Complex information processing without a neocortex: determining whether fish can solve complex tasks with `simple' brains

Abstract

The brain is responsible for producing a large range of diverse behaviours. Many complex behaviours, including language and abstract reasoning, are associated with the neocortex. A well-developed neocortex is often seen as synonymous with a high level of cognitive sophistication; however, while a well-developed neocortex must confer some survival advantages, it may not be essential for higher-order processing. The fact that some species of bird, which lack a neocortex, can perform tasks typically associated with the neocortex, suggests that alternative neural structures may be capable of implementing the same level of behavioural sophistication. Further studies focusing on the higher-order processing abilities of a range of species with different brains, especially those lacking a neocortex, are important to improve our understanding of the evolution of complex behaviours and determine how the development of the neocortex affects the cognitive abilities of animals. Fish represent a large vertebrate taxon lacking a neocortex, yet little is currently known about their higher-order processing capabilities. Therefore the overall aim of this thesis is to determine whether a species of fish can learn two higher-order tasks associated with the neocortex: 1) abstract concept learning, and 2) human facial discrimination and object recognition. Archerfish (Toxotes chatareus), a species known for knocking down aerial prey with jets of water, are used in all behavioural experiments. Using operant conditioning, the fish are trained to spit at a range of visual stimuli displayed on a computer monitor suspended above their aquaria.In Part I of this thesis, the relational rules archerfish use to solve a series of visual psychophysical tasks are explored. We show that archerfish can rapidly learn associative relationships between stimuli and when conducting the commonly used alternative forced-choice test, archerfish learn to avoid unconditioned stimuli rather than select conditioned ones. While this avoidance strategy was consistent throughout experiments, they showed flexibility choosing between well-known positive stimuli and unfamiliar stimuli, demonstrating that archerfish can apply complex decision rules to associative relationships. Archerfish were then tested for their capacity to learn abstract relationships between stimuli. Abstract concept learning can provide significant survival advantages as it enables animals to apply previously gained knowledge to new objects and situations. Three tests were used to determine if archerfish could learn abstract concepts and were based on odd-one-out, matched-to-sample and same/different tasks. Despite extensive training, only a small number of individuals showed any improvement during the training phase of the odd-one-out and same/different tasks and none could generalize when novel stimuli were introduced. While archerfish show an impressive ability to apply associative based rules to solve visual tasks, they show only a very limited ability to learn rules relying on abstract concepts.In Part II, the ability of archerfish to discriminate visually complex images was examined. Human faces were used as stimuli because correct recognition requires a high level of representational sophistication. Within the set of faces, very similar distractors exist (other faces) and the variability in appearance of a face under rotation makes recognition difficult (e.g. profile view vs. frontal view). In addition, human faces are an object class that is unfamiliar to archerfish precluding the possibility that specialist systems have evolved to process them. We found that archerfish can be trained to discriminate one human face from a large set of distractors. In a separate experiment, we found that they could also continue to discriminate two faces despite changes in orientation. The combined results suggest that some of the basic tasks involved in human facial recognition can be learned by archerfish and that they have a flexible recognition system capable of adjusting to changes in viewpoint.This thesis provides an extensive investigation into whether fish can perform higher-order processing. This work shows that archerfish can perform visual recognition tasks similarly to animals with a neocortex; however they are unable to apply abstract concept rules to visual tests. These data, combined with results from other studies, suggest that despite lacking a neocortex, archerfish have relatively sophisticated visual abilities. Further tests are required to determine if the limited ability of archerfish to learn abstract concepts is due to their speciesspecific ecology or if it is a limitation of all species of fish.