Reaching back to my master thesis, I am trying to
implement elements of the structuralist theory of
science into a computer program. Starting with a simple example, I use the
classical collision mechanics for a first attempt to work with theory elements
within a computer. A program written in PROLOG enables the entry of the theory
elements and optional data originating from a scientific experiment. It then
checks the theory parts (including the observed data) for completeness. Are the
available data sufficient to form a model, the program tells so and finishes. But
in case that there can not be found a valid model using the entered data, the
program can give some hints for filling this gap. It can propose a scientific
experiment and a method of measurement for gaining the missing data. By reentering
the new data into the computer, another approach for fitting the data with the
theory elements can be tried.
The experiences from this early work are used to establish multi-agent systems for scientific discovery and theory formation.
W. Balzer, C.U. Moulines, and J.D. Sneed, 1987, An Architectonic for Science, Dordrecht: R. Reidel
J.D. Sneed, 1971, The Logical Structure of Mathematical Physics, Dordrecht: R. Reidel
In this work I compare two approaches for doing
On the one hand I use a sequential multi-agent system called SMASS, on the other hand I use a properly distributed multi-agent system called DMASS (a short description of both systems is available here).
To perform the comparison, a model has to be found which is already available, quite interesting but not too difficult to implement. I chose the ‘Evolution of Support-Networks’ in the way Prof. Hegselmann established it. A detailed description of the model can be found in chapter 4 of the JASSS article: http://www.soc.surrey.ac.uk/JASSS/1/3/1.html, Understanding Complex Social Dynamics: A Plea For Cellular Automata Based Modelling.
Using the simulation systems and the mentioned model, I want to get answers to the following questions:
- Is one of the simulation approaches intuitively closer to reality compared to the other one?
- Which of the two approaches has a lower risk of producing simulation artifacts?
- Which simulation method is quicker to implement and easier to use?
- What is the gain in computing speed using the distributed multi-agent simulation system?
- Are there differences in the simulation results between the two simulation methods?
R. Suleiman, K.-G. Troitzsch, N. Gilbert (eds.), 2000, Tools and Techniques for Social Sciences Simulation,