Levels and Explanations Jon Opie (jon.opie@adelaide.edu.au) Discipline of Philosophy School of Humanities University of Adelaide South Australia 5005 Abstract It is a mainstay of the philosophy of science that reduction is a relationship between theories pitched at different levels of nature. But the relevant sense of “level” is notoriously difficult to pin down. A promising recent analysis links the notion of level to the compositional relations associated with mechanistic explanation. Such relations do not order objects by scale or physical type; one and the same kind of entity can occur at several levels in a single mechanism. I will sketch this approach to levels and consider some of its implications for our understanding of the relationship between cognitive psychology and neuroscience. Keywords: reduction; explanation; levels of nature; mechanism; mechanistic explanation Introduction It is a mainstay of the philosophy of science that reduction is a relationship between theories pitched at different levels of nature. So, for example, thermodynamics, which deals with certain bulk properties of matter, is said to be reducible to quantum mechanics, a theory operating at the level of the molecular constituents of matter. Likewise, the reduction of cognitive psychology to neuroscience would involve replacing a behavioural-level theory with a theory at the neuronal level. But the relevant sense of “level” is notoriously difficult to pin down, and attempts to offer an analysis that is consistent with scientific practice are plagued with difficulties. A promising new approach (Bechtel 2008; Craver 2007) links the concept of level to the compositional relations associated with mechanistic explanation. This approach has some unorthodox consequences, including that levels are local rather than global (in a sense to be explained), and that one and the same kind of system or entity can occur at several levels in a mechanistic hierarchy. According to the received view, by contrast, every material system belongs to one and only one level in a monolithic, global hierarchy. In what follows I will unpack these two competing approaches to levels, offer some grounds for preferring the mechanistic approach, and sketch a few implications for our understanding of the relationship between cognitive psychology and neuroscience. DN Explanation Once upon a time, to explain a phenomenon was to derive it from statements describing laws and background conditions. This view, most clearly expounded in the work of Hempel (1966), is known as the Deductive Nomological (DN) model because it treats explanation as the deduction of explananda from laws (Greek: nomoi) and boundary conditions. For example, to explain why the pressure on the walls of a gas- filled piston roughly doubles when its volume is halved, we invoke Boyle’s law. This law states that the product of pressure and volume for an ideal gas at a fixed temperature is constant (PV = c, where P is pressure, V is volume, and c is a constant determined by the quantity and temperature of the gas). 1 We can mathematically derive the measured change in pressure using this law, with the change in volume and the other criteria operating as background or boundary conditions. 2 What distinguishes one scientific discipline from another on the DN model (insofar as disciplines are regarded as primarily engaged in the business of explaining some range of phenomena), is the theoretical vocabulary, ontology, and proprietary laws or theories applicable to their respective domains. The various gas laws thus serve to pick out the discipline of thermodynamics, which (among other things) deals with gases conceived as fluids with a characteristic set of macroscopic properties. Kinetic theory, by contrast, deals with the behaviour of the microscopic constituents of matter, treated either as classical particles governed by the Newtonian laws of motion, or quantum systems governed by quantum mechanics. DN Reduction A question that naturally arises is how disciplines so conceived relate to one another. One possibility is that they are unrelated; this seems to be the right thing to say about cosmology and economics. But many disciplines operate in overlapping or ontologically related domains, and appear to be explanatorily connected. In particular, attempting to explain laws, as opposed to phenomena, usually requires one to cross discipline boundaries. To explain Boyle’s gas law we have to invoke mechanical and statistical principles which do not belong to classical thermodynamics. If we can 1 Additional criteria are implicit in the reference to an “ideal gas”. An ideal gas is a model system that is exactly described by the equation PV = nRT, where n is the amount of gas (measured in moles), R is the gas constant, and T is its temperature. A real gas approximates the behaviour of an ideal gas, but only under conditions of high temperature and low density. 2 PV = c; V 1 = ½V 0  P 1 = c/V 1 = 2c/V 0 = 2P 0