Overview

This book documents some experiments in “artificial societies”. The primary idea behind the book is being able to *grow* social behavior. The mission is broad and its goal is to model ideas from social science, and see them simulated and played out over time. While the approach that is used subsequently borrows from mathematics and artificial life, the chief ideas originate from social science.

The reasearch behind the book is funded by the 2050 project (run by the Santa Fe Institute, the World Resources Institute, and the Brookings Institution), whose mission is to investigate sustainable global system.

All of the conclusions in the book derive from running variations on a very small, simple simulation. The premise is that there are agents that can move about in a world, that can see a certain distance, and whose goal is to acquire food and not die. Food is dispersed at first around two locations. The means in which food is regrown changes depending on the parameters of the individual simulation. Later on, complicating elements are introduced, such as reproduction, other food types, trade, warfare, social networks, disease spreading, and debt and lending. With the complexity of the emergent behavior, it is easy to forget that the simulation is of a discrete grid that is only 50 by 50 squares.

While a great deal of interesting conclusions can be derived through these simulations, and the rules described are very simple and understandable, I find that not enough time is spent critiquing the underlying foundation of the simulation. The authors mention that vision works in the principal lattice directions, and that diagonal vision is not allowed to bound the agents “rationality”. What happens when vision within a radius is allowed? What happens if agents have a facing and can only view within a field around that facing? Agents can move, hop, to a place that they see, what if they could only move one square at a time? Agents are also capable of collecting all the sugar on a square after their move. What if agents cannot collect all of it at once? What if they have the option of collecting different ammounts? What if multiple agents can inhabit the same square? What if the world is continuous instead of discrete? If any one of these factors changes, what long term implications does that have for the subsequent rules described in the text?

All of these questions leave me wondering if the particular sets of rules that the authors described were carefully chosen for the dynamics that they produce. This is likely to not be the case, but the reasoning behind the low level elements of the simulations is still not addressed (although it has been made open), meaning that it can still be informed by bias. Because the simulations are evocative, there can be associations made between conclusions or elements of this model and the other system that are evoked.

Notes:

The author’s first concern is to approach social science in a radically different way than it is traditionally explored. The viewpoint taken is to explore bottom-up emergent patterns in social systems (as discovered by the simulations) instead of the traditional top-down means of looking at social science. Traditional social science tends to look at the world through a lens looking for one type of information or another. The authors aim for a new kind of social science, wherein macroscopic theories may be tested through generative simulations.

One of the more troubling flaws in the text is the equation of food or “sugar” to wealth. The flaw is troubling because it raises questions as to what, exactly, sugar or wealth is supposed to mean within the context of this simulation. If sugar is food, then agents are moving about to obtain food, creating something of a nomadic-scavenger idea. But if food is also wealth, then it seems like it is something to be hoarded and accumulated. Rarely ever do nomadic people hoard wealth. Wealth is necessary for survival, but in most societies, it is not something found while moving about. Wealth tends to exist to support societies that have grown to a size to no longer permit barter, and currency is currency because it is not used for anything else. It is thus troubling when sugar is called wealth and economical terms are used about its distribution.

Later on when pollution is discussed, the effect of one model of pollution results in an exodus of agents leaving one area to join agents in another area, which cannot support the new population. The lesson learned from this is that “environmental degredation can have serious security implications”. This is a trite revelation. In this case especially, so much is encoded into the model that it is far from realistic. The consequences of the model are nonetheless interpretable and evocative. Interpretation is associative, meaning that we connect real world ideas, rules, and consequences with the effects seen in the model, but whose real world analogue may be very different.

An example of especially arbitrary constants found in the text is a discussion of reproduction, which gives constants describing the lifespan of agents and how and when they can reproduce. The only difference between “male” and “female” agents in this simulation is that the “male” agents can have children on average for 10 turns longer than the “females”. There is no difference between how the males or females raise the children, and there is no gestation period for females. This distiction is the only one present, so why have it at all?

In a manner similar to Wolfram (although predating his book by almost ten years), the authors conclude that simple rules can generate complex behavior, and that changing the rules changes the fundamental ecology of the simulation. This finding is certain, as is the result that long term predictability is difficult. However, the applicability of these findings to other social systems outside of the 50×50 grid is less than compelling. This can be used for illustrating the fallacies in oversimplified models, but that seems to run counter to the authors’ intent.