|
The very traditional view of science as the formulation and testing of universal laws applies fairly well to the study of electro-magnetism, for example. The properties of electro-magnetism, for example, it is reasonably safe to assume that the properties of electro-magnetism are the same today, as they were yesterday, and the properties are the same whether in New York, or in Miami. But this view of science is completely inadequate outside of these, so-called, hard sciences. For example, philosophers of science have tried very unsuccessfully, to rope in evolutionary biology into this view of science. The fundamental problem is that the object of study changes, in space, and in time. If there is a universal law, it is highly contingent--the number of conditions in the if-clause, is enormous. And the most famous "law" in evolutionary biology, the Theory of Natural Selection, is itself entirely empty of physical content: its an algorithm, that can, or may not be, instantiated by a physical system, and because there are many ways in which the algorithm can be instantiated, getting the nitty and the gritty to actually formulate predictions requires the specification of an enormous number of background facts. Instead, evolutionary biology is a model-based science. Scientists evaluate models, models that only partially explain limited portions of the world, and these models usually explicate a process. We have families of models, describing Darwinian processes under different regimes. I do expect that some experiments conducted in psychology are non-repeatable, especially if they become famous. The question isn't whether it was science, but whether the target phenomena is important enough, and stable enough, to warrant our interest. In anthropology, Clifford Geertz famously argued that if there are laws of culture, they must be entirely uninteresting ones. His approach, instead, was rich description in place of theorization. His arguments are compelling, however, he too had this notion of science as universal law. I would argue that a model-based notion is much more productive, and much more accommodating to the problem domain. Let me go one step further: what one should focus on instead, is process by which phenomena observed are instantiated. Similar processes can result in quite different outcomes, and similar outcomes can be the result of very different processes. I work in the field of decision neuroscience. We work, largely, by linking human behavior and neuroimaging to fitted parameters in a family of reinforcement learning models (see: Reinforcement Learning: An Introduction, Sutton and Barto). Decision making processes are largely invariant, even when knowledge, and domain-representations, might change with changes in popular culture: that is, changes in behavior can be accounted for by changing the priors, or input information, into the model. |