Kandel on Psychiatry and Neuroscience

In 1998, Eric Kandel published an article entitled "A New Intellectual Framework for Psychiatry." (citation: Am J Psychiatry. 1998 Apr;155(4):457-69.) Kandel was trained in psychiatry, has authored one of the definitive texts of neuroscience, and he won a Nobel Prize for physiology and medicine in 2000—certainly bona fides that make his speculations about a framework for psychiatry worth reading. What framework did Kandel propose? Has it had any effect on psychiatry today?

In his introduction Kandel presents a very short history of the relationship between psychiatry and biology. Briefly: Freud was a neurologist, but the primitive state of biology in his time didn't allow for meaningful connections with psychiatry. Psychoanalysis, which flourished in post-WW2 America, rejected any biological connection, for a period even avoiding acceptance of psychopharmacologic advances in favor of a humanistic, social, individual grounding, until roughly 1980, when improvement in diagnostic precision, continued advances in pharmacology, and parallel developments in neuroscience science, all lead to increased calls for realignment of psychiatry with brain science.

Kandel describes his "common framework for psychiatry and the neural sciences," which he sets out as five principles (these are direct quotations):

Principles 2 and 4 describes results of neural science, which Kandel has had a direct hand in, but they don't have much connection to mental illness or psychiatry. Principle 3 makes two points: that genes are vitally important in determining brain structure and function, and that not all the functions (or dysfunctions) are determined directly by the genes, because of the environmental influences. This is as true for mental illness as for any other human behavior. (More about this below.)

Principle 1 is a statement of reductionism: all mental phenomena are (merely) the action of networks of nerve cells. Kandel presents this principle as "almost a truism" and "the basic assumption underlying neural science." On the surface, this principle is obvious to anyone working in context of scientific materialism: unless you believe that mental phenomena are associated with some vital force or some other anti-materialist factor, then this principle is apodictic. However, unless you are an omniscient being, it is nearly useless.

Reductionism has a simple meaning ubiquitous in modern science—that reduction is possible—and a more nuanced meaning the truth of which is contingent. A complex phenomenon, such as human behavior, appears to be reducible to the operation of neural networks, cells, molecules, and ultimately to quantum mechanical wave functions. Reductionism has had profound implications for how science is done—it replaced vital forces or mystical explanations with the mechanical explanations of some phenomenon through the interaction of simpler elements: reducing the biology of "life" to molecular biological implementations in genes, proteins, and other molecules. However, while the reductions replaced mysteries with mechanisms in a philosophically satisfying way, they left as actual scientific work the problem of making meaningful reductions. Thus, we're willing to believe that learning and memory can be implemented in brains, but we're left having no idea how this is actually done. Nor is it clear what the relevant target level of reduction should be. Learning could be, for example, be best understood at the level of brain regions, neural networks, individual neurons, synapes, gene expression, genes, or proteins.

One can adopt a purely materialistic approach to all science, and yet not believe that any good would come from trying to reduce all phenomena to physics. "Homicide" is a legal concept. It can be realized in an essentially unbounded number of ways, all of them composed entirely of atoms. But the level of atoms is useless for making any headway in understanding homicide as a phenomenon; the complexity would overwhelm, not enlighten.

Let's take a system simpler than the brain: a computer running a spreadsheet program. Assume that there's a mistake in one of the formulas in the spreadsheet. Since the spreadsheet runs on a computer, and the computer is a purely physical device, it could be claimed that the error in the formula be found, represented, and reasoned about by taking x-rays of the computer circuit boards, and measuring voltage traces along each wire path. However, it would be extraordinarily difficult to make any sense of such a description. Conversely, if the computer had a hardware malfunction, that error would have manifestations in how the spreadsheet program functioned. However, there is no reason to expect any simple description of the hardware error at the level of the spreadsheet formula.

Reductionism is useful when it advances scientific understanding; this depends on the levels from which one starts and the level towards which one is trying to reduce. A good rule of thumb is that the descriptions in the reduction should be reasonably compact. In the computer spreadsheet case, we are fortunate because there are explicit statements of how to implement each level at the next lower level—it's designed that way. The spreadsheet formula is written in some graphical interface for the spreadsheet, the spreadsheet is written in some high-level computer language that is compiled into (low-level) assembly language, assembly language is converted into machine code for that kind of computer, the computer hardware implements a well-described virtual machine, and, finally, the computer chips implement specify rules of a Boolean logic (and more) using certain time-varying voltages. You could try to understand the signal on a voltage probe attached to a particular memory chip pin as it relates to the formula in the spreadsheet. For our purposes here, the hardware world is completely deterministic, and could be written out in excruciating detail (for any given set of inputs to the computer). But no one would ever do so (except, perhaps, in a hardware-design simulator). What's more, very small changes in implementation choices would have large effects in the voltage signals without any changes in the spreadsheet. Said differently: the same spreadsheet can be implemented on rather different hardware configurations.

Kandel spends much of time talking about neural genes and gene expression. No psychiatrist seeks to understand mental function or dysfunction at the level of individual genes, unless there is an easily understood connection between a distubance at the level of the gene, and the manifest behavior of the organism. Consider Huntington's disease. We know what the gene disturbance is, and we can test for it. However, the connection between the gene disturbance and the manifest behavior of a patient with Huntington's disease is still a mystery, and knowing about the genetic error provides no help in guiding treatment, beyond genetic counseling for parents.

Throughout his argument Kandel confuses the possibility of reduction—an important philosophical and methodological advance—with its utility, which is decidedly contingent and has to be demonstrated, not asserted, for each domain in which is thought to hold.

Brain genes code mostly for neurodevelopmental programs, which are sensitive to the environment in which the brain develops. This is, roughly, ontogeny and learning. As a result, human brains are similar, not identical. Ten percent of the population is left-handed and have varying patterns of hemispheric lateralization of brain function; they are often screened out of brain imaging studies, because their varied lateralization would add too much noise to the data. Kandel comes very close to admitting this directly: "Since each of us is brought up in a somewhat different environment, exposed to different combinations of stimuli, and we develop motor skills in different ways, each brain is modified in unique ways. This distinctive modification of brain architecture, along with a unique genetic makeup, constitutes the biological basis for individuality."

To be fair, Kandel does admit that some personal and social phenomena might not be best understood as neural processes: "I hasten to add that formulating a relationship between social processes (or even psychological processes) and biological functions might not necessarily prove to be optimally insightful in elucidating social dynamics. For many aspects of group or individual behavior, a biological analysis might not prove to be the optimal level or even an informative level of analysis, much as subatomic resolution is often not the optimal level for the analysis of biological problems. Nevertheless, it is important to appreciate that there are critical biological underpinnings to all social actions." It's not at all clear how the last sentence saves his argument: it's either trivially true that social actions are implemented in biological systems, or it's left as a grossly incomplete scientific enterprise to find the meaningful reductions.

(In his reply written in response letters published about his article, Kandel admits, "One has to acknowledge that we are still far from establishing a biological foundation of psychoanalysis. In fact, we do not as yet have a satisfactory biological understanding of any complex mental processes. Therefore, it is quite possible that a convergence of biology and psychiatry is still a bit premature." If so, then why write that framework article? Who would that framework actually help?)

Kandel strives hard to make a connection between brain phenomena and psychoanalytic practice. "As the resolution of brain imaging increases, it should eventually permit quantitative evaluation of the outcome of psychotherapy." This is pure grandiosity. We don't know whether brain imaging will have anything useful to say about psychotherapy or not, until we have the results in hand. There's no reason to expect it a priori.

Psychiatry has often categorized mental disorders into two types: organic or functional. Organic disorders are those associated with clearly identifiable brain dysfunction. Alzheimer's disease is an example. Functional disorders are those with the link with brain dysfunction is not clear. Neurosis is a classic example. Kandel seems to think that his framework makes this terminology obsolete. He writes, "this distinction, now clearly outdated, is no longer tenable. There can be no changes in behavior that are not reflected in the nervous system and no persistent changes in the nervous system that are not reflected in structural changes on some level of resolution."

Kandel thinks that this organic vs. functional distinction is somehow anti-reductionistic. He's wrong. It is clearly the case that some disorders have meaningfully and compactly identifiable brain dysfunction, and some don't. It's not a claim that the disorders have no neural implementation—that's impossible, at least for those who practice scientific materialism. However, given the relatively primitive state of neural science we can't figure out how to implement the vast majority of our current list of mental disorders in large networks of squishy neurons, and it may turn out that the simplest, most useful description is not at the neural level, even when we get it all sorted out. While a few disorders might change from functional to organic, many won't, in the same way that some computer errors are best understood as hardware malfunctions, and others are best understood as software errors, even though no one would ever claim that computers are anything other physical objects.

It is not at all clear that the most useful reductions will be from psychiatry phenomena directly to elements of gene function or synaptic plasticity. I’ll bet such reductions wouldn’t be found. If I'm correct, then much of Kandel's framework is irrelevant. What we do need are the levels of abstraction for brain and mind behaviors that are each close enough to the neighboring levels to make reduction and implementation a tractable endeavor.

If Kandel had written an article saying that understanding human behavior (and behavior change) will require building a complex, multi-leveled, theoretical structure drawing on physical, biological, psychological, and social science, he would have been a good deal closer to the truth, but it would have left the neuroscience boosterism at the door.

Author: Steven Bagley

Date: 2013-09-28 Sat