This pre-study explores the potential use of Large Language Models (LLM), like ChatGPT and GPT-4, in decision-support systems for maritime navigation. The aim is to determine if these models could be part of decision support systems for identifying and resolving complex traffic situations, as doing so, would enhance maritime safety and efficiency, as well as reduce the environmental impact of shipping. The study benefits from previous research carried out in the COLREG2 project, COLREG 2 - Potential consequences of varying algorithms in traffic situations, which examined the limitations of algorithms meant to support the resolution of complex traffic situations. COLREG2 concluded that the evaluated algorithms had severe limitations and stated scepticism regarding the development of suitable algorithms in the near future. Shortly after the end of the COLREG2 project, large language models demonstrated their uncanny ability to understand, and to some extent, reason about complex texts and tasks. These abilities seemingly address the shortfalls of the algorithms evaluated in the COLREG2 project, making the use of large language models in decision support systems for marine traffic situations thought-provoking.

Authors
Luis Sanchez-Heres, RISE Research Institutes of Sweden
Reto Weber, Chalmers University of Technology
Fredrik Ahlgren, Linnaeus University
Fredrik Olsson, RISE Research Institutes of Sweden
Oxana Lundström, Linnaeus University

In cooperation with
Carl Petersson, Zeabuz
Tobias Husberg, Cstrider