Interpreting Animal Welfare through behavioural and computational approaches

New evidence indicates that Artificial Intelligence (AI) has potential to change how we study and assess animal health, welfare and ultimately well-being. This paper adopts a conceptual synthesis methodology rather than an algorithmic validation approach. AI technologies will enable improved methodologies for examining emotional states in animals through computer vision, wearable sensors, multi-modal data integration, and machine-learning techniques. However, neurobehaviour and psychophysiology can be computationally characterised. Automation may lead to disagreements with experts. Ethical and regulatory concerns such as dataset bias, stress-inducing data collection, uncertainty in automated inference, and potential conflict with expert judgment remain significant limitations. To address these limitations, the Integrated Climate Animal Emotion (ICEA) interpretive boundary framework, this situates emotional stress inference within environmental, behavioural, and psychophysiological context rather than treating emotion as a directly measurable output. Overall, AI systems can narrow plausible interpretations of emotional stress; emotional uncertainty cannot be eliminated and must remain subject to theory-driven human oversight.