Authored By: Syhal Sookdeo
University of Kwa-Zulu Natal
Introduction
The maritime industry is at the verge of a technological revolution. Maritime Autonomous Surface Ships (MASS) will minimize human error, which is the main cause of maritime accidents, and provide a cost savings and higher fuel efficiency.[1] The International Maritime Organization (IMO) has classified MASS into four levels of autonomy, from the most basic (Degree 1: automated processes with seafarers on board) to the highest (Degree 4: fully autonomous ships with no human intervention).[2] The engineering is available for both Degree 3 (remotely operated) and Degree 4 (fully autonomous) vessels, but much of the legal framework and regulations in which they operate are still based on centuries-old tradition.
The very essence of traditional admiralty law is based on the existence of a human master and a human crew. Current legal regimes for collisions, salvage and pollution resulting from these events are rules based on the conduct, negligence, or reasonable foresight of a ship’s human operators. The introduction of MASS throws up a huge legal question: if an AI-operated ship is involved in a collision or spills oil, who is legally liable? This article discusses the allocation of liability for MASS in the absence of any law. It first explores the impact of autonomous technology on the current system of human “master” dependence under UNCLOS. Secondly, it considers the application to algorithmic navigation of the subjective COLREGs standards, like “fair” or “good” “seamanship”. Third, it describes the expected movement from maritime tort to product liability and how this will impact on the challenges of “black box” AI. Lastly, it reviews these global changes from the perspective of the South African Admiralty, i.e. the viability of an action in rem against autonomous vessels.
The Traditional Liability Shift and the “Master”
The allocation of maritime liability must be the starting point to comprehend the disruption caused by MASS. Under the doctrine of classical maritime law, the master and crew of a ship are held to be agents of the ship owner and therefore are liable for their tortious acts/omissions in the course of their employment.[3] The rule of vicarious liability is that victims of maritime torts can sue a seafarer’s job, the ship owner, not the individual and usually impoverished seafarer.
The need for the master is a fundamental part of international conventions. Under article 94 of the United Nations Convention on the Law of the Sea (UNCLOS) all the states are required to make sure that their flagged ships are operated by a master and the officers who have the right qualifications, especially in seamanship, navigation and marine engineering.[4] In Degree 4 MASS, no one is at the helm, and navigation is done by means of complex algorithms that respond to the data from the sensors.
The legal question is why legal recognition as a “master” is an issue when determining an AI system. As it is now interpreted by the law, “master” refers to a natural person, and Degree 4 MASS at present are in violation of the UNCLOS. The familiar lines of vicarious liability start to get skewed when the definition is broadened to include the shore-based remote operator (for Degree 3) or the shipowner who deployed the autonomous system. The ship owner no longer has to trust the independent professional judgment of a human captain, but the pre-programmed parameters of a software architecture.
III. Navigational Standards and the COLREGs
The first issue in relation to autonomous vessels and liability is in the Convention on the International Regulations for Preventing Collisions at Sea 1972 (COLREGs). The COLREGs are the “rules of the road” for marine navigation and the responsibility for a collision is normally shared according to who violated the rules.
The COLREGs contain plenty of anthropocentric language that’s hard to apply to an algorithm. No vessel, owner or master shall be exempt from any consequences which may arise from their neglect to abide by the rules hereof or from the neglect of any precaution which may be necessary in accordance with the definition of “ordinary practice of seamen” in rule 2(a).[5] The “ordinary practice of seamen” is to be judged by an objective standard of what a reasonable human mariner would have done in a sudden crisis (the agony of the moment). An AI is not in panic, but it also doesn’t have human intuition. Is it possible for an algorithm to show “good seamanship”? Or does the algorithm only perform a “literal” mathematical application of the COLREGs, and is the human-crewed vessel expecting “good seamanship” in the manner of the algorithm?
Additionally, as written in Rule 5, every vessel is required to keep a proper look-out, “sight and hearing.”[6] A Degree 4 MASS requires a high level of technology including LiDAR, radar, optical cameras and infrared sensors and goes far beyond what the human eye and ears can see; however, the courts will be tasked with determining whether the electronic data processing meets the legal requirement of human sight and hearing.[7] If the optical sensors on an autonomous ship misidentify an object, due to the lack of a non-standard shape in the heavy fog, the same questions that used to be asked when a human lookout was exhausted or distracted are now applied to the algorithm.
If an AI ship violates the COLREGs, the owner of the ship would be responsible for the violation in accordance with maritime law. But ship owners will never be able to say that the cause was not their responsibility to operate the ship, but rather the navigational software was faulty, changing the nature of maritime litigation.
The Model Shift to Product Liability
With the human crew gone, the place of error moves from human carelessness to technology failure. Accordingly, in future, allocation of liability is likely to shift away from maritime torts (negligence and unseaworthiness) and towards product liability.[8]
In cases where a MASS collides with another vessel, the plaintiff, whether it be a competing shipowner or an environmental protection agency, will probably file a lawsuit against the owner of the MASS. The owner will be added to the AI software developer, the sensor manufacturer and the systems integrator as third parties to the action. Product liability principles are in operation and place the burden on proving that the autonomous system was defective in design and/or manufacture, or that the developer was not careful to warn users about the system’s limitations.
This change brings about new complexities. Machine learning algorithms, especially for autonomous navigation, tend to be a “black box. The AI learns and adapts from millions of navigational simulations, and the software developers themselves might not know how and why the algorithm navigates a particular way in a unique real-world situation.[9] If the AI takes a very unusual, but mathematically sound, shot that leads to a collision is the software “defective”?
Additionally, most software developers make use of End User License Agreements (EULAs) that have very broad exclusion of liability provisions. Domestic consumer protection laws have limited such exclusions, but commercial maritime contracts between software giants and the owners of ships will result in intense jurisdictional and contractual battles. A Software Developer’s contractual immunity will be outweighed by the maritime law’s non-delegable duty of shipowner to provide a seaworthy vessel, according to the courts.
The South African Admiralty Perspective: Actions in Rem
When it comes to enforcing liability against autonomous vessels, unique issues arise on the procedural level, especially in jurisdictions that has well-established admiralty laws like South Africa. South Africa’s Admiralty Jurisdiction Regulations Act 105 of 1983 (AJRA) regulates maritime claims and in particular claims for damage caused by a ship.[10]
One of the fundamental principles of South African maritime law, which is based on English admiralty law, is action in rem: a legal proceeding brought directly against the ship itself (res), and not the owner.[11] In South Africa, the claimant has to prove, amongst others, that an owner of the property to be arrested should be liable in an action in personam.
The use of the action in rem is a perfect example of its longevity, given the introduction of MASS. With an autonomous vessel that is responsible for an oil spill off the coast of KwaZulu-Natal the coastal state or injured party can arrest the vessel itself even if the shipowner is based in one jurisdiction, the software developer in another, and the shore control centre in a third jurisdiction. Jurisdiction is present when a ship is physically present in the territorial waters.[12]
But after being arrested, the issue of who takes an appearance to defend the vessel brings up the liability gap. Assuming the shipowner can prove that the entire collision was caused by a defect in the AI software that his vessel was operating that was never known to him, does the claimant lose his maritime lien against the ship? Strict liability may be the key to the development of South African and international admiralty law, and the shipowners of MASS may need to move to such a regime. In this approach, the shipowner would continue to be held liable to third parties for any of the damages that the autonomous vessel caused, and the owner would be able to ask the software developers for an indemnity in a separate, complex commercial litigation.
Conclusion
The development of MAS ships for the global supply chain is outpacing the development of the laws governing the ships. The traditional approach, where the human master is the responsibility for the vessel, liability, seaworthiness and good seamanship, is in direct conflict with Degree 4 autonomous vessels.
The IMO is developing a compulsory MASS Code but soon, in admiralty jurisdictions, courts will have to apply obsolete provisions, such as the COLREGs, based on AI. If the current legal climate is not to be fractured, maritime law should evolve so as to ensure that the innocent third parties and the marine ecosystem are protected, by imposing strict liability on the MASS operators. At the same time, the industry needs to create comprehensive product liability insurance mechanisms to ensure that the risk is shared in an equitable manner between shipowners and the software engineers that are, in effect, the ‘masters of the modern autonomous fleet’.
Reference(S):
Legislation
Admiralty Jurisdiction Regulation Act 105 of 1983 (South Africa)
Convention on the International Regulations for Preventing Collisions at Sea 1972
United Nations Convention on the Law of the Sea 1982
Secondary Sources
Davies M and Dickey A, Shipping Law (4th edn, Thomson Reuters 2016)
Gahlen S, ‘Civil Liability for Accidents Involving Autonomous Ships’ (2020) 26 JIML 124
Gold E, Maritime Affairs: A World Handbook (2nd edn, Longman 1991)
International Maritime Organization, ‘Framework for the Regulatory Scoping Exercise for the use of Maritime Autonomous Surface Ships (MASS)’ (2018) MSC 99/WP.9
Staniland H, ‘Admiralty Law’ in Joubert WA (ed), The Law of South Africa (vol 1, LexisNexis 2003)
Van Hooydonk E, ‘The Law of Unmanned Merchant Shipping – An Exploration’ (2014) 20 JIML 403
[1] E Gold, Maritime Affairs: A World Handbook (2nd edn, Longman 1991) 142.
[2] International Maritime Organization, ‘Framework for the Regulatory Scoping Exercise for the use of Maritime Autonomous Surface Ships (MASS)’ (2018) MSC 99/WP.9.
[3] M Davies and A Dickey, Shipping Law (4th edn, Thomson Reuters 2016) 315.
[4] United Nations Convention on the Law of the Sea 1982, art 94(4)(b).
[5] Convention on the International Regulations for Preventing Collisions at Sea 1972, r 2(a).
[6] ibid r 5.
[7] S Gahlen, ‘Civil Liability for Accidents Involving Autonomous Ships’ (2020) 26 Journal of International Maritime Law 124, 127.
[8] E Van Hooydonk, ‘The Law of Unmanned Merchant Shipping – An Exploration’ (2014) 20 The Journal of International Maritime Law 403, 415.
[9] R Veal and M Tsimplis, ‘The Integration of Unmanned Ships into the Lex Maritima’ (2017) 2017 Lloyd’s Maritime and Commercial Law Quarterly 303, 318.
[10] Admiralty Jurisdiction Regulation Act 105 of 1983, s 1(1)(d).
[11] H Staniland, ‘Admiralty Law’ in WA Joubert (ed), The Law of South Africa (Vol 1, LexisNexis 2003) para 131.
[12] Admiralty Jurisdiction Regulation Act 105 of 1983, s 3(4).
