Attention and predictive learning

Lead Research Organisation: Cardiff University
Department Name: Sch of Psychology

Abstract

The environments of animals have a predictive structure. Some stimuli signal events of importance - for example, the colour of a fruit might indicate whether or not it is ripe; whereas other stimuli are of no predictive relevance - for example, the size of a fruit might not indicate accurately how ripe it is. The brains of animals have evolved to learn about these predictive relationships, and theories of learning have been developed to describe the way in which this learning takes place. Although the theories do a remarkably good job of explaining how animals are able to appreciate the predictive structure of their environment, they do not provide a full account of predictive learning because they fail to take account of the fact that animals pay more attention to some stimuli than others. This omission is important because if an animal fails to attend to a stimulus, or attends to the wrong stimulus, then it will have difficulty identifying predictive relationships. One reason for this lack of understanding about the role of attention in learning has been the lack of a suitable methodology for identifying how much attention an animal is paying to a stimulus. As a consequence, not only do we have an incomplete understanding of how animals learn about the predictive structure of their environment, but we also have an incomplete understanding of the neural mechanisms that enable this learning to take place. The overall purpose of the project is to remedy this deficit in our understanding by making use of a new methodology we have recently developed in order to study the attentional processes of animals. A major concern of the project will be to study two very different theories of attention. According to one theory, animals will pay considerable attention to stimuli that are reliable predictors of events of importance, and little attention to stimuli that are irrelevant to the occurrence of events of importance. Experiments will test this proposal in the context of a discrimination where animals will be trained that two stimuli (A and X) that signal food, and two stimuli that signal the absence of food (B and X). The new methodology will then be used to determine if more attention is paid to the relevant stimulus (A), than the irrelevant stimuli (X). Additional experiments will identify the conditions under which these changes in attention take place. Experiments will then identify the neural structures that are responsible for these changes in attention by examining how the outcome of the behavioural studies are affected by damage to the hippocampus, for example. The second theory of attention assumes animals pay considerable attention to stimuli that have been experienced as inaccurate predictors of the events that follow them, because further learning about these stimuli is required; and little attention to stimuli that have been experienced as accurate predictors of their consequences, because learning about them is complete. Experiments will test this proposal for the first time in a direct fashion by training animals with two stimuli (A and B) that are always followed by food, and with another two (X and Y) that are intermittently followed by food. The new methodology will then be used to determine if more attention is paid to the inaccurate predictors (X and Y) than the accurate predictors (A and B). Experiments will then investigate the neural basis of these changes in attention by examining how they are affected by lesions to such brain regions as the central nucleus of the amygdala. It is likely that support will be found for both theories. Additional experiments will then identify the circumstances in which the different changes in attention take place. The results from the project will be used to develop a comprehensive theory of attention and predictive learning, and to identify the neural substrates of this theory.

Technical Summary

There is some disagreement about the changes in attention that take place during associative learning in animals, or if such changes take place at all. One reason for this uncertainty has been the lack of a suitable methodology for studying attention in animals. By using a recently developed methodology the proposed experiments are intended to overcome this obstacle. To demonstrate directly that animals pay more attention to relevant than irrelevant stimuli, two groups will initially receive an AX+ BX-, AY+ BY-, CX+ DX-, CY+ DY- discrimination. One group will then be used to test if an AX+ CX- discrimination is acquired rapidly, because it is based on previously relevant stimuli, and one group will be used to test if an AX+ AY- discrimination is acquired slowly because it is based on previously irrelevant stimuli. To test the claim that animals pay more attention to stimuli that are poor rather good predictors of their consequences, two groups will receive training with A+ B+ X+/- and Y+/-. One group will then receive an AX+ BX- discrimination, and another will receive and AX+ AY- discrimination. According to the foregoing claim the second of these discriminations will be acquired more readily than the first. Once the viability of the new methodology for studying changes in attention has been established, experiments will use the new designs to study the neural basis of attention. Thus the role of the hippocampus in tuning out irrelevant stimuli, and the role of central nucleus of the amygdala in enhancing to stimuli that are poor predictors of their consequences will be examined. Additional experiments will examine if changes in attention take place to individual stimuli, or to stimulus dimensions; if the changes are a consequence of top-down influences on stimulus salience, or of more peripheral influences based on orienting responses. Other experiments will examine the changes in attention to stimuli used for configural discriminations.

Planned Impact

The proposed research will enhance our understanding of both the theoretical and neural mechanisms of selective attention in animals. Who will benefit from this research? An increased understanding of the neural mechanisms of selective attention will be of value for the development of new drugs to combat such disorders as Alzheimer's disease, the cognitive deficits of ageing, and schizophrenia. Each of these disorders is characterised, in part, by an inability to focus attention on relevant information, and an inability to ignore irrelevant information. As we develop a fuller understanding of the neural mechanisms of selective attention, it will then be possible to develop new drug-based treatments for overcoming these deficits. The development of drugs that treat deficits of attention should enable patients to follow instructions, complete complex tasks involving sequences of behaviour, and acquire relevant rather than irrelevant information through learning. In short, such treatments will enhance the quality of life of the patients, and very likely help them to live more independent lives. An increased understanding of the theoretical mechanisms of selective attention in animals will be of value by allowing us to develop a more complete understanding of animal cognition and animal intelligence. One benefit of a growth of knowledge in this area is that it will point to ways for improving the welfare of animals kept in captivity. In order to provide the best possible environment in which to house animals it is not only important to be aware of their biological and social needs, but also of their cognitive capacities. A full appreciation of these capacities will permit the design of environments which an animal can most comfortably adapt to through the various mechanisms of learning it has at its disposal. The general public will also benefit from the proposed research. There is a widespread interest in cognitive disorders on the one hand, and animal intelligence on the other. By presenting our work within these general contexts it will be possible for a general audience to appreciate its significance, and to gain a fuller understanding of the origins of cognitive disorders and the nature of animal intelligence. How will the users be engaged? John Pearce has extensive experience of communicating with the general public through television (BBC2 series Animal Minds), radio, and the local and national press. He is also regularly asked to address a general audience. Mark Good has forged links with a number of pharmaceutical companies. Both investigators will build on this experience in order to disseminate their findings to the widest possible audience, and to exploit fully the therapeutic implications of their research.

Publications


10 25 50

publication icon
Cuell SF (2012) Changes in attention to relevant and irrelevant stimuli during spatial learning. in Journal of experimental psychology. Animal behavior processes



publication icon
George DN (2012) A configural theory of attention and associative learning. in Learning & behavior

publication icon
Gilroy KE (2014) The role of local, distal, and global information in latent spatial learning. in Journal of experimental psychology. Animal learning and cognition

publication icon
Horne MR (2012) Latent spatial learning in an environment with a distinctive shape. in Journal of experimental psychology. Animal behavior processes

publication icon
Horne MR (2013) The influence of excitatory and inhibitory landmarks on choice in environments with a distinctive shape. in Journal of experimental psychology. Animal behavior processes


publication icon
Kosaki Y (2013) Asymmetry in the discrimination of length during spatial learning. in Journal of experimental psychology. Animal behavior processes

 
Description The experiments used a new technique for investigating how animals distribute their attention during learning. We established they will attend to cues that signal important outcomes, and to cues that signal the absence of those outcomes. An offshoot of this research was the discovery of a set of findings that pose a serious challenge to a group of theories that state the learning is driven by error correction based on the collective predictive significance of all the cues present on a trial.
Exploitation Route To refine our understanding of the interaction between attention and learning.
Sectors Healthcare