INTRODUCTION
Artificial Intelligence (AI) is a tool using Data Enhanced Algorithmically Developed (DEAD) computing. The original algorithms are developed by humans and, if sufficiently sophisticated, they can allow an AI program to develop further algorithms (a set of instructions to be followed in calculations or other operations) based on vast data arrays made available to it.
These algorithms are essentially lines of “best fit” for multi-dimensional data arrays of many variables. Using multi-variable analysis, it can discern patterns that elude humans because of the plethora of data that they are faced with. Using a function called vector-embedding provides a probabilistic result, i.e. if this happens, that will happen (with the fact that the result has only an 85% certainty omitted), which is strangely similar to the ways humans think, but with added math.
Best Fit
It must also be noted that in making best-fit lines, it has been given the ability to manufacture data (a sort of fancy Autocomplete). A recent task assigned to AI in order to discover inflexion points in history over nearly 50 centuries (for the doubtful, the earliest known historical records are administrative, economic, and inventory-tracking documents from Sumer (Mesopotamia) and Egypt, dating to approximately 3400–3200 BCE.), identified over 10,000 such events. Just over 1,000 were checked by human researchers revealing only a 53% accuracy with the rest being backed up by invented references.
Above all, AI is derivative – it works with what it is given or can access via the Internet. The Internet in itself contains only a relatively small proportion of human knowledge and experience and a lot of historical data is skewed by historians and other chroniclers who omit, are unaware of or enlarge according to their own viewpoint.
The Pitfalls of AI
According to some estimates, only about 10–15% of the world’s textual, documentary, and archival materials have been digitised in any form. However, this refers merely to scanning or digital reproduction — it doesn’t mean the material is searchable or even publicly accessible online. It’s also only a smidgen of human knowledge, as by far the greatest source of human wisdom is transferred orally and not written down. So DEAD computing has access to, at most, 2-3% of the accumulated knowledge and experience of humankind.
Internet History Archives
Search engines only index a small portion of what exists digitally. The so-called “deep web”— which includes scientific databases, archives, institutional repositories, and unstructured content — is significantly larger but largely invisible to most web searches. Even among digitised materials, only a portion is structured and searchable as text, implying that less than 5% of the world’s textual knowledge is currently searchable in electronic form, and this knowledge is largely from Western sources. So, there is a lack of diversity in Large Language Models (LLM) and AI is host to a large number of noted and, also, unexpected biases (this is not referring to race or sex but to selection, stereotyping, bad filtering and data clean-up).
We can suppose that AI is like a rambunctious teenager, with large areas of ignorance and a tendency to fabricate data to suit its results.
This is not to underestimate what it can do which is a lot.
Why is this important for the maritime industry which has, to be frank, a high-consequence for error?
AI and Autonomous Shipping
An argument in favour of autonomous shipping normally follows the following path:
- Reduced Human Error
- Increased Human Safety
- Savings in Crew Count
- Increased Cargo Capacity
- Monitoring Cargo Conditions
- Monitoring Machinery Conditions
- Crew Space Optimisation
These points are political promises or half-truths.
Autonomous Perception and Navigation
53% of maritime accidents, according to a study from Cardiff University, are caused by grounding and collision, invariably attributed to human error and which AI could have a significant help, although not entirely eliminate.
The other 47% would be indifferent to AI assistance.
Increased human safety is a moot point since deaths and accidents aboard sea-going vessels generally occur because of illness and disease, person overboard, occupational accidents and suicide. None of these would be ameliorated by AI intervention.
AI could do little to lessen the injuries and deaths caused by, for instance, ferry capsizing, or on-board fires, loading or unloading accidents or vessel “loss of buoyancy”.
Discounting cruise ships and ferries, the crew numbers on maritime trade vessels of major categories vary between roughly 12 to 36 persons.
These are divided between those that work the bridge, those that work the deck and cargo, and the engine room team.
Vessel Control, Monitoring & Remote Operations Software
Engine room crew are already paired thanks to engine monitoring and automation. Normally, engine rooms have to have at least one competent engineer and one technician (oiler or rating) but can be classed as autonomous if they meet certain criteria but still require minimum safe manning.
Again, the bridge crew can be minimised, but minimum safe manning applies. Deck crew might be port-based for the most part, but that would be a risky option. Claims by Rolls-Royce that automation could cut labour costs by 90% sound far-fetched.
Increased cargo capacity is a bit of a pipe dream. No people on board certainly means that crew quarters can be removed, but such autonomy would require something called redundancy.
There are many mobile vessels working in hazardous locations that can be left unmanned. These all have a high degree of redundancy, whereby the loss of one engine, even one engine room, is dealt with by having several entirely separate engine rooms, all capable of individually running the vessel, so that flooding, fire, or technical failure does not cause a complete shut-down. These are all drilling vessels and support vessels. It is difficult to imagine any serious classification society certifying a merchant marine vessel with only one prime mover as safely autonomous.
The last three points, monitoring cargo, monitoring machinery and optimisation of crew space are already being done so AI does not bring additional benefits.
What AI can do is to stay at its post, observe the rules, not fall asleep and stay sober (alcohol has been banned in commercial shipping since the 2000s). It can learn which alarms must be addressed and which can be muted temporarily. The job of a watch-keeper on the bridge is generally boring, stressful and requires irregular hours. This is why the major factors that the 53% of maritime accidents that are collision and grounding are caused by inadequate look-out, failure in communication, poor judgement, fatigue, a very small proportion of insobriety and sundry other human errors.
It is estimated by Allianz Global Corporate & Specialty (AGCS) that roughly 90% of marine casualties and incidents are caused by human error, resulting in over US$1.65 billion in marine liability insurance claims. In a world of instant satellite communication, an on-board AI assistant watch-keeper would virtually eliminate collisions and groundings and would be a relatively economic solution.
Completely removing human crews would require a major shift in Merchant Marine Business Planning. An AI-driven Autonomous Ship might cost nearly ten times that of current ship designs and construction, depending on the degree of redundancy required, and would need port support geared for such a vessel.
Coastal cargo vessels, ferries and port tenders and tugs would be perfect candidates but large ocean-going traders with a lifespan of a few decades might never recoup the extra cost. Tight margins lead to the conclusion that full autonomy for ocean-going vessels is potentially some time away. Indeed, a Chinese coastal trader, the Zhi Fei operating between Qingdao and Dongjiakou, a distance of nearly 90 nautical miles across the Bohai Sea completed both and autonomous crossing and autonomous mooring and offloading. It is a relatively small box ship capable of 300 TEU with electric and hybrid propulsion. In Norway, the Yara Birkeland sets the standard for autonomous, zero-emission, electric container transport in Scandinavian waters, reducing reliance on trucking and there are at least two Russian vessels with autonomy operating in the Baltic. Semi-autonomous ship handling tugs are being trialled in Singapore and Abu Dhabi.
At the moment, world trade looks like it is heading into choppy water as world powers re-align to make way for the parvenu powers and the older major players start to decay.
Optimists suggest that ocean trade will continue to grow, predicting a nearly 30% increase in seaborne trade by 2030 and a rise in tonne-miles to 74,000 billion. If so, then there will be a much higher risk of marine accidents. This is a moot point, whether ocean trade grows, flattens, or diminishes slightly; then the benefits of adding AI to bridge operations and voyage planning are clear.
AN EU project called Maritime Unmanned Navigation through Intelligence in Networks (MUNIN) predicted savings of more than € 6 million per ship over a 25-year period in fuel consumption and crew expenditures (the most critical expense in vessel operating costs). As vessels migrate away from Marine Diesel Oil, savings might increase beyond that.
It is reported that at least €23 million has been spent on projects from Sea Machines Robotics, EU’s MUNIN, SINTEF’s Seatonomy, and Rolls-Royce’s Advanced Autonomous Waterborne Applications Initiative. There will also be investments in setting up terminals, port and onshore operations to monitor fleet movements, and deal with port repairs and load transfers where there are incompatibilities between the current marine fleets and the service providers. Stating this, the onshore investment requirement will likely be far higher than €23 million.
Conclusion
It is inevitable that automation and AI will impinge on every part of the maritime industry, and should unmanned ships become a significant factor in the future of the merchant navy, then the industry needs to act now in order to prepare our current and future workforce by reforming education and boosting training programs that support seafarers to work with AI and automation.
AI and automation are tools for assisting humans and it is highly unlikely that the ship’s crew can be replaced completely with technology. Shipping companies need to elevate human potential with technology by using available AI-powered tools to predict operational risks, navigation solutions, voyage optimisation, and such to reduce maritime accidents.
It would appear that there is excessive promotion of AI in shipping. Automation has been installed on new builds for several decades and has already helped reduce costs. AI will further assist, no doubt. Companies like SEA.AI, Kaiko Systems, Winward.AI, Greenroom Robotics, ORCA.AI, Diversemarine and Virtualworkforce.ai all offer AI enhanced bridge systems, enhanced AI-assisted inspections, corrosion detection, and predictive maintenance insights and business exposure assessment (probably not for the Persian Gulf and Red Sea at the moment).
Preliminary data is available as of 20th of March, 2026. This new report examines and evaluates the emerging autonomous vessel marketplace. Detailed analysis will discuss the varying degrees (1-4) of autonomy available – minimally manned, remotely manned and fully autonomous.
The report will cover the technical requirements of AMVs and outline the expected future market potential. This will include a thorough inspection on small boats used for specialist tasks; medium-sized vessels; larger international cargo vessels; military craft; subsea crawlers and subsurface vehicles. For more information on our autonomous vessel research, click here.
