Chomsky makes an valid point on the limits of statistical modeling and theorizing. Despite centuries of research and work, mankind can only model the smallest of natural processes with any accuracy and is forced to resort to gross simplification when modeling large ones. It is tempting to say that modeling and theorizing are just book-keeping: a way of tying up loose ends after the “real” science is done.
Conversely, traditional experiments, while generally precise, are an inefficient way of discovering new facts when dealing with large systems with many variables. Because of this, statistical models become necessary to narrow down and pinpoint the key factors in a system for closer, experimental, study.
Frankly, rather than pick a side in the modeling vs experimenting debate , I’m inclined to agree with Millikan, who stressed the importance of both theory and observation in the advancement of science. Without observation, theory is impossible, and without theory, observation becomes useless. Arguing about whether experimental research or theory is more important is like debating the relative merits of a car’s engine vs it’s steering column; it’s true that one is flashier and more fun to work on, but without both the vehicle cannot operate.
What must not be forgotten however, is that the field of statistical “Big Data” modelling is improving all the time, indeed the rise of Google and its search engine empire in the past decade lend credence to the idea that we are only scratching the surface of what can be accomplished by this technology. Big Data analysis promises to be a powerful new tool for science, both in the collection of raw data, and in its synthesis.