The practice of clinical neuropsychiatry often places me in philosophical doubt. Having been brought up in the tradition of biology and naturalism, I support the idea that mental phenomena are caused by neurophysiological processes and are therefore features of the brain. The version of mind—brain monism I espouse has been labeled "emergentism"; that is, mental processes are emergent properties of a complex system (the brain), just as solidity is an emergent property of the lattice structure of molecules and heat of the kinetic energy of molecules. As I argue for some kind of psychoneural identity, my scientific background pulls me toward a physical understanding of mental phenomena. I wish to understand macro phenomena in terms of micro phenomena. In other words, I want to make things quite simple. The risk of such a process is to make them too simple.
Let me take an example from everyday psychiatry. When a patient suffering from a depressive disorder asks me in an incredulous tone how drugs could relieve her depression, I usually launch into a monologue on the molecular basis of emotions, referring to a vast cast of neurotransmitters and neuromodulators and their probable abnormalities. For the more sophisticated, I may link these molecular explanations to the neuroanatomical sites that may be responsible for mood states, and yet again to vegetative functions. If I am still unable to convince the more thoughtful patient, it does not come as a surprise. Top-down or bottom-up, my explanations still require many leaps of logic. How, when, and where does altered serotonergic transmission become depression? Why does her self-esteem then take such a tumble, and why is she not able to sleep? Where do those derogatory voices arise from? I begin to question the adequacy of the reductionistic tendencies in my philosophical baggage.
In many ways, the concept of reductionism is confusing and has often been criticized. As an ideal, it was a feature of the positivist philosophy of science, a philosophy that is much discredited. Reductionism as a concept is, however, here to stay. The underlying concept is the idea that certain things might be shown to be nothing but certain other sorts of things. Any attempt to understand mental processes in terms of neurophysiological processes is therefore reductionistic in spirit. The pull in science is toward a physical understanding of all phenomena, and this is what we wish to achieve for mental phenomena as well.
A crucial aspect of our understanding is that there are different types (or levels) of reduction.1 The most important form for our purposes is ontological reduction. Ontology asks the question, "What is it?" This form has clearly been important in the history of science and has led us to sometimes surprising conclusions that tables are nothing but molecules, or that genes are nothing but DNA molecules. Mental processes, by this argument, are nothing but the movement of molecules in the brain. The kind of ontological reduction we are particularly interested in is property ontological reduction. Some examples of this in the natural sciences are that the heat of a gas is nothing but the mean kinetic energy of molecule movements, or that solidity is nothing but molecular movements in lattice structures. A similar reduction is implicit in the statement, "Depression is nothing but decreased transmission of serotonin in the brain."
Intratheoretic reductionism has been another favorite concept with philosophers.2,3 A large part of our discussion on reductionism involves understanding psychological theories in terms of neurobiological ones. Typically such a reduction is a complex affair, and the results are sometimes surprising. That the pitch of a sound is no more than its frequency is a surprising understanding that could not be achieved without first realizing that sound itself consists of a train of waves. It is not knowledge that we can reasonably expect someone to conjure up or even comprehend without priming by an authoritative source. It is this kind of reduction that has led to major advances in science. The science of optics was no longer the same once light was demonstrated to be electromagnetic radiation. Furthermore, the causal powers of the reducing phenomena should explain the causal powers of the reduced entity. We therefore need a neurophysiological theory powerful enough to also be adequate as a psychological theory.
Another term to be introduced at this point is supervenience.4 Mental states are said to be supervenient to neural states such that a difference in mental states involves a corresponding difference in neural states. As an angry person gradually becomes calm, his neural state correspondingly changes. This supervenience is causal; that is, the neural phenomena cause the mental phenomena. The same neural state will produce the same mental state, but the reverse does not follow, as a particular mental state may be caused by different neural states. As discussed later, this has consequences for our descriptions of mental states; although it may be possible to substitute mental-state descriptions for neural-state descriptions in particular instances, a general substitution may never be feasible.
In order to understand how neurophysiological processes may possibly explain complex mental phenomena, we must study brain activity at many levels. The brain is organized as a composite of structures at a spatial scale that varies over many orders of magnitude (from Angstrom unit to meter): molecules, synapses, neurons, networks, layers, maps, systems. The nature of the information processing that occurs is different at each level, and for us to be able to build an empirical superstructure requires research to occur at all levels of organization. It is as important to understand what happens at the level of the synapse as it is to understand the working of networks of neurons and neuronal systems. Neuroscientific theories must co-evolve so that research at one level provides input into research at both higher and lower levels. Increasingly subtle and sophisticated behavioral studies in experimental psychology and ethology have greatly deepened our understanding of what exactly are the psychological capacities that need to be understood, thereby clarifying the molar phenomena for which neurobiology seeks mechanisms.
There are a number of neurophysiological techniques that are capable of resolution, whether temporal or spatial, at different levels of analysis. These techniques are beginning to provide neurobiological data that have a bearing on the time-honored philosophical issues pertaining to the mind. In the process, they are changing our very approach to mental phenomena and psychiatric disorders. Take the example of neuroimaging. Computed tomography and magnetic resonance imaging provide images of the brain that capture its history over several decades. Functional imaging enables us to visualize the working of neuronal systems in time frames ranging from minutes (positron emission and single-photon emission tomography; PET and SPECT) to seconds (functional MRI and electroencephalographic mapping). Microanatomical and histochemical techniques have also advanced sufficiently to examine brain structure in fine detail. Using microelectrodes, iontophoresis, and voltage-sensitive dyes, we can examine neural events over times as brief as a few milliseconds, thus providing excellent spatiotemporal resolution. It is no wonder that philosophical questions are beginning to be addressed by neurophysiologists.
We are confronted with a serious problem, however. The complexity of mental life is so incredible that one wonders if the structure of the brain can support such a development. Is the human brain complex enough to cause all of our thoughts, feelings, and actions? To answer this question, we must examine the architecture of the brain with an argument borrowed from Churchland4 and Flanagan.5 The single neuron, complex though it is in its architecture, is a relatively simple piece of information processing material. It is from the organization of networks of neurons that one can see the emergence of complex mental processes. It is conservatively estimated that the brain has about 1011 neurons, with some estimates being 10- to 1,000-fold higher. If each neuron has 1,000 synaptic connections, and each synapse has, in turn, 10 possible activation levels, there are a staggering 10 to the power of 100 billion possible neural states. If 99.9% of these states are nonfunctional, and of the functional states, 99.9% are unconscious, we still have an unimaginable 1099,999,999,999,994 neuronal states for conscious mental life. Compare this with an estimated 1087 primary particles in the universe, and it comes as no surprise that the brain has sufficient power to generate the most complex of mental states.5 The brain is very energy efficient as well. A neuron uses roughly 10—15 joules of energy per operation, whereas an efficient silicon chip typically uses 10—7 joules per operation. The brain is therefore 7 or 8 orders of magnitude more energy efficient. The fastest digital computers are capable of around 109 operations per second; the brain of the common housefly performs about 1011 operations per second when merely resting.
The brain also has some constraints, a major one being evolutionary. The brain has evolved naturally, and nature does not have the luxury of dismantling anything that is not optimal and starting all over again. It can only modify what has gone before. The brain is limited in its size, which is constrained, among other things, by the size of the female pelvis. In the human brain, the total length of wiring is about 108 meters, and it has to be packed into a volume of 1.5 liters. A large part of the brain is necessary to carry out the vital functions of life, such as breathing, controlling circulation, moving, and sensing, and only a portion is devoted to cognition.
Having satisfied ourselves that the task of reducing mental states to neurophysiological ones is not obviously hopeless, we ask, "To what extent can we understand the underlying neuronal states?" We know that neurons network with each other. The functioning of some simple networks has been studied sufficiently to enable modeling of the various states that the network can be in.6 In such cases, we may be able to use a bottom-up approach. A model for decision making in the insect brain is the consequence of such an approach. The complexity of the human nervous system makes reliance on an exclusively bottom-up approach impractical, and multiple levels of analysis are necessary. One can see three possible domains for such an analysis—phenomenology, neuroscience, and psychology—and each of these is important.
Neuroscience has made excellent gains in the understanding of some mental processes. A major neuroscientific quest has been to understand sensory qualia, or why things seem that they seem to be. A notable example is visual perception. Starting with the early work of Hubel and Weisel,7 the anatomical basis of vision has been extensively investigated.8 From the earlier work, we understood how a line is perceived as a line and how a moving object is noted for its movement. The complexity of the visual cortex and its interconnectivity have now been well mapped out in the nonhuman primate as well as in man.8 We understand that different aspects of the act of perceiving an object have neuroanatomical correlates. The dynamic process of the activity of the visual system in not simply seeing, but also perceiving, is illustrated best by PET investigations. Some examples follow.
It has been shown through some elegant experiments that the cerebral blood flow responses to four types of visual stimuli (words, pseudo-words, strings of letters, and false fonts) are different and reproducible.9 The response to pseudo-words and words (i.e., meaningful stimuli) is particularly prominent. It is possible to distinguish from these data the neural correlates of various characteristics of the stimuli: visual features, letters, orthographic regularity, and meaning. The brain analyzes the visual features of the stimuli, regardless of their meaning, at one level, and it analyzes the visual word form at another. A lexical experiment using PET10 again demonstrated how different brain states correspond to different tasks. This experiment examined brain activation during a set of hierarchical language tasks: passively viewing words, listening to words, speaking words, and accessing their meaning, and demonstrated distinct brain activation for each.
Let us invert the process and argue that PET imaging information is available on the brain state of an individual but that this individual is unable to communicate her mental state in any other form. Is it possible then to infer the mental state from a knowledge of her brain state as visualized on PET? If we can, it would be a perfect example of an intratheoretic reduction. My thesis is that possibly we can infer mental state in this way—perhaps not completely, and perhaps not with certainty, but with a high statistical probability. There are examples of many other mental states for which corresponding neural states have been studied and consistently demonstrated. Such demonstrations are not restricted to PET investigations. Other functional imaging techniques with good temporal resolution have achieved similar results (examples are functional MRI changes with auditory perception or event-related potentials associated with cognitive processes). These experiments are a good demonstration of our ability to reduce mental phenomena to neural processes that have some specificity and reproducibility. There are, however, many limitations to this form of reductionism.
The first issue is whether such reductionism is applicable to simple mental states, such as visual perception, but would fail if we extended it to more complex states, such as emotions and thinking. An error in this argument is that intuitions about the simplicity or complexity of a mental state in neural terms may be misleading. An example is the neural complexity of what appears to be a simple activity: holding a pen in a pincer grasp to write. Computational scientists have not yet been able to simulate this function in robots. Furthermore, the growth in our understanding of the brain has been exponential in the last 10 years, and what seems too complex now may no longer be so in another 10 to 20 years.
The epithet category error is sometimes considered sufficient to reveal the naked nonsense of reductionism. The argument goes like this: the category "mental" is remote in meaning (i.e., completely different) from the category "physical," and it is therefore absurd to talk of the brain as feeling or seeing, just as it is absurd to talk of the mind as having neurotransmitters or conducting current. This objection has been repudiated by some philosophers.4 Categories are linguistic entities that reflect the worldview at the time of their construction. We have seen major changes in the worldview in the past without the necessary changes in the categorical epithets, which makes the theories look funny until the underpinnings are revealed. To talk of heat as molecular motion is on the surface a category error; to equate mass with energy is another category error that has scientific respectability. Instead of categories, we should refer to levels of analysis and acknowledge that different levels are necessary to understand different phenomena.
Does the association between neural states (as seen, e.g., on PET) and mental states imply that there is causal relationship? The problem here again is in the question being asked. It is more comprehensible scientifically that A is B than that A causes (separate thing) B. Paul Churchland11 refers to what he calls the Betty Crocker Fallacy. Betty endeavors to explain how a microwave oven works. She says that when you turn on the oven, the microwaves excite the molecules in the food, causing them to move faster and faster. They therefore bump into each other more, increasing the friction between molecules, and as we all know, friction causes heat. Betty Crocker still thinks heat is something other than molecular kinetic energy; something caused by but actually independent of molecular motion. Scientists, however, do not think so, for a number of very good reasons. Mental states are in fact neural network states; they are also neuronal states and molecular states at different levels of complexity.
If neural states are mental states, is it the case that a mental state is always the same neural state, or can different neural states represent the same mental state and vice versa? Let us take the example of the neurological phenomenon prosopagnosia, an inability to recognize faces. Lesion studies show that damage related to prosopagnosia includes parts of Brodmann areas 18 and 19, and sometimes 20, 21, and 37. The neural states are therefore likely to be different from case to case even though the mental phenomenon is the same. But it is not difficult to imagine how different networks may reach the same functional outcome. It does not present a problem for reductionism.
Another argument presented against psychoneural identity has been referred to as the binding problem. How does the brain integrate signals, separated in space and time, such that one experiences a unity? For example, even though pathways subserving motion are different from those for color vision, we see a white car passing by as a unity. Binding across time helps us understand a sentence or a conversation. Although the binding problem is not solved yet, there are early inroads into it. One suggestion is that temporal binding occurs because of the conjunction of specific and nonspecific neuronal loops through synchronized response episodes.12
There seem to be powerful arguments that we can reduce some mental phenomena to certain kinds of neural states. Does this mean that the phenomenological study of mental states would no longer be meaningful? The answer is clearly negative. We would still need all levels of analysis. Let us take some examples. First, consider working memory. There have been major advances in the understanding of the neurobiological basis of working memory, especially its neuroanatomical localization. Nevertheless, to understand the interactions of the working memory system and the outside world, cognitive models are extremely useful. One such model is the Soar Cognitive Architecture model,13 according to which there is a single long-term memory—a production system—that is used for both declarative and procedural knowledge. Interaction with the outside world occurs via interfaces between working memory and one or more perceptual and motor systems. Other models that have been presented for implicit and explicit memory have guided our understanding of the underlying neuronal networks.
Let us take a clinical example. Patients with obsessive-compulsive disorder (OCD) have been shown to have abnormal neural states, with increased metabolic rates in the caudate nucleus and orbital gyri.14 This abnormal neural state relates dynamically to the obsessive fear, and a patient with OCD confronted with this fear has an abnormal response.15 The underlying basis of this abnormal response is possibly at the molecular level, although this process is incompletely understood. It has been suggested that 5-HT1A serotonin receptors may be abnormal in OCD and may provide the molecular link for the abnormality.16 The question is, will we ever reach a stage when a PET or other similar scan of a patient will be sufficient to diagnose OCD without the mental state being available? Perhaps it would be necessary to have not only metabolic information, but also neurotransmitter and receptor information. The possibility is tantalizing and would be the realization of a reductionist's dream. It would, however, not take the mystery out of the choice of obsession or compulsion in OCD, nor would it minimize the social context of many of the symptoms. Far from making phenomenology redundant, it would make the analysis of the symptoms more relevant in achieving a complete understanding of the disorder. OCD provides an example of another issue of importance in the discussion on reductionism: the role of psychotherapy and cognitive-behavioral therapy in the new biology of the mind. The efficacy of behavior therapy in OCD and the normalization of cerebral blood flow abnormalities associated with the improvement17 illustrate the utility of a top-down approach and further suggest that psychology will continue to complement neurobiology even in therapeutic approaches.
One can consider many other clinical examples. Schizophrenia-like psychoses have attracted a great deal of attention, and this is a field in which brain development, acquired brain injury, drugs of abuse, electrophysiological disturbance of epilepsy, an adverse environment, and other factors intertwine to produce a lasting brain and behavior abnormality. It is here that reductionistic strategies are likely to be at their most powerful and may provide a final common pathway to schizophrenia.
The above analysis suggests that reductionism of mental processes is not only attainable, but also desirable. This does not mean that all reductionism will proceed to a molecular level. The development of neural connections is determined by an individual's genes and their products, the expression of which is influenced by social and environmental factors.18 Learning produces changes in gene expression, and drugs used to treat mental illnesses act through a similar mechanism.19 The question confronting us is whether mental processes can be reduced to the level of gene expression and the activities of neurotransmitters, receptors, and modulators. Such reduction of mental processes is the ultimate dream of some and the final dread of others. It is no longer disputed that all behavioral and mental states are associated with some neurobiological processes. However, our reductionistic understanding of mental states does not go beyond the concomitant neural states as imaged by the latest imaging techniques. The level of understanding at the molecular level is patchy, and any attempt at a framework either leads to large gaps or wild generalizations. What is encouraging is that we are developing frameworks, such as the one for OCD described above, that may one day lead to a much bolder reduction.
In science, whenever a lower-level explanation is sufficient it becomes paramount, and there is no reason why this will not happen for human behavior. Science continues to deal with different levels of explanation; for example, solidity can continue to be of scientific interest even when the reduction to lattice structure is possible. For the human mind, the higher-level analysis is even more meaningful, and will continue to remain so. We could, for example, have a classification of "depressive" disorders referring primarily to neurobiological states, with only a secondary reference to mental states. But it would not be a bottom-up explanatory strategy alone. A top-down approach will remain valid, and psychological theories will continue to hold currency. The current classifications of depressive disorders will not totally lose their usefulness. Top-down treatment strategies will also continue to play a role in therapy, and the empirical test will decide the relative merits of intervention at the molecular, cellular, network, or mental levels. This will be a psychiatry in which the debates of mental versus physical, psychological versus organic, will lose their relevance.