Dylan Marks

SSP205, Spring 2005

Paper on Axelrod

A Review of Axelrod’s “Advancing the Art of Simulation in the Social Sciences”

Axelrod is most renowned to me as the inventor of tit-for-tat, a game strategy that is most useful in cooperation environments and one which is highly successful in tournaments involving the prisoner’s dilemma. He has done much in the work of simulations and models and his paper represents his desire to advance the simulation field to a higher level of maturity. Axelrod’s paper provides an overview of the current state of simulation as a tool of science. His work describes the benefit of using simulation as a scientific tool as well as the difficulty of sharing simulation work among the scientific community. The paper also provides us with an outline of how to carry out simulation work and of comparing the results of other simulation work.

Axelrod’s Simulation Uses

Axelrod provides the reader with much evidence claiming that simulation is indeed a growing tool to the social scientist with its many references in various journals. However, these references are spread across all journals and disciplines, meaning simulation has no real home and there is no mature simulation field that scientist can easily refer to.

Being the young field that it is Axelrod defines what simulation is in the social sciences and what it can do to help scientist. The preliminary definition of simulation he gives is “driving a model of a system with suitable inputs and observing the corresponding outputs.” This seems like a reasonable definition since the model should drive itself and come to an outcome based on what you give it and when it is initialized, but is a model based on agent simulations. For more complicated simulations such as flight simulators, the user seems to help drive the system in real-time and there are no real outputs except for updates as they are driving.

Axelrod suggests that simulation has more purposes of which prediction, proof, and discovery are most important for science. Prediction being most closely associated with the above definition in providing the model and inputs and letting it predict the outputs.

A simulation can also provide an existence to a proof according to Axelrod’s paper. He cites that Conway’s “Game of Life” simulation proves that complex behavior can result from simple rules. I agree with Conway’s simulation as providing evidence for this proof but believe that most proofs done through simulation are those that find relationships using simple rules to provide complex behavior, such as the Schelling 1974 model. For other formal proofs such as those in mathematics, I believe it would be more difficult to run a simulation that gives such a proof given the premises, though I know of some programs that have premises and operations built in and they can check to see if a proof is correct or not with logical checking.

One of the most innovative uses of simulation in the sciences is for pure discovery. Science has thrived on observation for discovering new relationships and facts about the world. In simulation, the model is built from relevant components of the real world such as rules and relevant agents. The end result is a discovery taken from the history and behavior of the simulation, thus discovery can be made from the premises givenn.

Simulation as Science

Axelrod claims that simulation is actually a third way of doing science fusing deduction and induction. Of course with induction we make observations to come to conclusions about what is occurring, while in deduction we start off with premises and rules and are able to come to conclusions given these premises. Simulation combines the two by using the given premises and rules, and in the end we do empirical analysis on what has happened in our simulation world.

One of his most powerful claims when talking about simulation as a tool is that “simulation can be used as an aid to intuition.” In order to prove a point or further our curiosities, traditional thought experiments have been performed and debated. With simulation the assumptions are built into the model and the experimenter can view rather quickly whether his intuitions are correct or not given those assumptions. These results are what Axelrod comes to call the emergent properties of the simulation.

Why not just deduce these intuitions without the need for such a simulation? Axelrod makes a good point regarding this. Scholars usually use the rational choice model in game theory and not because it offers realistic advice to the decision maker, but because it does allow for assumption. For more realistic results and decision making, adaptive techniques are needed which is impossible for deduction. Simulation must be used. I agree with Axelrod here because in a highly dynamic system with non-linear paths, there is no way to deduce an end result as suggested by chaos theory.

The most beneficial technique when using simulation as a tool for discovery is agent based modeling. This technique creates a model with independent agents interacting with one another and the environment. The end result of the simulation reveals emergent properties which are a result of bottom-up processes from the agents. Agent based modeling is especially useful in the social sciences because of the nature of our societies. Humans can always act as the agents and experimenters can test out their theories of societies in these simulations.

Axelrod’s Simulation Research Process

Most of Axelrod’s recommendations for the research process of simulation I find as standard for any types of scientific research. “Analyzing your results and sharing what you have found with others.” The most interesting parts of Axelrod’s discussion of the process are the reproducibility of simulation results. I believe he is making the assertion that the reason little reproduction is done in simulations is the difficult nature of the process. However, I feel if anyone found a simulation relevant enough they would attempt to reproduce the simulation themselves as they would in reproducing any other scientific experiment. Perhaps the problem is the worthiness of simulations to do reproduction, and this shows how it is still not accepted as a third way of obtaining useful results, only more fodder for theory.

Should simulation become a more respected method of doing science, I believe Axelrod’s documentation on the problems of reproducing simulation science will be greatly useful as it addresses many preliminary problems such as computer compatibility and code ambiguities. Axelrod’s work provides many useful guidelines for simulation research as a whole and does establish groundwork for the field that he hopes will mature.

References

Axelrod, R. (2003). Advancing the art of simulation in the social sciences. Japanese Journal for Management Information Systems.