Post 5 of 5 from Mazviita Chirimuuta on The Brain Abstracted (Open Access: MIT Press).
The last of this series of posts summarises the conclusions regarding philosophy of science more generally that emerge from this study of simplification in neuroscience. The question of realism may have already occurred to you. Once we buy into the idea that no theory or model in neuroscience can represent the complexity of a neural system in its entirety we are left wondering if it follows that all such theories and models are false or at least in some way epistemically deficient. This is a worry that is conditioned by a common scientific realism which asserts that the best theories of a mature science accurately represent entities and relationships just as they are in nature, independently of us. This sets up a strong notion of veridicality as a condition of adequacy for scientific representations, against which all representations that contain abstractions and idealisations are bound to fall short. However, we should be aware that the realist’s standards are not themselves realistic: once it is acknowledged how widespread and indispensable these simplifications are across the natural sciences, not only neuroscience, the modus ponens becomes a modus tollens. The failure of traditional realism to make sense of how science can be successful despite its reliance on strategies that fail to represent the true complexity of things in nature is reason to explore alternatives.
The alternative I propose in Chapter 2 (Footholds) is called haptic realism. It is loosely Kantian in inspiration, asserting that scientific knowledge is the product of an interaction between researchers and the items they investigate. The active contribution of the scientist cannot be discounted, meaning that the traditional realist’s standard, that science at its best should represent things as they are in themselves, independently of the way that humans have chosen to interact with them, is an unobtainable ideal. Haptics is active touch and I propose this as metaphor for the way that scientific research, and hence the resulting knowledge, involves deliberate manipulation and shaping of the objects investigated, in order that the practical goals of research, such as medicine and technology, can be reached.
Once it is recognised that science is not a disinterested pursuit of knowledge for knowledge’s sake, but an activity directed at producing knowledge that is utilisable, it is easier to see how it is possible that some of the most prized products of scientific research can be full of idealisations that depart deliberately from the truth of observable empirical facts. The point here is that idealisation is a method for taming complexity, and complexity is more often than not an obstacle to practical efficacy. Science conceptually and materially shapes its objects with a view to isolating the causal dependencies most of relevant to human ambitions of manipulation and control. I have spoken here of science in general. I think these points apply not only to neuroscience but the gamut of biological sciences, these all being connected in some way to biomedical goals even when classified as basic not applied research. Most of the physical sciences can be understood as having connections with application, however indirect, though there may be exceptions such as cosmology.
In Chapter 8 (Prediction, Comprehension, and the Limits of Science) I pick up the Kantian line of thought on there being limits to the epistemic reach of science — specifically, regarding the goal of understanding the natural world. Based on work in the historiography of science, nicely summarised by Peter Dear in his 2005 essay, “What is the history of science the history of?”, I argue that the objects of scientific understanding – the things that modern science gives us understanding of – are not, strictly speaking, natural objects; rather, they are artefacts to some degree or other constructed from the products of the natural world in such a way as to better facilitate modelling and theory building. These artefacts may be concrete material ones, such as deliberately bred model organisms that render certain causal relationships more transparent than their wild-type cousins, or they may be abstract conceptual constructions that stand as proxies for the systems in nature that are in themselves opaque. As surprising as this may sound, this historical tradition tells us that celestial mechanics is in some way a theory of machines, for machine building was a necessary precondition for the development of the discipline of mechanics, and the conceptual construction of all objects in the heavens as operating in a machine-like way remains the pre-condition for this form of scientific understanding. The point is that science did not leave anthropomorphism behind when the clockwork universe replaced the teleological one of the Aristotelians. It is just that instead of analogising the forces of nature to the psychological drives of human beings, the analogies came to be drawn instead from the inventions made by human beings.
I ask in this chapter what happens when human inventions begin to be so complex that their makers can no longer be said to have a good understanding of their workings, and yet they are still used as a means towards scientific understanding. I argue that this is the situation neuroscientists find themselves in with respect to the many layered artificial neural networks (ANNs) now used as proxies for actual brain networks. I describe how the explanatory goals of neuroscience are altering in response to this situation. The brain abstracted (in an ANN) is now the strange, silicon frontier of neuroscientific understanding. There is more that could be said but I will leave it there.
Thank you for reading these posts, and special thanks to the commentators for reading and engaging with the book.