Artificial intelligence (AI) is increasingly becoming incorporated into diagnostic imaging for image interpretation, triage, reconstruction, quantification, workflow optimization and report generation. The effectiveness of such tools goes beyond the algorithms but also relies on how the AI-derived information is interpreted and translated into actions. Human-in-the-loop systems allow trained medical practitioners to mediate between the AI-derived information and the action taken by the practitioner. This review outlines the potential of Physician Assistants (PAs) to assist in safe and efficient use of artificial intelligence in radiology practice. Evidence has been found to show that the inclusion of AI-assisted interpretations improves selected diagnostic tasks. However, external validation and clinical evidence are key factors to consider. Studies conducted recently have shown degradation in performance on external data, whereas new research on human-AI collaboration has identified automation bias, user interface, integration in the workflow, skill effect and governance as factors contributing to clinical value. PA can add value through provision of clinical context, patient preparation, urgency identification, communication and coordination, discrepancy reporting and quality improvement activities. Safe implementation of AI in radiology involves local validation, training in roles, performance monitoring, escalation process, privacy and multidisciplinary governance. The outputs from AI systems must only be treated as supporting information to make decisions and should never replace qualified radiological analysis.