During my long term work on the autonomous ship solutions and on board intelligent system design, I had the pleasure of talking to investigators from various Maritime Accident Investigation Bureaus (MAIB). Through these talks, I realised , that one scenario, which has to be considered and studied when designing autonomous ships is the accident and post-accident scenario. After managing the situation of an unmanned autonomous ship in the short time before the accident, during the accident and shortly after the accident, bringing the ship to a safe state, determine the amount of damaged caused and the ships ability to safely return to port, the task of performing an accident investigation follows. Such investigation would address what initiated the sequence of events leading to the accident, which decisions were made and why and how were they executed?
On the conventional vessels, MAIBs extract data from the VDR, access historical AIS data from VTS stations in the vicinity, go through the ship’s papers and documents, interview witnesses and question the crew. Besides the mountains of data available, one of the crucial sources to get a decent understanding of what happened and what caused the accident is the interview of the crew(s) on board the vessel(s). In case of the unmanned autonomous vessel, there will be a lot of data available to be collected and analysed, but there is no crew on board to interview, as the vessel is managed and controlled by the on board artificially intelligent control system(s). The question is then, what decisions were made by the on board intelligent systems and what caused the systems to make these decisions. What data was available and how well a situational awareness was accomplished by the on board systems. For an accident investigator, the autonomous ship may pose a new challenge and require completely new competences, because someone has to examine the on board algorithms, which are part of the decisions process and analyse the data received and crunched by the algorithms to determine, if there were any flaws in the algorithms or erroneous or missing data in the machine learning process and if any current data could have misled the AI. In other words, the traditional interview technique will have to be replaced by a new way of analysing the algorithms and data just before and around the time of the accident. But it may also be necessary to look at the data used for teaching the machines. The data provided for the machine learning purpose. The MAIB investigators have to build strong IT understanding and be able to comprehend how the design of the AI and machine learning parts may affect safe operation and decision making. The centuries of experiences build in maritime accident investigation using maritime inquiry as a way to get the full picture may not apply.
In case the control was transferred to an on shore control centre, there will be a remote operator as well to be considered in the equation, and it will require not only recording data on board the ship but also on shore, which will pose a new challenge in how to accomplish this the best way.
The MAIBs and the maritime industry will have to get together and rethink how to collect, store and retrieve data from the on board ship systems and on shore systems, and it will be necessary to find ways for the accident investigators to analyse the algorithms and get a good understanding of what triggered the on board sequence of decisions or lack of decisions leading to the accident.
Accidents have been a key source of information and learning to the maritime industry allowing us to understand what may increase the risk of an accident and how to mitigate this risk. We will still to need to collect and analyse data from accidents to determine causes, get wiser, improve procedures and design safer systems. This will not change with the introduction of intelligent ship systems and autonomous ships, but the way the work is conducted will change dramatically.
This guest blog is shared by courtesy of Mads Friis Sørensen – M/S MARTA – MARitime Technology Advising.
