Sunday, September 07, 2008

Bounded Rationality?


I’m still finishing the last few chapters of Kauffman’s reinventing the sacred. In his chapter “Living into Mystery”, Kauffman gives an interesting description of models that deal with overfitted and underfitted expectations. This tied in nicely with one of my all time favorite papers by Willis (2004) that uses the idea of entropy to model the evolution and degradagation cycles of organizations. First let’s take a look at the logic behind Kauffman’s (& Vince Darley’s) model.

The idea is that with a little bit of historical information’s predictive models tend to be fairly simple – say a few Fourier wavelengths. As more historical information accrues, the models get more complex – say a dozen Fourier wavelengths. In this case, increased complexity implies increased precision. Increased precision increases the chances of disconfirming evidence. Models never fully match chaotic reality. This leads to a phase change (my interpretation) where simpler models become more robust.

“This changing pattern of the time series is itself generated by the increasing fragility of ever more precise models. In turn, the models the players build of one another undergo an oscillation between simple, but robust models that yield self consistent behavior for some time – temporary ration expectations that then lead to increasingly complex models, until their increased fragility leads to disconfirmation and a new pattern of behavior of the players, creating the nonstationary behavior.” (Kauffman, 2008, pp. 240)

Willis’ idea is that organizations cycle between scientific management(highly order organizations with low entropy) and chaotic management (weakly ordered organizations with high entropy). Most people would think of these two end points in terms of an overly bureaucratic company that is rigid and inefficient and a grass roots company in touch with its roots, but in need of better structure. Companies cycle between these two states, passing the highly productive complex realm along the way. During de-evolution there is a small, but non-zero chance of complete collapse. Thus companies tend to come and go.

Both these angles illuminate the cycling that is an inherit part of education. I think they also illuminate some people’s entry points into religion. Religion offers robust moral models that can grow in complexity until they require reformation and re-interpretation (de-evolution). The self- introspection religion encourages provides a chance for Willis’ entropy cycling to occur before one extends either side of the cycle too far. I think the difference is that the group adaptive characteristics of religion allows people to enter a stable, faith based, loop of high order and low entropy. Similarly I can envision new atheists in a similar loop on the low order, high entropy side of things. Neither loop really leverages the power that exists on the edge of chaos. In a similar light, people have a hard time stepping far enough away to see the benefits of what Kauffman would consider “bounded rationality” (and by this I don’t mean to imply rationality isn’t useful, only that it has its limits and simpler more robust models can sometimes be better than more precise, complex models).


Kauffman, S. (2008). Reinventing the sacred: A new view of science, reason, and religion, Philadelphia: Basic Books.

Willis, R. (2004). A complexity and Darwinian approach to management with failure avoidance as the key tool. In Complexity theory and the management of networks: Proceedings of the workshop on organizational networks as distributed systems of knowledge (P. Andriani & G. Passiante Eds.). Imperial College Press: London.

No comments: