48 Does Automated Behavioral Monitoring Inevitably Lead to Improved Pig welfare?

Abstract

Abstract Academics, producers, veterinarians, technology developers, and even the general public seem to be talking about using technology to solve agricultural problems. The starting point for many proposed technological solutions to help manage pigs is the automated detection of specific behaviors. These behavioral data then need to be translated into actionable information to assess welfare, and ideally to provide a management solution to ensure good welfare for the monitored pigs. However, detecting behavior of animals, particularly of relatively homogeneous animals living in large groups at high densities, such as pigs on commercial farms, is not a trivial problem. Most functional behaviors are complex and made of multiple smaller elements, including postures and simple motions. Training technology to detect behaviors with accuracy that mimics a human expert is difficult, due to behavioral complexity as well as to variation in how each individual does a behavior and how physical and social environmental context affect performance of a behavior. Disappointingly, many technologies that claim to detect performance of behaviors are in fact only detecting the proximity of the pigs to resources or to each other and then assuming that pig is interacting with the resource or social partner in a particular way. However, a pig standing near a feeder is not always eating, nor is a pig near another pig always deliberately interacting with that other animal. It is necessary to closely evaluate automated behavior detection solutions to determine if these do directly detect the behavioral outcomes of interest. The next problem is connecting the detected behavior to a specific welfare outcome. Animals typically perform behaviors for biologically motivated reasons related to maintaining physiological homeostasis, responding to disease challenges, or to expressing their emotional state. Yet making a direct link between performance of a certain behavior and a subsequent welfare state is not straightforward. For example, grooming is often thought of as a comfort behavior, but animals that are distressed may groom as a displacement behavior. Therefore, context and other measures are needed to complement the behavioral information before drawing a welfare conclusion. A final problem to consider is whether detection of a welfare problem inevitably leads to improved welfare. In some cases, such as when a technology identifies the start of an outbreak of treatable illness, action is possible. In other cases, the technology may detect a more intractable problem, such as lameness in a sow. If we cannot effectively cure the underlying problem or eliminate the associated pain or distress, does this raise ethical dilemmas or reduce trust in the swine industry? While technologies do hold promise to better monitor and manage pigs in labor efficient ways, we must mindfully develop solutions that monitor pigs' behavior and cautiously evaluate their ultimate impacts on pig welfare.