On the recent weekend of October 29 – November 1, I attended this year’s conference of The Computational Social Science Society of the Americas, at Santa Fe, New Mexico. This Society emphasizes computerized agent-based modeling. So their work bears some relation to my project in this blog. My project, to give a clue to newcomers, is exposition of a new model of our human experience, the Resource-Patterns Model of Life (RPM). Below I will comment upon two of the papers presented at Santa Fe. The second of those papers led me to read a few papers by Robert Axtell. I will end my comments here with what I found in Axtell’s papers. Overall I found affirmation in the conference. Affirmation, that is, of the approach to modeling I use in this RPM project.
Mirta Galesic presented her paper with Daniel Barkoczi, “Social learning strategies, network structure and the exploration-exploitation tradeoff”. The “social learning” subject of this paper catches my attention because social learning could describe the main question of interest that arises in RPM. Evidently this paper is only a small part of a body of literature on social learning now available. This literature has a context which bears some relation to RPM, so it will need to be studied as research is undertaken on social learning within RPM.
Digression: Before proceeding with the remainder of this post let me tell that I struggle to understand what I am doing on this blog. In a sense my drive is clear: I am promoting RPM which gives a better understanding of some important facts of life. But I have paused over questions such as: What is a model? Does conjecture have a place in science? I want my methods of both science and communication to withstand scrutiny. So I am looking for guidance, looking for what I may learn from the experience of others — especially those who have labored to communicate new agent-based views of science.
With that said let me say that in RPM the agents and their world remain, at this stage, mostly a thought experiment. I have started to computerize the agents as I reported earlier, but that effort did not promise enough rewards at this stage to justify the time it would require. So I am presenting a thought-experiment agent-based model of economic and psychological life.
Matthew Koehler presented a paper “Exploring Organizational Learning and Structuring”. But for me the most interesting part of Matt’s presentation was a long digression with which he started. In this digression he talked about the degree of specificity of agent-based models, praising what he had learned from Robert Axtell’s papers (see below). Matt showed a table (unfortunately not in the available paper) in which one axis represented the degree of specificity. Thought experiments were shown at the low-specificity end of that axis. I was heartened to see this because it seems to endorse the usefulness, within a broad view of science, of thought experiments at the early stage of development of scientific theories.
Matt also offered a suggestion for how to make a paper about a computerized agent-based model more meaningful to a reader. This was to lay out natural-language sentences, the sentences describing the agents and their world, in the first column of a table. Then in the second column give a reference to the computer code, that is to the line numbers of computer code in an appendix with the paper. This linking of natural language to computer language would empower a careful reader to understand exactly what is meant by the unavoidably fuzzy meanings of natural-language sentences. It may be impossible for some readers to gain a comfortable feeling of comprehension without such specifics. That may be helpful for me with exposition RPM. In some cases I have told more detail than I had believed necessary at first. But still I do not know how much detail to spill.
Robert Axtell, cited by Matt Koehler, has written in two papers (see references below) about a helpful way to categorize agent-based models. He suggests models be evaluated based upon their correlation at two levels with the empirical world. The two levels are the micro level of the agents and the macro level of system-wide developments. In each of these levels correlation with the empirical world is ranked, ranked to be either qualitative or quantitative, with quantitative judged to be better. This categorization ranks a model with a number from 0 to 3, with 3 being awarded to models which correlate with the empirical world at both micro and macro levels.
To equate Axtell’s terms “micro” and “macro” with what we use in RPM, recall that RPM allows for consideration of life in four or more levels. Almost all of our attention in RPM will focus upon the psychological implications of how living things on one level may advance to the next higher level. So Axtell’s term "micro" might refer to any level n in RPM, and then "macro" refer to level n+1. Also, micro might refer to the critters and macro to the organizations of critters which form in response to resource patterns.
RPM would be ranked at the lowest level in Axtell’s categorization scheme, because it is not close to empirical quantification at either micro or macro level. But that is alright, because RPM is still in the early fuzzy-language stage of paradigm development.
R. Axtell (2005). “Three Distinct Kinds of Empirically-Relevant Agent-Based Models”. Brookings Institute, 30 September 2005.
R. L. Axtell, and J. M. Epstein (1994). "Agent-Based Models: Understanding Our Creations". Bulletin of the Santa Fe Institute, Winter 1994, pp 28–32.
Daniel Barkoczi, and Mirta Galesic (2015). “Social learning strategies, network structure and
the exploration-exploitation tradeoff”. CSSSA 2015 link.
Matthew Koehler, Luciano Oviedo, and Michael Taylor (2015). “Exploring Organizational Learning and Structuring”. CSSSA 2015 link.