Maritime AI: the digital revolution that can save millions, if done right

"It's a stubborn beast that regularly hallucinates and gives you a six-letter word when you ask for seven." Geert Schouten, director at Shipbuilder, doesn't mince words when it comes to the current state of AI in the maritime sector. "But don't underestimate what it can already mean for your business. A container ship stuck sideways in the Suez Canal costs millions, and in shipbuilding, there are many of these 'small Suez Canals' hidden in your processes."

While tech giants pump billions into generic AI systems, the maritime industry struggles with the question: jump in or wait? Is it hype or a helpful tool? And especially: how do you prevent your company from falling behind competitors who are using AI smartly?

According to Schouten, who has been pioneering with language models and AI solutions for shipbuilders since 2019, many companies fall into the trap of unrealistic expectations.

The current status of maritime AI

Where does the maritime sector currently stand when it comes to AI implementation? According to Schouten, we're just at the beginning. "We've just started. That applies to almost everyone worldwide," he says. "Billions are being invested in generic AI. Of course, billions are not being invested in specific maritime domain knowledge."
Shipbuilder, with their SENSE system, is already five years ahead of many developments in the market. "A ship's construction specification can contain 300 pages, which we can run through our software and automatically convert into a structured data network. And we also provide a reliability score on that."

Small steps versus big promises

An important pitfall in AI implementation, according to Schouten, is that companies want to take too big steps. "Throwing everything in leads directly to hallucination," he warns. By 'hallucination,' he refers to the phenomenon where AI systems generate incorrect or irrelevant information as if they were facts."You can't force a language model to do or not do certain things. It's a stubborn beast," Schouten laughs. "They suffer quite a bit from what they call 'hallucinating.' If you Google AI and hallucination, you'll find the most brilliant stories."The solution? "Start with small steps. First analyze one sentence, then another. Take small steps, because that already provides so much added value by enormously accelerating the entire process."

Domain knowledge: the missing link

A crucial aspect that Schouten says is often missing in generic AI systems is specific maritime domain knowledge. "Training a model is a specialist job. Maritime domain knowledge is not 100% available in public models, that's a limitation. And your own company knowledge is certainly not available there."This explains why generic AI platforms like ChatGPT cannot simply be deployed for complex maritime issues. "Designing, specifying, building, engineering, and maintaining a ship is extremely complex. There's a lot of knowledge in people's heads, especially a lot of experience."

That one missed requirement can cost you millions: AI & requirement management
Where maritime AI is already proving itself, according to Schouten, is in structuring and analyzing requirements, thus requirement management. "Take a shipyard that outsources work to an electrical installation contractor. You can't say: 'I'm throwing it over the bridge, I'll hear when you come to pull the cables.' That's not a smart way of working."

With AI support, the working method changes enormously. "If you combine the process of collaboration with AI tools, you can categorize requirements much better. So: These are requirements from production, these are requirements from the electrical installation contractor, these are requirements set by the shipyard, and these are requirements from the customer. Through this classification, you can prepare the engineering package for, for example, the electrical installation contractor much better."This structuring of requirements is, according to Schouten, essential for reducing failure costs and miscommunication. "The difference between good and bad requirement management can save millions in a large shipbuilding project. And this is precisely where AI is already practically applicable."

Maritime AI: applications in the chain

Where can AI already add value in the maritime sector? Schouten sees opportunities in different phases of the process, starting with tender management:"Analyzing what the customer actually wants is already an interesting given. Has that customer told everything he or she wants? Or are there still gaps? If you analyze those requirements that are provided with AI, you can take each sentence separately and say: 'What does this actually say? Is it open to multiple interpretations? And what questions should I then ask my customer?'"A good start is crucial for the entire process, according to Schouten: "If you've done your sales phase well, that doesn't mean everything you've offered is perfect, but you do have insight into which part is perfect and which part is not well filled in."This contributes to cost savings: "If you have to figure out what is exactly needed during the production phase, then that is just a hundred times more expensive than when you've figured it out well in the initial phase. AI can help tremendously with that."

Reducing failure costs: the real business case

The ultimate business case for maritime AI, according to Schouten, is drastically reducing failure costs. "The costs of a ship must decrease by 15 to 20%. That's not a voluntary wish, but a hard necessity for the maritime industry to remain competitive," he states unequivocally."Look," Schouten clarifies, "if you have to figure out what's exactly needed during the production phase, then that's a hundred times more expensive than when you've figured it out well in the initial phase. AI can make those uncertainties visible early in the process."However, he warns against an instrumental approach: "Maritime artificial intelligence is not a goal in itself, but a new tool. And as with any tool: give a carpenter a hammer with a loose head, and that head flies through the window at the first blow. That can also happen with AI if you don't handle it carefully."Schouten concludes with a call to the sector: "I tell everyone: start experimenting now. Start small, learn fast, and build on successes. The hype is that everything can be done in a few months, but in the harsh reality of shipbuilding this is definitely not the case. AI is not a magic bullet, but it is a crucial tool for those willing to learn how to use it. I know it can save millions, if you do it right."