Evolution of preference for consonances as a by-product

Recent theoretical studies of evolution of visual signals in animals have revealed that biased preferences for symmetric patterns or simple coloration can evolve in the absence of positive fitness effects. In this paper, we study the evolution of biased preference for auditory signals. In music theory, intervals between a pair of auditory signals are classified into consonances and dissonances. Consonances are more comfortable to listen to than dissonances, and often have a frequency ratio close to a ratio of small integers. By examining the preferences shown by a three-layered network as a simplified model of an auditory system, we assess why we find consonances comfortable and dissonances uncomfortable. When the network was trained to accept monotones accompanied by harmonic tones and to reject random signals (noises), it developed a preference for consonances rather than dissonances. This suggests that the preference for consonances may have evolved as a by-product of training for a simple task, such as distinguishing mother's voices from noises, rather than as a result of being taught one-by-one. When the network was trained to favour a consonance and to reject a dissonance, it did not generalize the preference to other consonances or dissonances.