This article is based on our deep dive report into artificial intelligence in the maritime industry. The full report is available for subscribers to the Thetius Intelligence platform here. If you don’t have a subscription but want one, request a callback from our team.
In 2019, 45% of technologists surveyed believed we’d have general artificial intelligence (AI) by 2060. When we get there, AI will be capable of independently learning to carry out new tasks that aren’t related to what it already knows how to do. But we don’t need to wait for general AI—narrow AI is already here.
Narrow AI is restricted to one narrowly defined task. In the same way that you can’t use GPS to propel a ship, a narrow AI designed to identify fires on CCTV can’t optimise your engine performance. However, within those boundaries, narrow AI provides an incredible leap forward.
Unlike humans, computers don’t get distracted, fall asleep, or get offended if you disagree with them. They’re excellent at the boring, repetitive tasks that most humans hate. When it comes to data analysis, computers leave humans trailing far behind.
Since 2018, there’s been an 11% increase in projects and organisations claiming to use AI in their operations. Despite that, we’re only scraping the surface of the range of potential uses for AI. Even in the notoriously conservative maritime industry, we can’t escape the fact that AI is already here.
What is AI?
At its root, AI is a computer program that aims to learn, think, and respond like a human. Programmers who write normal computer programs must specify how they’ll respond to every conceivable scenario. AI doesn’t work like that. Instead, it identifies patterns in data and uses those patterns to work out its own responses.
Several technologies fall under the AI umbrella. In 2021, machine learning (ML) and neural networks (NN) are the most popular. ML is a type of AI known as supervised learning. To teach a ML algorithm to differentiate between container ships and tankers, you feed it training data—thousands of labelled photos of container ships and tankers—and it learns to tell them apart.
ML is limited by the practical challenges of converting and labelling data. Unsupervised learning, like that used in NN, overcomes this problem. It finds patterns in unlabelled data and uses these patterns to make predictions.
Even for neural networks, the quality of the data poses a problem—if the designer isn’t careful, the AI might learn the wrong patterns. A classic case is that of the AI intended to identify wolves in photos. Instead, because all the photos of wolves contained snow, the AI learned to identify snow.
Worse, if there’s any bias in the training data, it also shows in the AI model. A clear example of bias was the Amazon’s CV screening AI. The training data showed that being male was an indicator of success, so the AI rejected female candidates. Similarly, a study into the use of AI to predict juvenile re-offending in Catalonia showed a strong racial and gender bias. That bias reflects an obvious pattern in the historical probation data, regardless of whether it’s a fair and accurate reflection of the actual risk of re-offending. While humans have similar biases, the impact of a biased AI in widespread use could be much larger and harder to challenge.
Like all technology, AI relies on other technologies to function. Digitalisation, the internet of things (IoT) and big data are the main underlying technologies that support AI. Digitalisation is the process of using digitised information to improve your systems; the IoT connects physical objects, such as sensors and cameras to a network, simplifying digital data collection and transmission; and big data provides a means to process and analyse the data.
Barriers and Challenges
If digitalisation is a key supporting technology for AI, data silos are a key barrier. Data silos result from the lack of interoperability between systems. The differing formats of data make it difficult to collaborate or use the data without first spending resources converting it to a compatible format.
Solving data silos will take a complete rethink of the culture of competition. In the short term, organisations can try to break down the silos within their organisation. Looking to the future, alliances such as the Open Industry 4.0 Alliance aim to ensure operators and OEMs collaborate, to establish a common platform and semantics, and to ensure interoperability with no vendor lock-in.
The law is another key barrier to widespread AI. On the surface, there’s no relationship between AI and maritime law. However, the law needs to regulate AI.
The Comite Maritime International (CMI) and the International Maritime Organisation (IMO), along with many governments, are working on regulatory scoping for marine autonomous surface ships (MASS). But there’s more to maritime AI than MASS, and in an international industry the legal grey areas can curtail its use. The absence of international agreements make companies rightly wary of relying on AI.
Current Uses of AI in Maritime
The maritime industry has come a long way since the days of oars and sail. Clipper ships’ sleek hulls replaced caravels and galleons, then windjammers, with their labour-saving brace-winches and smaller crews, replaced clippers. Steam replaced sail, propellers replaced paddle wheels, HFO and diesel replaced coal. Cranes replaced union purchase, and containers replaced boxes, bags and barrels.
Each development replaced its predecessor because it was more efficient, safer or offered cost savings. AI optimisation follows this trend. It’s already helping to optimise fuel, maintenance, operations, paperwork, port calls, logistics, voyage planning and more. With regulatory and commercial pressure towards optimisation, these uses will only expand.
The maritime industry is among the most dangerous work environments on the planet, with seafarer death rates more than twenty times higher than workers ashore. Seafarers and other maritime workers face high-risk tasks, from oil-spill cleanup and search-and-rescue operations to firefighting and tank inspections.
AI-driven robotics can take over high-risk tasks, from tank entry to underwater hull inspection. In the near-future, automation will reduce the number of crew exposed to the dangers; one day, unmanned MASS might even eliminate the risk of going to sea at all. Even now, AI drones are being used for hold, hull and tank inspections, saving time and money.
After container fires led to several deaths, a National Cargo Bureau survey found 55% of containers failed inspection with one or more deficiencies, including misdeclared or improperly stowed cargo. AI can help to identify these containers before loading, improving safety.
Tracking and surveillance
It’s undeniable that tracking and surveillance have sinister connotations and have been misused throughout history. However, it’s equally inescapable that they have the potential to be a huge step forward in safety and security.
While regulations and procedures have come a long way in improving safety, they’ll never prevent every problem. Alerting the bridge when people fall overboard, tracking passengers in emergencies, identifying illegal fishing, pirates and vessels in distress, stopping fires as soon as they start, and recognising stressed or fatigued staff are just a few of the current AI tracking and surveillance tools in use in the maritime industry.
Future Uses of AI in Maritime
The industry’s come a long way, but we could go so much further. Ashore, AI is already used in countless ways that aren’t yet available in the maritime industry.
From cancer diagnosis to medical chatbots, the use of AI in healthcare is increasing. Despite that, it isn’t widely used in maritime medical care. However, with regulatory and financial pressure, it’s only a matter of time before AI-supported telemedicine comes into use at sea.
By law and custom, the ship’s master is, “master under God,” legally responsible for compliance with the ever-expanding body of laws, conventions, rules, recommendations and regulations. But they are not lawyers. Ashore, lawyers already use AI, Even simple chatbots can help laypeople with legal decisions. While maritime again lags behind industry ashore, given the complex legal environment faced by masters, this is an area ripe for disruption.
For ships, stability is critical. Before departure, officers calculate stability based on reported or estimated cargo weights. Incorrectly declared weights, miscalculated stability, and commercial pressure lead to serious marine accidents. AI systems could calculate stability in real-time, simply by monitoring the ship’s movement, saving time and improving safety.
We’re living in a future that, just ten years ago, was the stuff of science-fiction. Because of its international nature, the maritime industry faces unique barriers to AI adoption; however, as the technology matures, its use will encompass every aspect of the industry, improving both safety and ease-of-use.
When Steve Jobs unveiled the iPhone in 2007, no-one foresaw the countless uses for today’s smartphones. In the coming years, we can expect to see AI become as familiar as smartphones. Particularly at sea, the biggest barriers to AI adoption are not technical or legal: they’re human.
There’s a common quip that seafaring is the second-oldest profession. Whether or not it’s true, the industry has been around for a long time and is mired in tradition. Those time-tested ways of doing things have saved countless lives; however, unless we address the industry’s resistance to change, those traditions will become an anchor holding us back.