Some Virtues of Modeling With Both Hands

Webb distinguishes two endeavors she calls animal modeling and animat modeling and advocates for the former. I share her preference and point to additional virtues of modeling actual biological mechanisms (animal modeling). As Webb argues, animat modeling should be regarded as modeling of specific, but madeup, biological mechanisms. I contend that modeling made-up mechanisms in situations in which we have some knowledge of the actual mechanisms involved is modeling with one hand—the good one—tied behind one’s back. The hand that is used in animat modeling is constructing and evaluating models by whether they behave in the right way—do they exhibit the particular phenomenon one is trying to understand? The good hand that is disavowed seeks to use evidence about the mechanism employed in real living systems both for inspiration in designing the model and for evaluating the model. Denying oneself use of one’s good hand both limits one’s access to valuable evidence for evaluating a model and denies oneself access to a potent discovery strategy. Webb draws attention to one reason to employ the good hand—if models are to be relevant to biology (and not just characterize hypothetical mechanisms), then the component parts and operations specified in the model must in some way map onto those in actual biological organisms. Especially if one accepts the possibility of multiple realizations, then if one only uses behavior to evaluate the model one may well have described an alternative realization than that found in real organisms. To determine that one has modeled the actual realization, it is necessary to compare the proposed mechanism with the actual mechanism—does it consist of the same parts, operations, and organization. I have argued elsewhere (Bechtel & Mundale, 1999) that multiple realization is highly suspect if we demand that alternative realizations be compared at the same fine-grain of behavioral detail as employed in claiming that the realizations are actually different. If there is only one realization once we are rigorous in evaluating behavioral equivalence, then models of made-up mechanisms that meet the behavioral test will also correspond to models of actual systems. In fact, however, researchers typically work with a much coarser-grained analysis of the phenomenon to be explained, and in such circumstances the plausibility of mechanisms with very different design generating the same phenomena increases significantly. If the goal is nonetheless to produce models that correspond to actual mechanisms, the sooner we release the good hand and start to use it, the sooner we will arrive at realistic models. Many animat modelers, however, reject the assumption that the goal is to produce models that correspond to specific actual mechanisms. They advocate instead the importance of idealized or general models—models that represent the breadth of biological possibility not the specific realizations studied in actual biology. Webb correctly notes that in any modeling of a mechanism, specific assumptions are made about the mechanism. She then questions what makes a made-up mechanism more general than an actual mechanism. This is an important issue and raises a broader question about the nature of generalization in the life sciences. In philosophy, generalization is usually assumed to take the form of universally quantified statements