Automated Operant Conditioning Devices for Fish. Do They Work?

Abstract

Simple SummaryAutomated training devices are commonly used for investigating learning, memory, and other cognitive functions in warm-blood vertebrates, whereas manual training procedures are the standard in fish and other lower vertebrates, thus limiting comparison among species. Here, we directly compared the two different approaches to training in guppies (Poecilia reticulata) by administering numerical discrimination tasks of increasing difficulty. The automated device group showed a much lower performance compared to the traditionally-trained group. We modified some features of the automated device in order to improve its efficiency. Increasing the decision time or inter-trial interval was ineffective, while reducing the cognitive load and allowing subjects to reside in the test tank improved numerical performance. Yet, in no case did subjects match the performance of traditionally-trained subjects, suggesting that small teleosts may be limited in their capacity to cope with operant conditioning devices.The growing use of teleosts in comparative cognition and in neurobiological research has prompted many researchers to develop automated conditioning devices for fish. These techniques can make research less expensive and fully comparable with research on warm-blooded species, in which automated devices have been used for more than a century. Tested with a recently developed automated device, guppies (Poecilia reticulata) easily performed 80 reinforced trials per session, exceeding 80% accuracy in color or shape discrimination tasks after only 3–4 training session, though they exhibit unexpectedly poor performance in numerical discrimination tasks. As several pieces of evidence indicate, guppies possess excellent numerical abilities. In the first part of this study, we benchmarked the automated training device with a standard manual training procedure by administering the same set of tasks, which consisted of numerical discriminations of increasing difficulty. All manually-trained guppies quickly learned the easiest discriminations and a substantial percentage learned the more difficult ones, such as 4 vs. 5 items. No fish trained with the automated conditioning device reached the learning criterion for even the easiest discriminations. In the second part of the study, we introduced a series of modifications to the conditioning chamber and to the procedure in an attempt to improve its efficiency. Increasing the decision time, inter-trial interval, or visibility of the stimuli did not produce an appreciable improvement. Reducing the cognitive load of the task by training subjects first to use the device with shape and color discriminations, significantly improved their numerical performance. Allowing the subjects to reside in the test chamber, which likely reduced the amount of attentional resources subtracted to task execution, also led to an improvement, although in no case did subjects match the performance of fish trained with the standard procedure. Our results highlight limitations in the capacity of small laboratory teleosts to cope with operant conditioning automation that was not observed in laboratory mammals and birds and that currently prevent an easy and straightforward comparison with other vertebrates.