Advances in the field of artificial intelligence (AI) have contributed to considerable progress in the development of automated vehicles in recent years. While press coverage has focused on autonomous cars and drones, autonomous shipping has also quietly been making headway with the UK ship register registering its first unmanned ship in late 2017.
AI is a field of computer science that includes machine learning, natural language processing, speech processing, robotics, and machine vision. It is already used across a wide range of industries in increasingly significant, and at times highly disruptive, ways. Many people assume that AI means Artificial General Intelligence (AGI) – that is, intelligence of a machine which performs any intellectual task as well as, or better than, a human can perform it. Or to put it another way, AGI is AI that can meet the so-called “Turing Test”: a machine's ability to exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human. In reality, we are some way off the emergence of AGI, although we already benefit from forms of AI which perform specific tasks.
Automation is not new in shipping and is already used for cargo systems, navigation and in engine rooms, but the wholesale adoption of unmanned ships (i.e. ships that are remotely controlled by humans from shore) and subsequently fully AI-enabled ships (i.e. ships with AI systems that can make fully autonomous decisions) is still some way off.
Nonetheless, there are ambitious plans within the industry to develop autonomous shipping. Norwegian company Kongsberg is aiming to have manufactured a containership which will travel a 37-mile coastal route in southern Norway and be fully AI-enabled by 2020, and Rolls Royce’s development programme is aiming for fully AI enabled autonomous ocean-going ships to be a common sight on the oceans by 2035.
The adoption of AI-enabled ships is, however, likely to vary according to shipping sector, with more risk averse sectors of the shipping industry (such as passenger ships) less likely to adopt fully AI-enabled ships for many years to come, while dry bulk and containers could begin adopting this technology much sooner.
The various benefits which commentators believe will drive forward the development of unmanned and/or AI-enabled ships include:
Advocates predict that both unmanned and AI-enabled ships will reduce costs, not only by cutting crew costs, but also by allowing more efficient use of fuel and space (given that a deckhouse and crew quarters will no longer be required). The bigger prize will be to eventually assimilate AI-enabled ships into an integrated and AI-enabled end-to-end supply chain with intelligent ports, AI-enabled trucks (something that commentators anticipate will lead to even greater efficiencies).
The shipping industry is facing a shortage of skilled crew wanting a life at sea. As a result, experienced crew are increasingly expensive (e.g. the master of an LNG ship is a substantial outlay). It is hoped that unmanned and AI-enabled shipping will negate this shortage by enabling fewer crew to manage more ships and also create more attractive shore-based jobs.
Advocates argue that unmanned and AI-enabled ships will improve safety overall. Analysis by Allianz estimates that human error is behind approximately 75% of marine liability losses. The use of sophisticated unmanned and AI-enabled ships will reduce human error and remove crew from high risk environments (removing the risk of injury and loss of life). However, conversely unmanned and AI-enabled ships will result in a changing risk dynamic. For example, fire risks and environmental risks could become more acute if there is no crew onboard a ship to deal with an incident promptly if it arises.
There are also a number of issues which will need to be addressed before unmanned and/or AI-enabled ships become a reality, including:
Regulation and liability
The existing regulatory regime does not contemplate unmanned or AI-enabled ships, and so is not fit for purpose in the context of such ships.
The Master of a ship is a key person in the operation of a ship. He/she is a representative of the flag state and has many onerous duties in respect of safety, management and both civil and criminal liability of the ship. The Master’s absence will have to be addressed in numerous regulations.
Safe manning levels are required in order for a ship to be considered seaworthy under key regulations, such as the UNCLOS and SOLAS Conventions. However, safe manning levels are subjective under the UNCLOS and SOLAS Conventions, which allow different jurisdictions the discretion to stipulate different manning requirements.
This raises the prospect that different jurisdictions could treat unmanned/AI-enabled ships in diverging ways. An uncertain regulatory environment created by such divergences in interpretation could lead to unpredictability in the way risk will fall in the event of damage (which could inhibit the adoption of unmanned/AI-enabled ships).
To add to this uncertainty, with regard to AI-enabled ships, traditional arguments on liability following an incident may also shift up the chain to the AI manufacturer (or others involved in the implementation of the system, such as those providing the data enabling the system to learn).
To facilitate the development of unmanned/AI-enabled shipping, the regulatory regime will need to be updated to provide more certainty about risk, responsibility and liability. Shipping is an international business, so updating legislation needs to be on an international level (through the International Maritime Organisation (IMO)) as well as at a national level. This process has already started following the meeting of the IMO in June 2017, where it agreed to commence a regulatory scoping exercise for unmanned/AI-enabled ships. However, it is anticipated that an updated international regulatory framework is unlikely to be in place until at least 2028.
When pricing risks, insurers need to be able to establish that the insured events will occur within a predictable range and that there is a causal relationship. This may be difficult where AI-enabled behaviour goes beyond a predictable range of foreseeable events. For example, AI determining the course of an AI-enabled ship may alter its reaction to certain events as a result of machine learning. The outcome may be unexpected (perhaps not within the normal range of human decision-making). As a consequence, coverage will have to be reviewed and policy exclusions considered with care.
This will become increasingly important for both unmanned and AI-enabled ships. Significant safeguards will be needed to protect against the hacking of, and taking control of, the systems controlling a ship. There have already been reported cases of crewed ships being hacked. For example, in February 2017 hackers reportedly took control of the navigation systems of an unnamed 8,250 TEU container ship en-route from Cyprus to Djibouti. The potential risks associated with, for example, hackers taking control of an AI-enabled ship, where there is no crew onboard to try to regain control of the ship, are even more serious (for example, the use of shipping to “weaponise” otherwise wholly innocent commercial activity, or simply to disrupt such activity). The UK Department for Transport recently released a new Code of Practice for Cyber Security for Ships in an effort to improve practices in this area for manned ships. However, this is an area which will need continued development in order to ensure the secure use of unmanned and AI-enabled ships.
Embedding human ethics
For AI-enabled ships to be accepted commercially, they will need to abide by certain minimum legal and ethical standards. Multiple scenarios demanding ethical and legal judgments that are difficult to foresee must be agreed in advance, and must be consistent. For example, if a ship gets into difficulties how should the system balance damage to property against damage to the environment? It is likely to be impossible to foresee every eventuality, therefore businesses producing AI-enabled ships will need to have a transparent and defensible process in place for the AI controlling a ship. More information on AI and ethical-legal problems can be found on the Norton Rose Fulbright Toolkit for Ethical-Legal Problems.
Keeping humans in the loop
The complexity of AI systems, in combination with the plethora of circumstances the AI systems could encounter on the seas, mean that constant monitoring of AI systems, and keeping humans “in the loop”, may be required. However, while keeping humans “in the loop” may help to achieve accountability, it may also limit the intended benefits of autonomous decision-making.
In order to train the systems running AI-enabled ships to deal with the range of circumstances that might arise, significant data sets will be required. It is unclear if there is sufficient data currently available, or if industry participants will be prepared to share their data sets. Another risk is poor quality data, and in the event of an incident arising from poor quality data, where will liability fall? Could the system developer shift liability by blaming the poor quality of data provided by a shipowner?
Given the global nature of shipping, in order to harness the full benefits of AI-enabled shipping, common technology standards need to be adopted across the industry. However, developers in the shipping industry (or the shipowners producing AI systems themselves) are likely to want their particular technology standards to prevail. What if a particular jurisdiction insists that only their AI system should be used on ships wishing to dock in their ports? As well as the practical issue of having to have a specific system on board, this could also raise security issues if, for example, geopolitical forces led a foreign power to want to take control of foreign ships using the AI system.
Day to day running
Unlike commercial aircraft, that only spend comparatively short periods of time in the air before landing, deep water merchant ships can spend weeks away from port. It is unclear whether technically it will be realistic for a ship to be unmanned for such long periods during which machinery needs to be repaired, routine maintenance undertaken and the effects of heavy weather remedied by experienced crew.
In order to unlock the full potential of unmanned and AI-enabled ships, collaboration will be required across the industry (including at a state and international level) and also across the supply chain in order to facilitate the development of a cohesive strategy.
It is unlikely that development of the technology required to run an unmanned or AI-enabled ship will be the factor that delays its widespread adoption, as there are other issues which may take longer to solve. In particular, international collaboration is required to update the regulatory regime and it is important that the international shipping community is involved at all levels. Legal systems will need to be updated on an international and a national basis in a clear and consistent manner in order to address how legal responsibility for loss or damage caused by AI systems will be apportioned. If the regulatory regime is updated on a piecemeal, state-by-state basis this will still result in an uncertain environment which could hinder the adoption of unmanned and AI-enabled ships.