Dynamics of mimicry evolution

We simulated mimicry evolution by allowing three populations to coevolve: two populations of senders and one of receivers. Artificial neural networks were used to model receivers, and it was assumed that recognition was inherited. The senders’ signals consisted of nine dimensions. Changes to receivers and senders were caused by random mutations during the course of the simulation. Whereas it paid both types of senders to elicit the same response from the receiver, it benefited the receiver to respond in this way only towards one of the sender types. The receiver was thus in conflict with one of the senders, e.g. as in Batesian mimicry. Monotonically increasing response gradients caused the appearance of the model and the mimic to move in the same direction. Mimicry evolved because the mimic approached the model faster than the model moved away. Even after mimicry was established the model and the mimic were constantly changing in appearance. Our results conform with what is known in comparative psychology and ethology about how animals respond to stimuli. Several of our results are a direct consequence of recognition and have not, to our knowledge, been reported before, showing the importance of considering the recognition mechanism in detail when studying mimicry.