AI tools like ChatGPT have been grabbing the headlines for months. MAN People asked experts from Group IT, PrimeServ Marine and Information & Cyber Security to give their assessment of this technology.

Thomas Alders, as the AI Ambassador in IT at MAN ES, you can surely share how artificial intelligence works with us.

IT systems that refer to AI are nothing new. The theoretical basis for systems that simulate neural networks has been taught for decades. The idea is to simulate the basic structures of the human brain, with protracted training phases resulting in an optimized weighting of artificial nodes and connections. Neural networks differ fundamentally from classic IT systems that run algorithms. What AI systems do is display intelligent behavior by analyzing their environment and taking action with some degree of autonomy. With the free accessibility of ChatGPT, OpenAI has created a huge media buzz. The system really does feel at many points like you are entering into a dialogue with a ‘universal knowledge machine’. ChatGPT is based on a combination of new technologies. ‘Chat’ is a reference to the dialolgic character of the system. ‘GPT’ stands for ‘Generative Pre-trained Transformer’, a pre-trained system that transforms text input to generate meaningful responses in natural language. The model has assimilated a huge amount of input during a pre-training phase, including web pages, books, news, etc. During pre-training, the model learns to predict an appropriate next word in a sentence, which helps it grasp grammar, semantics and contextual understanding. But beware – the system can also ‘hallucinate’, i.e. assert complete nonsense with great confidence. So AI still needs human intelligence for control. The evolution of the GPT model might, however, lead to a self-optimizing ‘AGI’ (Artificial General Intelligence) in the future.

Ingi Ulfar Helgason, you are in charge of a large AI testing ground at PrimeServ Marine. What are some of the projects that have already benefited from AI?

If we are to achieve our ambitious growth targets like ‘Move to Triple 10+’, we will have to employ technologies that make us more innovative, creative and productive. We already use Auto Quote to generate quotes automatically, and our team in Greece uses AI to create catalogs for our vessels. In another project, we employ machine learning technologies to sort emails, reducing the workload by two-thirds. So we have already found ways in which AI can help us. With AI we can prioritize the important stuff. We have to be careful, though, and make the right decisions, as AI is a reality that we need to adapt to.

Elena Furini, what are the right decisions when it comes to information security?

The potential opportunities, but also the risks, of AI are enormous – many of them have yet to be researched and evaluated. Despite all the enthusiasm, it is important to remember that AI is also used by criminals for malicious attacks. For example, deep fakes could impersonate board members. There are some general rules that help mitigate the risks: first, we must not disclose any internal or company-
relevant data until the tool in question has been reviewed and approved by MAN ES. There is no data-processing contract with OpenAI, and it is extremely unclear how data protection requirements, such as GDPR, are met. When using AI tools like ChatGPT, you need to be aware that the information can be used by the vendor for any purpose. Here, it is really important to read the terms of use carefully. This helps us establish how intellectual property rights are protected – which in turn frequently depends on the sources.