The global flow of goods and materials is made possible by the trillion-dollar shipping industry. With 11 billion tons of cargo transported by ship each year, responsible, safe, and reliable practices are essential. Assurance is necessary to reduce risk and ensure shipping companies operate safe, reliable, and compliant vessels.
Why do we need to assure and verify AI?
With the increasing efforts around autonomous operations, the assurance of AI systems is critical. An AI system may deviate from its intended functionality due to unpredictable environmental conditions, or because the machine learning that is used to understand the autonomous vehicle environment has been unable to make sense of the data. Assurance is fundamental in providing clarity and confidence for end users when navigating new technologies that have little user experience. This is ultimately crucial for their broader adoption. Another aspect of AI is how we can ensure its ethical use as it becomes increasingly integrated into everyday life.
There is a growing number of cyber criminals at play that use AI to manipulate and falsify data. In shipping, AI-generated content poses a risk of being used to create forged documents, such as cargo manifests, certificates of origin, or safety inspection reports. Another example is the manipulation of sensor data that can be used to tamper with ship systems. Generative AI could be exploited to alter sensor data transmitted by ship systems, enabling attackers to falsify or manipulate readings related to navigation, weather conditions, or cargo status. This could mislead ship operators or automated systems, potentially resulting in incorrect decisions or unsafe navigation.
For example, in December 2019, a spoofing event between the Italian island Elba and the French island Corsica proved how easy it was to fake maritime AIS data and generate incorrect position information.
Thousands of AIS streams were received and assumed to be Dutch-flagged naval units. But closer investigation revealed the signals were fake. This type of incident poses a severe threat to the safety of navigation as other vessels may change course to avoid collision with non-existent hazards. This type of incident highlights how relatively simple it is for maritime AIS to be hacked and inaccurate position information to be generated as a result.
How do we assure AI?
It’s evident that AI requires assurance, but how does that look? If a piece of software is using AI, how do we know that the software is safe and capable of making decisions that do not endanger human lives?
The first step is to determine what safe behaviour means and then to ensure that the AI system is capable of behaving safely in all conditions it may encounter. Classification societies like Lloyd’s Register (LR) play a key part. They provide confirmation and confidence to end users that ships are built to the applicable safety standards and in compliance with classification rules and regulations.
With the emergence of new technologies, solutions, and practices in the industry, class societies must adapt and evolve their assurance processes to help ensure safe and successful integration while effectively managing associated risks.
In 2018, LR created an assurance framework, which has been applied to AI applications. It takes a four-phase verification and validation process that examines a company’s readiness to be a technology provider, the technology and product itself, testing of the technologies involved, and the validation of the product in service.
To learn more about the areas highlighted in this article, take a look at our thought leadership report, produced in collaboration with Lloyd’s Register titled, “Beyond the Horizon, Opportunities and Obstacles in the Maritime AI Boom” This report offers an in-depth analysis of the current state of operational AI in the shipping industry, shedding light on the latest trends, groundbreaking developments, and successful implementations. You can download a copy of the report here.
