"Growing Artificial Societies"

[From Richard Kennaway (2009.10.16.2154 BST)]

Bill asked for some more information about the work I mentioned a few days ago, "Growing Artificial Societies" by Epstein and Axtell, where they perform computer simulations of populations of simple agents interacting by simple rules. Here's a summary.

For those who may not read to the end of this, I'll say at this point that there's associated software at http://sugarscape.sourceforge.net/. It doesn't look very easy to use. Their instructions about how to compile and run it seem unnecessarily complicated: double-clicking the Sugarscape.html file worked for me on a Mac. It should work on Windows and Mac OS X. The user interface is obscure and I can't see any way to change parameters, save and load worlds, or plot graphs of the form contained in the book. All I can see how to do is start and stop the simulation. To devise new types of agents would require progamming in Java.

The book begins by setting up Sugarscape, a terrain consisting of a grid of discrete points, each having a supply of "sugar", a resource that the creatures that live in Sugarscape need to consume. On each time step, each creature looks for the unoccupied point having the most sugar among those it can see, moves there, and acquires all of the sugar there. Each creature uses up its stock of sugar at a constant rate, and dies if it ever reaches zero. Each terrain point regenerates its supply at a constant rate up to its capacity.

The creatures vary in how far they can see and how fast they use up sugar. The terrain points vary in their sugar capacities.

These are maximising creatures, not controllers: they always try to add as much sugar as they can find to their stock. This applies throughout the book.

That's the simplest model. It results in the creatures congregating at the areas having highest sugar production. The population dwindles to reach the carrying capacity of the land. Those with high consumption rate and low vision tend to die first.

The concept of pollution is brought in. It's assumed that sugar consumption creates a substance called "pollution", which reduces the apparent value of a terrain point to the creatures. Pollution can diffuse to neighbouring cells.

"Culture" is introduced. An agent's cultural beliefs are simply a string of bits. When two agents meet (i.e. can see each other), there is a certain chance that a single bit of one agent's culture will be changed to agree with the corrresponding bit of the other's. In the long run, this results in all agents converging to a single culture, unless the distribution of resources in islands results in multiple communities isolated from each other, in which case each community converges separately to a single culture.

In the "sugar and spice" world, there are two resources, called "sugar" and "spice". Each creature needs both: if its supply of either reaches zero it dies. The two substances have different distributions. As in the simpler world (and in all the models in the book) each creature moves on each turn to the point among those it can see which it values the highest. In Sugarscape the value was simply the amount of sugar there, but in the sugar and spice world, the value depends on the creature's own supplies: it will value a resource in inverse proportion to its own stock of it divided by its rate of consumption, i.e. the time it would take to run out if it never got any more.

Trade is introduced. Two creatures that can see each other can exchange sugar for spice. The same valuation formula is used as for grazing. The authors devise a formula to determine how much sugar one will give to receive how much spice from the other, both benefiting from the transaction (i.e. receiving something they value more than what they give). A market economy then develops in which the exchange rate of sugar to spice is observed to approach an equilibrium with some random noise. In general, the fluctuation of prices in individual trades is greater, the less able the agents are to trade (e.g. by having shorter range vision, or lower population density, so they cannot find as many trading partners).

In some circumstances the exchange rate exhibits small or large oscillations. Famines and gluts can be imposed; the economy generally settles into a new steady state, though sometimes with quite a long transition period.

The book builds models of geographical seasons, war, mating, reproduction, disease (with transmission, immune systems, and multiple diseases), and finite lifespan. They observe epidemics, measure the Gini index (a measure of economic inequality), and exhibit various emergent phenomena. For example, in one case, the total population undergoes oscillations with a 115 "year" cycle, even though the average individual lifespan is only 80 years.

The authors draw a conclusion that "In complex systems there may be highly indirect and counterintuitive ways to induce social outcomes from the bottom up. [I.e. by changing the rules according to which individuals act.] Combinations of small local reforms -- 'packages' exploiting precisely the nonlinear interconnectedness of things -- may result in desirable outcomes in the large. Complexity beckons us to think that way." However, I am not convinced by the implied exhortation to design rules for real societies so as to bring about desired outcomes. The fact -- emphasized by the authors -- that small changes to rules can lead to unexpected results, implies that the differences between the simulation and a real society, even if small, may also result in the real society behaving in ways very different from the model. It seems to me that the models they build are at most suggestive, not predictive.

Searching Google, I get the impression that there was an explosion in this area in the mid-90s (of which this book is a part), but that it has declined in the last five years or so.

ยทยทยท

--
Richard Kennaway, jrk@cmp.uea.ac.uk, http://www.cmp.uea.ac.uk/~jrk/
School of Computing Sciences,
University of East Anglia, Norwich NR4 7TJ, U.K.

[From Bill Powers (2009.10.17.0715 MDT)}
Richard Kennaway (2009.10.16.2154 BST)

On each time step, each creature looks for the unoccupied point having the most sugar among those it can see, moves there, and acquires all of the sugar there. Each creature uses up its stock of sugar at a constant rate, and dies if it ever reaches zero. Each terrain point regenerates its supply at a constant rate up to its capacity.
...
These are maximising creatures, not controllers: they always try to add as much sugar as they can find to their stock. This applies throughout the book.

That's the simplest model. It results in the creatures congregating at the areas having highest sugar production. The population dwindles to reach the carrying capacity of the land. Those with high consumption rate and low vision tend to die first.

It doesn't sound to me as if there's any underlying principle or testable model here -- it's more like a board game than a model, with arbitrary rules set up to appear interesting.
...

"Culture" is introduced. An agent's cultural beliefs are simply a string of bits. When two agents meet (i.e. can see each other), there is a certain chance that a single bit of one agent's culture will be changed to agree with the corrresponding bit of the other's.

Do they ever indicate what difference it makes whether someone belongs to one culture or another?

The authors draw a conclusion that "In complex systems there may be highly indirect and counterintuitive ways to induce social outcomes from the bottom up. [I.e. by changing the rules according to which individuals act.]

The fact -- emphasized by the authors -- that small changes to rules can lead to unexpected results, implies that the differences between the simulation and a real society, even if small, may also result in the real society behaving in ways very different from the model.

I don't think they're even suggestive. If you make certain rules for your board game and people play according to those rules, there will be consequences. Make the rules complex enough, and the consequences will be surprising, or at least different from what you might have expected. But just giving familiar names to the board pieces doesn't mean the game has anything to do with any real system, or that the course of play tells us anything about a real system. This is really not modeling at all; it's a sort of Cargo Cult version of modeling.

Best,

Bill P.

[From Richard Kennaway (2009.10.17.1819 BST)

[From Bill Powers (2009.10.17.0715 MDT)}
>"Culture" is introduced. An agent's cultural beliefs are simply a

string of bits. When two agents meet (i.e. can see each other),

>there is a certain chance that a single bit of one agent's culture
>will be changed to agree with the corrresponding bit of the other's.

Do they ever indicate what difference it makes whether someone
belongs to one culture or another?

It makes no difference at all. "Culture" is just those bit strings, nothing more.

-- Richard Kennaway