Dull, dirty, and dangerous facets of human-AI collaboration for psychotherapy

Abstract

Despite an abundance of optimism surrounding the transformational potential of artificial intelligence (AI) in psychotherapy, it has become increasingly important to consider ethics and the risk-benefit tradeoff of these novel systems. Responsible AI (RAI) approaches within the field of Human-Computer Interaction (HCI) have produced a number of principles and tools to support more ethical AI design. However, it remains difficult to apply these principles and tools in psychotherapy contexts that must consider both provider interests and patient safety. A paradigmatic shift from human-AI interaction to human-AI collaboration and teaming further complicates matters and makes it all the more important to discuss appropriate solution spaces for AI, especially as factors like autonomy, accountability, transparency, and trust become distributed across dynamic teams. Through the lens of an updated Dull, Dirty, and Dangerous (D3) framework, this perspective piece offers a way of navigating human-AI collaboration in psychotherapy and the likely risks that might occur in practice.