Bark to the future: The welfare of domestic dogs during interaction with a positively reinforcing artificial agent

Domestic pet dogs typically spend a large amount of time inside the home environment, including hours alone. An audio/food dispensing automated device (an artificial agent, hereafter agent) was evaluated for enrichment potential. The agent issued owner-spoken verbal cues to dogs, and food rewards for correct behavioural responses. The dogs’ welfare during interaction was examined using Qualitative Behavioural Assessment (QBA), and results were compared with quantitative behavioural measurements. QBA is a ‘whole animal’ measure, which describes and quantifies the emotionally expressive quality of an animal’s demeanour in a specific context, using descriptive terms such as ‘excited’ or ‘curious’. Adult pet domestic dogs (n = 17) were observed in two experimental conditions: 1) Agent; researcher and owner present (A+RO); 2) Agent; researcher and owner absent (A-RO). The agent was remotely triggered to call each dog by name and issue up to four repeated, randomised and previously trained verbal cues e.g., ‘spin’, ‘roll’, ‘sit’, ‘up’ (maximum 16 cues). In a baseline condition (Owner; researcher and agent present) (O+RA) the dog’s owner replaced the agent’s role. QBA using a fixed list of 20 descriptors was performed by three observers for each dog in each condition, using video recordings (51 clips) and Principal Component Analysis (PCA) applied. QBA principal component 1 (PC1) was the same in all conditions, characterising dogs as ranging from ‘attentive/interested/anticipating’ to ‘conflicted/apathetic’, with the majority of dogs at the former end. QBA PC2 revealed variability in dogs’ interactive styles, ranging from ‘calm/concentrating’ to ‘aroused/excited/persistent’ in agent conditions, and ‘calm/wary’ to ‘excited/aroused’ in baseline. Inter-observer agreement was high on all PCs (e.g., PC1 A+RO Kendall’s W = 0.88; PC1 A-RO W = 0.83; PC1 O+RA W = 0.77). Quantitative continuous behaviour sampling revealed no significant differences in state behaviour between conditions; ‘looking up’ (p = 0.494); ‘ears up’ (p = 0.662); ‘cued response behaviour’ (p = 0.630); these behaviour categories correlated significantly with QBA PC1 in each condition (average r = 0.800; r = 0.780; r = 0.793 respectively). Indicators of positive anticipation, sustained engagement, competence, and motivation were identified throughout testing and in all conditions. These findings suggest that positively reinforcing interactions with an agent or an owner are equally rewarding for dogs within the context of testing. This study is the first to assess welfare during interaction with an automated device and, using a novel application of QBA, suggests the overarching experience is pleasurable for dogs.