To obtain rewards and avoid punishments in complex and dynamic environments, animals learn to make predictions about upcoming events and select appropriate courses of action to satisfy their biological needs. A common approach for assessing learning and choice behaviour is the use of the appetitive two-choice discrimination task. Despite the widespread use of this method, exactly how animals learn to adapt their choice behaviour through trial-and-error in order to maximise reward remains unknown. Indeed, many studies focus purely on metrics of choice performance (i.e. correct or incorrect) which do not necessarily provide insight into how response selection changes throughout the task acquisition process. Furthermore, averaged data from multiple subjects is normally used to assess the learning curve, which produces inaccurate estimates of the emergence of performance improvements in individuals. Finally, much of our understanding of learning and decision making has, of necessity, been born out of studying discrete elements of behaviour in reduced environmental settings and so does not necessarily allow behaviour to emerge in a naturalistic fashion.
The aim of this thesis was to construct a rich profile of behaviour in mice as they acquired a simple visual discrimination task in a complex testing environment. A fine-grained statistical analysis of each individual performance record was executed to systematically identify distinct phases of task acquisition that were then deeply interrogated for a range of behavioural measures that went far beyond simple metrics of choice performance. This close analysis revealed how different choice strategies emerge, develop and interact under different task contingencies. The findings suggest that choice behaviour is highly structured throughout task acquisition and adaptation to change, even when performance is at chance levels. Importantly, multiple abrupt shifts in performance were observed confirming that the discrimination learning curve of individual animals does not necessarily manifest as a gradually increasing function, as proposed by conventional models. Finally, a subtle but detectable role for dopaminergic signalling in the substantia nigra pars compacta during the earliest stage of task acquisition is characterised.