June 16, 2014
John Holland

John Holland's Emergence is a brief introduction to … well, emergence; the principle that simple interactions between simple components can, without the benefit of a grand guiding intelligence, lead to complex, ordered phenomena as long as there are enough interactions going on. The quintessential illustration (which Holland mercifully doesn't beat on as much as he could have) is ant behaviour: While an individual ant is an incredibly simple organism living its life according to a quite simple set of rules, the many interactions between individual ants leads to highly sophisticated colony behaviour even without any kind of central guiding intelligence.

This is essentially the inverse of principle of reductionism from philosophy of science: In reductionist science, phenomena are decomposed into interacting component parts, and gaining an understanding of the component parts then allows us to understand and make predictions about the whole. Emergence is essentially the bottom-up companion to top-down reductionism: The principle of emergence holds that from the many interactions governed by "micro-laws" will emerge "macro-laws" that we can then use to make predictions about higher-level phenomena while ignoring the micro-level. Essentially, this is similar to the concept of abstraction levels in computer programming.

The book attempts to extract some general rules about the concept of emergence, through computer modelling of emergent phenomena - models ranging from playing checkers (board games being highly emergent systems), through Conway's Game of Life, to simulating a central nervous system. The central contribution is Holland's mechanism of constrained generating procedures, which he demonstrates are a useful means of modelling how "macrolaws" can follow from "microlaws", which are in fact exactly what emergence is about. They're much like the menagerie of simple automata studied by theoretical computer scientists, in fact, except that they're explicitly connection-oriented (precisely because they're used to model systems that attain complexity through interactions and connections).

It's not exactly light reading (although the heavier math has been placed in grey boxes, allowing the mathematically disinclined reader to skip it), but the writing is remarkably clear and concise. Whenever something raised a question in my mind, Holland ended up addressing it explicitly a few pages later.

Definitely recommended, particularly if you are the kind of person to maintain an interest in the weird intersection between natural science and theoretical computer science.

Powered by Plutonium