Whoa, No-Go: Evidence consistent with model-based strategy use in horses during an inhibitory task

Abstract

It is thought horses lack the prospection and brain architecture required for goal-directed, reflective model-based learning which considers future states. We investigate whether horses can use model-based strategy in an inhibitory task. Twenty subjects were trained for three sessions of a Stop-Signal paradigm using positive reinforcement (+R) for correct responding. All twenty failed to reach learning criterion. Subjects continued to touch in “Stop” contingencies indicating horses either: cannot complete Stop-Signal tasks; need further consolidation; or are utilising model-based cost-benefit analysis. Subjects underwent a further three sessions with the addition of negative punishment (+R/-P) as a cost for errors of emission (EE). If horses lack the ability to complete Stop-Signal tasks, EE would remain high across both treatments. If horses found Stop-Signal difficult but the introduction of cost aided their learning, EE would gradually decrease throughout the +R/-P condition. If horses built a cognitive model of the task but developed a strategy of indiscriminate responding in the +R condition, EE would suddenly decrease with the introduction of cost. A significant, immediate reduction in EE was observed when cost was introduced (p=0.02) that remained stable throughout the +R/-P condition providing evidence consistent with model-based cost-benefit analysis in horses.