Automated computer vision monitoring of uncertain decisions, anxiety-like behaviour, and positive and negative affect

Decision-making is a fundamental feature of animal behaviour that can influence, and inform us about, animal welfare. An animal’s choices may reflect its affective state, lead to positive or negative welfare consequences, and tell us about its wants and needs. Typically, animal decisions are measured in terms of choices made and time taken to make them. However, the animal’s actual behaviour during decision-making is rarely quantified despite the fact that it may reveal important information about decision-making processes, the certainty with which an animal chooses, and its underlying affective state. Measuring the dynamics of behaviour during rapid decisions is challenging and one likely reason for a lack of research in this area. However, modern computer vision methods incorporating deep learning algorithms create new opportunities for investigating behavioural markers of animal decision-making. They allow us to address questions including: how the probability with which behaviour predicts choice outcome changes across the decision period; what movements, postures or expressions contribute most to accurate choice prediction; how individuals vary in the speed with which their choices can be predicted (a marker of certainty) and how this is related to affective state. Automated computer vision tools for using expressed behaviour to quantify decision dynamics could have a number of animal welfare applications including: provision of granular information on timing and certainty of choices enhancing both preference-testing and judgement bias research; detection of uncertainty, apprehension and anxiety-like states informing management or training of animals faced with new husbandry procedures and equipment; modelling decision-making processes using choice behaviour as a complement, or even alternative, to invasive neural studies of evidence-accumulation.