Many academic papers suggest different solutions how the COLREGs may be implemented in algorithms but the parameters controlling how close quarters situations are avoided may potentially differ depending on the type of used algorithm and its settings. When tuning a collision avoidance algorithm for a specific ship and voyage, the effects and potential consequences are basically unknown without an in-depth understanding and testing of the algorithm. This report highlights the potential effect of a limited number of input parameters of an algorithm and simulations indicate that the variances in parameters and their values result in different actions taken and that the predictability of autonomous ships in a traffic situation may be poor. As it is initially expected that autonomous ships will need to follow the existing COLREGs due to the mixed environment of both manned and un-manned ships, it becomes imperative for there to be clear and universally accepted requirements and standards for autonomous ships. These standards need to ensure that all autonomous ships not only follow a common set of rules and algorithms in traffic situations, but also that such set of algorithms reflects how professional mariners would handle the situation. The comparison of track patterns of autonomous ships and human operated ships in the simulations performed is a convincing argument that traffic scenarios handled by autonomous ships must be benchmarked against human operated ships, and that even simulations with combinations of manned and unmanned ships should be performed. The “orderly” track patterns of manned ships may also be regarded as a testimony of the strength and elegance of the COLREGs as a legal document, as it effectively balances the need for a set of clear, concise and universally understood regulations for preventing collisions at sea, with the flexibility to accommodate the unique circumstances of different types of ships and changing maritime conditions. To include all factors influencing human decision making in traffic situations and to potentially incorporate seafarer experience, flexibility and seamanship into artificial intelligence will require machine learning, more advanced neural networks, and a massive amount of data.