Serendip is an independent site partnering with faculty at multiple colleges and universities around the world. Happy exploring!

You are here

Evolution and computer modelling

JoshCarp's picture
I've been interested by computer modelling approaches to evolution for some time now, and I've come across a few papers on the subject that might offer something to a discussion of emergence. A sizeable part of evolutionary theory cannot practically be submitted to direct test. Evolution can only be observed in rare cases--among bacteria in petri dishes or finches on the Galapagos islands. This is where computer modelling comes in. It has been applied extensively to ideas about the evolution of cooperation in social species. Studies of this sort typically examine simulated interactions among members of a population, pitting them against each other in games like the prisoner's dilemma. Members of the population follow different strategies with different levels of cooperation. After some predetermined number of rounds of the game are played, a new "generation" is born, with its proportions of cooperators and defectors determined by the relative success of each strategy during the previous generation. This process is iterated over hundreds or thousands of generations. The entire simulation can be run with parameters--say, the payoffs and punishments of the game, or the initial proportions of each strategy--altered. Some configurations give rise to fixation of one strategy, others to a stable polymorphism, others to steady oscillations in strategy frequency, etc. Some studies are surpassingly complex, incorporating lots of variables that might affect "evolutionary" outcomes. Some studies add smaller social networks within populations, with transfer among networks; individuals become more or less satisfied with their networks depending on the outcomes they meet with and can switch groups if very unsatisfied. Direct and indirect reciprocity, major topics in the theoretical literature, have been examined too. I think simulations like this have a lot of potential to expose details about the process of evolution, and also about how the strategies of individual actors interact to affect the behavior & development of the aggregate.


PeterOMalley's picture

My friends did something that is sort of related in high school: they wrote a neural network for what we called 'noids that had given inputs (I think they were seeing straight ahead, current energy, etc) and 3 outputs: turn left, turn right, and move forward/attack. These 'noids were in a virtual world where food would appear randomly and they had to eat food to gain energy/survive, and would spawn children once their energy surpassed a given level. The kicker was that the children could have mutated neural networks (that is, the morphology and weights of the links could be slightly randomly changed). The neural nets would start out entirely random, and over time you would definitely see an evolution as creatures that would initially behave randomly would be replaced by ones that would seek out food. It was interesting to note that they very rarely evolved to attack. I have the source code if anyone is interested in seeing it. I'm sorry if this was technical or didn't make sense; reply and I'll explain it more.