AI and seafarers

Could seafarers use AI based tools onboard? Many are already using them, even if unauthorised, we heard. But when under stress, seafarers should not have to worry about whether the material they see is accurate

When seafarers are under stress, they don’t have time to consider whether an AI based software tool is giving them the right answer, they need to get the ship out of the port, said Phrixos Papachristidis, CEO of Hellespont.

He was speaking at People Tech Maritime in Hamburg in April.

As a shipowner, Mr Papachristidis would prefer to employ seafarers who know how to run ships without asking ChatGPT, he said. And he would prefer seafarers with a question to find the answer by another means, so they are not reliant on a computer.

As a manager, you don’t necessarily want to make life too easy for crew, in case they stop thinking for themselves, he said.

People using these tools need to remember they may need to check its results, he said.

And AI tools do not necessarily reduce training, they can also increase it. “AI requires training in itself. The manuals are going to get bigger and bigger.”

“I’d like to think [AI] is going to be easy, [but] I’ll believe it when I see it.”

Answering a question with large language models can turn out to be more time consuming than doing it the traditional way, he said.

In one example, he asked Copilot to make a cashflow model for a potential ship investment. He asked the model to be checked by a professional, who said it was “just wrong”.

A friend working in real estate did something similar and used multiple AI tools to evaluate each other’s results. He got a good answer in the end, but it “ended up being more work”.

The biggest use of such tools may be in market research, where “sometimes it finds the corners of the internet a Google search won't.”

Fabian Feldhaus, Flagship Founders

Many seafarers are already using large language model tools onboard via their mobile devices, including CoPilot and ChatGPT, said Fabian Feldhaus, managing director of maritime investment company Flagship Founders.

Shipping companies would have more control if it was done over company systems.

Most LLMs are programmed to always give you an answer, and never say that they don’t know, he said. This is not what you want in a safety critical environment. Tools for seafarer use would need to be trained to say when they do not have an answer.

However, Flagship Founders has invested in Sealenic, a company which makes a ChatGPT like tool designed specifically for seafarer use. The shipping company has full control of how the data is managed, he said.

The best way to get value out of a LLM is to ask it questions in natural language, such as “what are the biggest problems with port state control in Houston.” Many people still use it like a search engine, entering terms such as “Houston port state control” and expecting a certain answer, he said.

“In the very near future the technical barrier to connecting to systems [onboard] will go down to almost zero,” he said.

Natural language interfaces

Language model based tools could provide natural language interfaces to other software tools. This would make the interfaces more human, said Holger Ritter, managing partner with DRYNET, a maritime digital infrastructure company.

They could also work in the seafarer’s home language, and so reduce misunderstanding and errors, he said.

Some companies are developing agentic tools which can sit on top of multiple software tools, so you have one interface customised to your company, said Fabian Feldhaus from Flagship Founders. “I think it’s a very interesting approach.”

Recruitment AI

Many shipping companies use AI based systems to manage job applications.

Many shipowners are flooded with CVs, and do not have time to go through them all, said Alexander Held, head of commercial with maritime human resources company TURTLE.

AI tools can extract data about good candidates into shipping company systems.

This means that seafarers can apply for roles just by providing their CV and answering some questions, rather than having to make many cumbersome applications, he said.

These systems are also screening and rejecting CVs. They could reject a CV from a good candidate, if they don’t use the right key words on their CV, said Columbia’s Arun Sankar.

To make sure its seafarer recruitment process has plenty of human input, Columbia has around 20 crewing officers worldwide supporting crew applications in person, he said.

AI in crew training

AI tools can help with seafarer training. You can ask them anything, including questions you might not be comfortable asking a superintendent, said Speros Stamatis of gas training company EduGas.

Training for senior officers is different to training for junior officers, because junior officers want to build up the knowledge, senior officers want to understand multiple different cases and scenarios. AI tools can customise the training to the needs of the student.