Liveblogging the conference: Jim Griesemer

Jim Griesemer is one of my favourite philosophers. Here he's discussing the work of Herbert Simon on dynamical boundaries.

What Simon said was that subsystems are nearly decomposable - systems are hierarchical and their internal relations are stronger than their cross level relations - and that this is a criterion for the dynamic specification of such systems.

He discusses the Hora and Tempus example. Jim asks, how do either of them make watches at all? He asserts that scaffolding - any structure or element that facilitates the development of a system's skills or capacities - is sufficient to explain this, and that it undercuts Simon's example. This is based on Bill Wimsatt's 1974 paper.

Self-scaffolding may have arisen by internalising external scaffolding processes. Modularity in the evo-devo sense, has emerged as a developmental concept; this is a kind of scaffolding concept. What makes a part a module? Near decomposability? Do Simon's ideas help understand modularity. Sort of...

Simon sees evolution as a satisficer rather than a rationaliser, and that hierarchical systems have an advantage over other kinds of system. In the Hora and Tempus example, there is a probability of interruption and a number of steps needed to complete a watch. Hora has 1000 steps, while Tempus makes only a 10-step subassembly. The probability that Hora will complete is very low. Simon calculates Tempus' advantage as 4000:1. [Damn, I got that reversed. Hora wins, not Tempus. Reverse the names.]

Jim is not disputing the advantages of hierarchical systems, qualitatively. But that's not to say that near decomposability is good enough for evo-devo. He has arguments based on the failure rate distribution: is it a Poisson process? Systems reliability theorists do not take this assumption lightly. Failures early and late do not happen the same way.

Failure at different parts of a process do not have similar consequences. So what's implied by the assumptions of the parable? Do they give any help about systems in general? Are they true of biological systems that evolve? Wimsatt argued that evolved systems will be internally complex and not decomposable. So Simon doesn't have an argument for his view (yet).

So, what's left out from Simon's description? Similarities in process between Hora and Tempus drop out of comparison. Tempus would need to be a kinesthetic genius to have an assembly that could be held together by hand, and the likelihood that at step 999 the assembly would fall back to elemental pieces is very low.

There must be some hierarchical functional organisation if the product is to function as a watch. If the finished watch functions as a nearly decomposable functional system and the assembly was not hierarchical, where is the "magic emergence" step? Part 1000? The assembly process is not clearly described in the parable. Fingers support nonstable moments and therefore scaffold the assembly. Handling times of assemblies between makers are bound to be different, so the assumption of a Poisson process is violated.

Are we to suppose that if interrupted the 1000-step assembly the whole thing would fall to bits? There must have been intervening stable states. If Tempus needs a fancy scaffold, then Simon's description fails. The quantitative argument may be wildly unlikely.

Wimsatt pointed out that in developmental systems you often have complex descriptions that cross systems. The strong point of his argument is that one can question the assumptions of modularity - assembly and disassembly have different consequences. We should be wary of using near decomposability as a criterion of evolving systems.

Evolution may favour internalisation and entrenchment of scaffolding, and hierarchy is a byproduct. There are ways to think about this problem but it's going to take a much richer set of concepts.

I hope I got this all right.