What is a „mechanistic explanation“?23.03.2017
What is a ‚mechanistic explanation‘? In my experience, many people tend to misunderstand this in terms of a ‘mechanical explanation’ especially when knowing that a mechanistical explanation should establish causal claims. So, is this a revival of 18th century mechanical thinking? Surely, not. In my work on philosophical aspects of neuroeconomics, I submit strong non-reductionist claims, and I think that mechanistical explanations allow for just that, making non-reductionist causal claims. I also adopt ‘naturalism’ in the sense of causal physical closure of the world, but I do not think that this implies that, for example, ultimately neuroscience would be able to explain all human behaviour in terms of neurophysiological facts.
‘Mechanistic’ is derived from ‘mechanism’ (and not ‘mechanics’). A mechanism is a complex entity that generates certain phenomena, after receiving some input. Mechanistic explanations aim at opening the black boxes of mechanisms and at understanding how the different parts work together. Complexity means, for example, the existence of multiple levels, and mutual relationships between parts and whole. That is why another term for ‘mechanistic explanation’ is ‘constitutive explanation’: What constitutes the entity that has the causal capacity to generate a certain phenomenon?
Let me give you the example that we discussed at our recent workshop of the Witten group, herding. Economists explain herding on financial markets (resulting in ‘bubbles’, for instance) as a possible equilibrium state of a market in which private and public information is processed by rational agents. From the mechanistic perspective, this explains nothing. Indeed, other sciences that deal with herding look at many different phenomena which might have causal relevance, such as psychological contagion of emotional states across agents. In this case, a mechanistic explanation may indeed move down even to the most basic level of neurophysiological phenomena, such as ‘mirror neurons’. But in my view, that would be necessarily incomplete. Other disciplines such as econophysics would emphasize network structures, such as centrality of networks, asymmetries of communication flows etc. These are certainly not neurophysiological facts.
A major task of mechanistic explanations is to integrate these different disciplinary views on a phenomenon into one single description in which these different views identify parts of the mechanism. That is why mechanistic explanations are non-reductionist: We put together the neuroscience and the network perspective in one picture, thus identifying different levels of the mechanism which are not reducible to the other. Yet, they are causally connected in generating the phenomenon in question, the ‘herd’.
I think that mechanistical explanations are the current methodological gold standard in organizing cross-disciplinary research. They impose intellectual discipline, beginning with simple questions such as, ‘where are the boundaries of the mechanism?’, ‘what are the parts of the mechanism?’, and so on. For example, considering herding, most economists just think of herding being a phenomenon of ‘the market’, thereby falsely hypostasising the equilibrium state. That overlooks the basic fact that ‘the herd’ can only exist because there is the other side of the market, that is agents who do not follow the herd, and transact with the agents in the herd. So, the interesting question is, why are some agents part of the herding mechanism, and others not? The boundaries of the herding mechanism are not congruent with the boundaries of the market.
It seems to me, even though mechanistic explanations pay full respect to complexity, they still are amazingly simple in terms of claims and structure – conventional economics often does it just the other way round, approaching simple phenomena by means of ever complex models.