An important goal of the research on fish hearing is to develop models of sound perception and behavior based on information encoded in auditory nerve cells. We first define questions using behavioral and psychophysical experiments on the hearing and sound perception capacities of goldfish. We then investigate the response of single auditory nerve cells to the same set of sound stimuli, and develop hypotheses on the dimensions of neural activity and computational strategies that the brain actually uses in making decisions about sound sources. As an example of this approach, experiments were carried out to determine the possible neural bases for sound intensity discrimination and perception. Behavioral studies measured the minimum detectable difference between two sounds that differed only in intensity, and the effect that sound duration had on this threshold. A model was developed hypothesizing that judgements about the intensities of sounds were made using the number of action potentials (spikes) produced: as sound intensity is raised, the number of spikes increases. We investigated the performance of an "ideal observer" in using the number of spikes to decide which of two sounds differing in intensity was presented. Using the principles of the Theory of Signal Detection, the predicted performance of an ideal observer making decisions only on the basis of the number of spikes was compared with the behavioral performance of the goldfish. This analysis showed that an hypothetical, ideal observer performed very much like the goldfish, both in terms of the minimum detectable intensity difference, and in terms of the change in this threshold caused by manipulating sound duration. This sort of modeling provides support for the notion that goldfish use the number of spikes from one or a small number of cells in this task, and like human and other vertebrate listeners investigated, functions very much like an ideal statistical decision-maker.
There are several major conclusions from this sort of modelling work. First, it suggests that ideal, statistical decision-making processes are common and primitive features of many and perhaps all vertebrate nervous systems. Among other things, this finding permits us to view human information processing in a general biological context, and gives confidence in the development of plausible animal models for human brain processes. Second, the similarity between fishes and mammals in these processes helps establish the fish nervous system as the simplest model for vertebrate brain functions involved in sensory decision-making. Present and planned experiments extend these models to performance and decision-making with respect to more complex sounds such as those that vary in amplitude over time.