(This post is a spin-off from this essay by Fredrik Liljeros and me.)
The use of computers and numerical techniques (except regression analysis) has always been outside of the mainstream social and behavioral science. At the same time, computational social science was not born later than computational physics or chemistry (although it is a bit hard to pinpoint the first article in either of the fields, it is clear the developments were parallel). It can be hard today to motivate a simulation approach. It was, of course, even harder in the 1960s when few people actually used a computer. Gilbert and Hammel (1966) felt compelled to explain that
… we regard a computer as a clerk with several virtues but one which is no better than its instructions. To appreciate the way in which this clerk operates, one should realize that a computer is, in principle, nothing more than a desk calculator and a note pad. In using a desk calculator, an operator selects items from the note pad or other storage location (such as his own memory) and transfers them into the calculator. He then performs certain manipulations of the data so entered by punching particular buttons in a certain sequence, reads the results, and transfers these back to the note pad. A computer combines the desk calculator and the note pad (storage) in one system so that storage, retrieval, and manipulation of items are mechanized. The program for the computer mechanizes the operator in directing the computer which items to select from storage, which operations to perform on them, and which results to place back in storage.
When explaining what computers actually were was out of the way, the real challenge began—to explain the methodology of simulation. So how did the pioneers of computational social science motivate their use of computer simulation?
As nothing new
With or without quotations from Francis Bacon’s Of Simulation and Dissimulation, many authors make the point that simulation has been a methodology since before digital computers.
Simulation may be traced back to the beginning of time—be it the make-believe world of the child at play, or the adult make-believe world of the stage.
… writes Harry Harman in a 1961 white paper for the military subsidiary System Development Corporation. Herbert Simon also points to pre-digital computer simulation in his Sciences of the Artificial (1969). In particular, he mentions the MONIAC—an analog simulator of Keynesian economics.
To facilitate causal (or computational) reasoning
Another idea—often championed by the prolific behavioral scientist couple John and Jeanne Gullahorn (see e.g., this paper)—is that breaking down the code in appropriate blocks and linking them by mechanisms corresponding to the real system can help researchers to reason about cause and effect. This argument was formalized by the computational thinking movement (maybe not a movement, but anyway) of the 1980s.
As a way to do experiments otherwise impossible
This was perhaps the most common motivation and one that possibly fallen slightly out of fashion. Quoting the psychologist Kenneth Colby
Before the computer program we had no satisfactory approach to huge, complex, ill-defined systems difficult to grapple with, not only because of their multivariate size but also because of a property of elusiveness which in psychology is mainly a function of vagueness in that the limits of inclusiveness of conventional terms are unclear.
The “complex, ill-defined system” Colby has in mind is nothing other than “a neurotic.” For another example, Ithiel de Sola Pool mentions this point in his essay “The Kaiser, the Tsar and the Computer” about simulating international political conflicts for the benefit of decision-makers.
As proof-of-concept models
Another motivation was that computer simulations can validate verbal reasoning. Nowadays, this usually goes under the name of proof-of-concept modeling (at least in the biological sciences). Personally, I think this is where social simulation really has a future. Edward Feigenbaum writes in his essay Computer simulations of human behavior (1963)
One of the advantages of computer simulation is this one, of guaranteeing sufficiency and completeness. The computer simulation model will not operate if you forget anything. If you fail to take into account some necessary mechanism that, in the verbal description of a theory, you might readily pass over, the computer simulation will not run. Thus, one is forced to focus attention on all of the mechanisms—those that are not well understood as well as those that are well understood—and all of the subtle effects and interactions which are implicit in one’s model. In a sense, this business of accounting for all of the information processing is a very powerful mental discipline.
For forecasting and scenario testing
Following the excitement of the Club of Rome’s Limits to Growth—that relies on computer simulations to argue that the Earth cannot sustain an exponential population growth—there seems to have been several attempts to start large-scale, real-time forecasting of social processes. The journal simulation, mentions several attempts to a “world simulation” (see here and here). Recently, we could hear echoes of these attempts from FuturICT’s proposed “Living Earth simulator.” Scenario testing was also a motivation for simulation in management science who tried to simultaneously investigate the external and internal processes of a company