'Reductionism' is a tenet of modern neuroscience. Only its version and explicitness vary among subdisciplines and their practitioners. Neuroscientists attempt to explain mental faculties by brain faculties, hence mental phenomena by biology. In that they are different, for example, from orthodox "behaviourists. Clearly, biology at large has accomplished some of its most impressive triumphs so far by adhering to the radical reductionist approach, epitomized in state-of-the-art molecular and cellular biology, genetic engineering ("neurogenetics), and the Genome project. Whether ardent reductionism also fits all the needs of memory research is, however, still a debatable question.
In science and the philosophy of science, 'reduction' (reducere, Latin for 'to bring back') is employed in different connotations (Nagel 1979; Mayr 1982; Dudai 1989). In its most common use, it refers to the mere process of analysing a complex phenomenon by dissecting it into elementary components. This is 'constitutive reductionism' (definition 1, 'description'). In the context of our discussion, it means that one attempts to identify brain, neuronal, or molecular correlates ("criterion) of learning and memory. This type of reductionist approach is accepted by all neurobiolo-gists and practised by the majority of them; ethologists and behavioural psychologists can still excuse themselves from preaching the reductionistic "zeitgeist. In the course of practising constitutive reductionism, neu-robiologists take 'reductive steps'. These are shifts in the level of analysis from the level of a "system as a whole to the level of its components. For example, a shift in the analysis from that of molar electrical activity in "cortex to that of individual cortical neurons, or from single neurons to individual molecules in the "synaptic membrane, is a reductive step. In addition, constitutive reductionism almost always involves 'simplifying steps'. These are procedures taken to facilitate experimental analysis, without altering intentionally the level of analysis. For example, proceeding in the analysis of single neuron activity from in situ to a brain slice (e.g. "hippocampus), or removing part of the tissue and hence decreasing the number of cells in a ganglion (e.g. "Aplysia), while still maintaining the cellular level of analysis, is a 'simplifying step'. This "methodology is best epitomized by Johnson (1751), who noticed that 'Divide and conquer is a principle equally just in science as in policy'. A 'simplifying step' may also mean switching to a simpler organism or circuit that display the phenomenon, process, or mechanism in question ("model, "simple system).
More rigorous than constitutive reductionism is 'explanatory reductionism' (definition 1, 'explanation'). It assumes that, ultimately, the knowledge of the components will explain properties of the system as a whole. This means, in our case, that having once understood the properties of neurons and molecules, one should be able to show how these properties are necessary and sufficient to explain learning and memory. Most neurobiologists practise constitutive reduc-tionism in the hope of achieving explanatory reductionism. Others doubt whether in the neurosciences, satisfactory explanatory reductionism is always feasible.
Most demanding is ' theory reductionism', i.e. reducing a theory, including all its concepts and laws, into another, more inclusive or basic theory (definition 2; e.g. Nagel 1979). This would mean that, having a biological theory (such as a future theory of brain function), one would be able to reduce it without residue into a physical theory. Profound doubts are often expressed whether this is appropriate and feasible (for a modern "classic, see Fodor 1974). 'Theory reductionism' requires that 'bridge laws', or 'correspondence rules', be established to enable shifts from the terminology of one theory to the other. A limited concept of 'correspondence rules' is also useful in the less stringent, descriptive, or explanatory reductionist approaches, because it focuses attention on the need to formulate systematic relationships between findings at one level to those at another. For a selection of the literature on explanatory and theory reduction in the neuroscience, see Putnam (1973), Searle (1990), Schaffner (1993), Churchland and Churchland (1998), Crick and Koch (1998), and Fodor (1998). For issues related to other facets of reductionism in biology, which reflect on neurobiology and especially on "development and neurogenetics, see Hull (1981), Sober (1994), and Kirschner et al. (2000). And for a bit more on the philosophy involved, see Brentano (1874) and Kim and Sosa (1999).
Within the context of the present discussion, two issues deserve special attention. The first is what do we expect the relationships to be between properties, processes, and events at a lower level to those at a higher level, and vice versa. It is safe to assume that even conservative psychologists will agree that mental events somehow relate to physical events. But what does the relationship mean? For example, the two may correspond to each other property by property, and ultimately unique mental states would correspond one by one to unique brain states. If this is true, then having identified a certain physicochemical brain state in sufficient detail, we will be in a position to deduce from a set of physiological data what the "subject "recalls. This argument applies to multiple levels of brain function: knowing the molecular details will be expected to tell us precisely what a cell encodes, monitoring the electrical activity of a "cell assembly will be expected to tell us exactly what the assembly represents, etc.1 An alterna tive possibility is that a given mental/behavioural state is encoded by different molecular and physiological states. This would mean that indeed, a particular mental/behavioural state corresponds to some physical events in the brain, but the former does not necessarily correspond to a unique configuration of the latter. This view might induce chagrin in many practising neurobi-ologists who attempt to read into defined codes, representations, and computations. The problem is not merely philosophical. When we devote our career to determining with great pains the activity of "signal-transduction cascades within neurons, or of circuits within brain regions, do we expect to be able, at the end of the day, to conclude from the lower-level data what the overall higher-level meaning is, and at what level of accuracy? Do we have an 'uncertainty principle' operating at the level of neuronal representations? Also, is knowing more lower-level details always means knowing more about the higher-level functional state of the system (Alberts and Miake-Lye 1992; Barkai and Leibler 1997; Sanes and Lichtman 1999; Kirschner et al. 2000)?
The second issue, related to the first one, is how far should one attempt to reduce a system without losing the characteristic properties that had provided the incentive for the research programme in the first place. In other words, how much of the system properties are 'emergent', i.e. appear only at a higher level of organization or function, because of some interactive or integrative properties (Pepper 1926; Meehl and Sellars 1956)? Why not aim at reducing the description of brains to the language of the elementary particles of matter? Most reasonable people will ridicule such a proposal, yet there is nothing in it that contradicts ardent reduc-tionism. If not elementary particles, why not atoms? Or molecular motifs? When do reductionistic aspirations cease to amuse and become a serious scientific goal?2 It seems that in the case of memory research, the clue lies in the definition of "memory. Once we agree that memory is retention over time of "internal representations, we should look for the level at which behav-iourally meaningful representations are encoded. Lower levels (lower than cellular) will provide us with valuable information on mechanisms that subserve memory, but are unlikely per se to identify what a specific memory is. Memory research, as in any other branch of biological research, thus requires pragmatic, 'focused reductionism' (Dudai 1992). Otherwise, we may find ourselves in a situation so enchantingly illustrated by Geertz (1983): "There is an Indian story, about an Englishman who, having been told that the world rested on a platform which rested on the back of an elephant which rested in turn on the back of a turtle, asked... what did the turtle rest on? 'Another turtle'. And that turtle? Ah, Sahib, after that it's turtles all the way down.'
Selected associations: Criterion, Level, Method, Paradigm, Zeitgeist
1This argumentation is intentionally restricted to the pragmatic point of view of the experimenter. What is argued here is only whether there is one-to-one, or one-to-many, or many-to-many correspondence between the particular neuronal state and the particular behavioural or mental state. Nothing is claimed about the quality or type identity of the corresponding states, e.g. whether some events are of a 'physical type' and others of a different, 'mental type' (e.g. Block and Fodor 1972). Further, the correspondence may reflect correlation, super-venience, or causality ('criterion).
2This is actually still another manifestation of the classic 'Sorites paradox', which is presented in *insight.
Was this article helpful?