That derisive comeback, throwing cold water on what would otherwise be a profound discovery, is repeated on a constant basis in science. While they are mostly not miniature versions of Arthurian castles designed for cinematic effect, scientists use models all the time, and criticizing other scientist’s models is an essential part of the process.
Most obviously, in experimental biology disciplines like neuroscience, scientists use animal models. That means that the mouse or rat or monkey or what-have-you is meant to be a stand-in for the real system of interest (often human), and it is used to answer those pesky questions that researchers are incapable of investigating in the actual organism. By the way, this is not to dismiss the many neuroscientist and other biologists who are not primarily interested in human application of their work, but who instead study animals to understand fundamental principles of nervous systems or ecosystems or evolution. That’s science too, but their animal models are more of an exemplar than a substitute. In any case, by way of example, there are several good reasons why neuroscientists do experiments with cute furry creatures. First of all, mice are cheap and easy to raise, and their nervous systems are comparable to ours in many important ways. The most important reason, however, is that they happen to be very easy to manipulate genetically. This makes them our nearest kindred vertebrate in which scientists can exert really precise molecular control over all aspects of development. This means, for example, that neuroscientists can breed mice in which a particular subset of neural types in a particular circuit express a protein that makes them glow for easy identification. Or, they can express another protein that makes the cells activate or inactivate when illuminated with a particular color of light. More speculatively, this genetic control might soon mean the ability to allow millions of neurons to record their own activity patterns or layout a whole brain circuit diagram in a single experiment. Obviously, this kind of experimental power is extremely attractive for the experimentalist.
Scientist also use more ephemeral models as well. These can range from the sort of rule-of-thumb abstractions (conceptual models) or simple metaphors for unobservable cause and effects to explicit mathematical models with thousands of variables. Whether they admit it or not, all scientists use conceptual models to organize their thinking about observations and experiments, just as all people organize their perceptions around more or less explicit understanding of how the universe works. Researchers who do mathematical/computational models (as I have) are required by the technique to specify all the moving parts in their models and to make explicit exactly how those parts interact. In that way, quantitative modeling disallows certain kinds of hand-waving (“the super-ego keeps the id in check”) while introducing others (“This symbol I wrote down means ‘supply’”).
With scientists using the range of models, from unconscious conceptual schemata to differential equations to living animals, it’s no wonder criticizing models makes up a large part of scientific discourse. So what makes a good model? Trick question. There is no such thing as a good model, at least not outside of the context of a particular scientific question. It all depends. Want to study executive function and decision-making? Cell cultures in a petri dish may not be the right choice. On the other hand, if you are studying the basic molecular functions that all living cells use, that petri dish might be just the thing. Of course this example makes the point trivially obvious, but scientists who use different model systems often speak at cross-purposes. The choice of model systems always involves trade-offs, some of which are carefully thought out and some of which are simply a matter of intellectual disposition (to-MAY-to, to-MAH-to). For example, in studying processes underlying human diseases, some people like the precise control of a cell culture system and others like the immediacy of clinical drug trials. The central tension is between the simple models that can be more tractable and more complex models that soon become as difficult to understand as the systems they are trying to explicate.
Just as an aside, some animal rights activists have claimed that cell-cultures and computational models can replace many animal experiments. While I’m absolutely in favor of reducing the suffering we cause animals as part of biomedical research, as someone who has used all three of these techniques, I can say this about people who claim animal research can be supplanted by the other techniques: they know nothing about biology and less about computers. Unless they are ethical absolutists on this question, animal rights activists would be sensible to support techniques (like those proposed by the BRAIN Initiative) that would allow collection of huge amounts of data from individual animals in one experiment.
So you have to be content with the fact that there is no sweet spot. Reductionists will always fault systems scientists for not understanding the underlying mechanisms, and systems scientists will always disparage reductionists for not explaining the emergent phenomena. It’s important to realize that despite the bickering tone this discussion may sometimes take, it is essential to understanding how findings from one model apply (or don’t apply) to others.
Photo Credit: Classic scene from the classic movie Monty Python’s Holy Grail. From a pre-CGI era, when special effects meant building miniature replicas of big things and getting the camera up real close.
Update 12/01/2013: Fixed a couple of minor typos/redundancies.