A systematic review of human-AI interaction in autonomous ship systems

Automation is increasing in shipping. Advancements in Artificial Intelligence (AI) applications like collision avoidance and computer vision have the potential to augment or take over the roles of ship navigators. However, implementation of AI technologies may also jeopardize safety if done in a way that reduces human control. In this systematic review, we included 42 studies about human supervision and control of autonomous ships. We addressed three research questions (a) how is human control currently being adopted in autonomous ship systems? (b) what methods, approaches, and theories are being used to address safety concerns and design challenges? and (c) what research gaps, regulatory obstacles, and technical shortcomings represent the most significant barriers to their implementation? We found that (1) human operators have an active role in ensuring autonomous ship safety above and beyond a backup role, (2) System-Theoretic Process Analysis and Bayesian Networks are the most common risk assessment tools in risk-based design, and (3) the new role of shore control center operators will require new competencies and training. The field of autonomous ship research is growing quickly. New risks are emerging from increasing interaction with AI systems in safety–critical systems, underscoring new research questions. Effective human-AI interaction design is predicated on increased cross-disciplinary efforts, requiring reconciling productivity with safety (resilience), technical limitations with human abilities and expectations (interaction design), and machine task autonomy with human supervisory control (safety management).