Artificial Intelligence is – depending on your perspective – humanity’s greatest technological advance or the greatest risk to its survival. The truth as ever lies somewhere in between.

With the technology still in its infancy it’s difficult to tell what the true impact will be, but a recent webinar organised by Riviera Maritime Media heard that even if its uses in maritime are few so far, its ability to support smarter shipping is not in doubt.

One of the traditional challenges in improving vessel performance is that the data gathered presents wide variations. Speed and power performance of a ship is full of deviations, reflecting everything from the master’s choices to charterer requirements making it hard to collect representative data.

This makes it hard to model with traditional techniques but neural networks find such variations easier to handle. In his work as Associate Professor of AI for Marine Applications, University of Southampton, Adam Sobey used AI to predict vessel power demand and got just a 2% prediction error in real weather conditions.

“That’s better than models or a towing tank. We get a 4% error for most vessel data and we can fuse the results to make fleet predictions. It’s not necessary to equip every ship, you can extrapolate with high accuracy,” he said.

Combining this data with draft and trim optimisation and air lubrication of the hull could provide considerable savings in fuel and thus CO2 emissions. Optimise the voyage using algorithms for weather, tides and port delays to adjust speed and arrival and he reckons owners could comfortably see a 7% annual increase in timecharter equivalent earnings.

French start-up SINAY provides a similar approach to tackling voyage performance, deploying algorithms to optimise navigation to increase efficiency and cut costs. The company has made a ‘Google Map’ of the sea to identify environmental characteristics that are difficult to compute conventionally but which dramatically affect efficiency.

This involves training the AI on real data that has been cleaned and verified against AIS data to provide higher confidence in the information on actual vessels and positions. “AI is not magic and it’s not going to make magic, it needs to be trained and built, it needs data inputs,” said Marie Besson-Leaud, Product Experience Manager.

Holding back wider use of AI is a lack of resources and a lot of misconceptions which need to be addressed among maritime users, she added. Many people have a notion of AI, but not how the applications can be relevant to their domain.

Sobey batted away the notion that AI automatically means ‘autonomous’ shipping is somehow a closer prospect; people will still be at the centre of seafaring, he argued. “AI augments how we do these jobs and protect people from danger. Machine learning is well adapted and has been used for decades but true AI with full automation process is in its infancy,” he said.

“With autonomous shipping it’s a gradual process of moving towards it. There are background drivers but most of the time there isn’t a business case. You still need people and removing the master doesn’t save you any money, you still need people to take leadership responsibility,” he said.

Besson-Leaud noted the psychological threshold around autonomous vessels and similar barriers in aerospace. User research shows real concern around AI replacing humans which is not the case for now. “AI is a support to human activity, even a semi-autonomous vessel will still have humans onboard,” she said.

For any start-up in this space, data is the new oil in the AI context and until now, we have been wasting a precious resource. In future there will be a need to transfer knowledge to avoid it being forgotten and to use data rather than throwing it away.

Besson-Leaud finds that data owners are still reticent about sharing “and you need a strategy on data sharing and let your partners have visibility”. As a result it may be some time though, until all the potential applications are realised.

“Ports in particular say to us, we have the data don’t know what to do with it,” she adds. “But AI is interesting because existing data can be integrated into the modelling, you can feed it with data already have provided it is clean and structured.”

But does AI have other implications – if it were possible to predict the weather and therefore a ship’s path in relation to it then would insurers find that valuable? Sobey thought not. Despite assumptions about weather becoming better and better, the values are of interest but still largely theoretical.

“The master is still in control so any navigational AI still relies on their vigilance in taking decisions, what the AI is doing is providing general guidance and visibility,” he said. 

Building a true neural network for navigation would require capturing all available shipping routes and port data and building these into an integrated VTS that could support traffic management, optimisation and vessel safety.

Sobey thinks regulation isn’t moving fast enough to support this; “We need data standards to do AI modelling and the lack of standards has a huge impact on our work, we need them in order to progress. We need an ontology, the building blocks so that the implications of change can be seen clearly”.