holland, 1979). 
  6. See, for example, Eduardo D. Sontag, Mathematical Control Theory: Deterministic Finite Dimensional Systems (New York: Springer, 1998). 
  7. See, for example, Paul Broca, “Sur le principe des localisations cérébrales (On the Principle of Cerebral Localization),” Bulletin de la Société d’Anthropologie 2 (1861): 190–204. 
  8. Diffusion tensor imaging, a variation of MRI, is focused primarily on the movement of water molecules in brain tissue. This method allows for the exploration of the connections between cerebral areas and can assist with diagnosing neurological diseases. Unfortunately, the technique is quite sensitive to artefacts associated with motions such as swallowing, breathing, peristalsis, and physical movement. The image resolution thus remains relatively low and the reconstruction of the structure of white matter fibers does not always comply with their real structure in the brain. In particular, it is impossible to distinguish the finest fibers. Contrary to diffusion imagery, which establishes the physical links between cerebral areas, functional connectivity MRI shows the interaction between these territories. Some areas are activated simultaneously for a given task, even if they are not directly connected. 
  9. Charles Roy and Charles Sherrington, “On the Regulation of the Blood-Supply of the Brain,” Journal of Physiology 11 (1890): 85–158. 
  10. In a healthy brain at rest, there is a physiological coupling between the cerebral blood flow rate (CBF), the cerebral use of glucose, and oxygen consumption reflecting local integrated synaptic activity. Whereas this coupling is usually maintained in cases of reduced synaptic activity due to a chronic pathology, this is not the case in instances of increased synaptic activity associated with a particular task. Instead, this translates as a local increase in CBF and consumption of glucose, but one that is proportionally weaker than oxygen consumption. Thus the variations in blood flow rate by far exceed the oxygen needs of the tissues which are the seat of a change in neuronal activity. Such a decoupling suggests that a portion of the energy of the neuron would be produced in anaerobiosis. However, this metabolic pathway is twenty times less productive than the aerobic pathway. The mechanism of local variations in flow rate remains unexplained because the capillary vessels involved do not have their own muscular cells which would enable autonomous regulation of their diameter. 
  11. When neurons become active, an increase in the local rate of blood flow in the affected region of the brain supplies more oxygenated hemoglobin (HbOxy) than necessary. Because this excess of oxygen is not captured by the cerebral tissue, the venous capillaries surrounding the tissue become enriched with HbOxy, displacing deoxygenated hemoglobin (desoxyHb). fMRI measures these variations in desoxyHb. However, fMRI has a fundamental limitation: the difference in response time between a synaptic activation and its vascular correlate. 
  12. See, for example, Nozomi Akanuma et al., “Lateralising Value of Neuropsychological Protocols for Presurgical Assessment of Temporal Lobe Epilepsy,” Epilepsia 44, no. 3 (2003): 408–18. 
  13. Joshua Carp, “The Secret Lives of Experiments: Methods Reporting in the fMRI Literature,” Neuroimage 63 (2012): 289–300. 
  14. Russ Poldrack, “Can Cognitive Processes be Inferred from Neuroimaging Data?” Trends in Cognitive Sciences 10, no. 2 (2006): 59–63. 
  15. Édouard Zarifian, “L’outil, les fonctions et le sens. Voir le cerveau penser est un mythe ou un fantasme (The Tool, the Functions, and the Meaning: To See the Brain Thinking is a Myth or a Fantasy),” La Recherche 289 (1996): 117–19. 
  16. The work of Pierre Karli on aggression behavior provides an experimental demonstration. Michel Bitbol dedicates most of the 700 pages of his book La conscience a-t-elle une origine? (Does Consciousness Have an Origin?) to the subject. See Michael Bitbol, La conscience a-t-elle une origine? (Does Consciousness Have an Origin?) (Paris: Flammarion, 2014). 
  17. Pierre Lévy, La machine univers: Création, cognition et culture informatique (The Machine Universe: Creation, Cognition and Computer Knowledge) (Paris: La Découverte, 1987), 186. More radically still, according to Hubert Dreyfus: “Le cerveau reçoit et transforme de l’énergie, un point c’est tout. (The brain receives and transforms energy, that’s all.)” Hubert Dreyfus, Intelligence artificielle. Mythes et réalités (Artificial Intelligence: Myths and Reality), trans. Rose-Marie Vassallo-Villaneau (Paris: Flammarion, 1984), 220. 
  18. Erwin Straus, Du sens des sens: Contribution à l'étude des fondements de la psychologie (The Meaning of the Senses: Contribution to the Study of the Foundations of Psychology), trans. Georges Thinès and Jean-Pierre Legrand (Grenoble: Jérôme Million, 1989), 271. 
  19. Erwin Straus, Du sens des sens: Contribution à l'étude des fondements de la psychologie (The Meaning of the Senses: Contribution to the Study of the Foundations of Psychology), trans. Georges Thinès and Jean-Pierre Legrand (Grenoble: Jérôme Million, 1989), 195. 
  20. Jean-Pierre Changeux and Paul Ricoeur, La nature et la règle : Ce qui nous fait penser (What Makes Us Think?) (Paris: Odile Jacob, 1998), 84. “The brain,” Michel Bitbol remarks, “is decidedly the most notable absentee of any feeling experience … My feeling brain cannot be felt spontaneously.” Michel Bitbol, La conscience a-t-elle une origine ? Des neurosciences à la pleine conscience : une nouvelle approche de l’esprit (Does Consciousness Have an Origin? From Neurosciences to Full Consciousness: A New Approach to the Mind) (Paris: Flammarion, 2014): 563.