Though a fun paper, Technology Review's write-up of it is worthless. At most, all you should take from reading this paper is that "cartels" potentially can emerge without coordination, not that they do or that they are likely to. Or that this phenomenon has anything at all to do with the dynamics of prices at your neighborhood gas-station.

As anyone who has studied complex systems understands, the evolution of interactions among even very simple agents is enormously sensitive to initial conditions. (That's often taken as the very definition of chaos in deterministic systems.) So it is almost always wrong to draw strong conclusions about a complex part of the real world based on your computational model of it: vary one parameter a little bit and you end up with something that looks completely different.

This is why agent-based models, like the one employed in this study, are widely despised in economics and the other social sciences despite offering a way to do experiments that would otherwise be impossible in these fields. We simply don't know enough about people and how they interact to build meaningful models, and we probably never will.

For example, an economist might point out that the model presented in this paper doesn't account for the spacial distribution of gas stations and consumers, for buyers' and sellers' expectations of future prices, or for any number of conceivable and inconceivable factors—almost all of which would affect aggregate outcomes in the model, I would bet, just as the authors found that the speed of reacting to price-changes did. (That was the only parameter the authors varied, by the way.) And of course we could model the problem in many fundamentally different ways to start with.

In other words, this is much different (and harder) than building an N-body model of the dynamics of Alpha Centauri, where's there's basically one force (gravity) that operates in the same way on a bunch of things that can all be described adequately by a single parameter (mass).