Study suggests methods to increase transparency for artificial intelligence use and reduce its bias, including through legal means and engaging with local communities.
The use of artificial intelligence (AI) is becoming increasingly common when deciding where offshore energy projects such as wind farms, oil platforms and pipelines should be placed in the ocean, a process known as marine spatial planning (MSP). UNESCO defines MSP as ‘a public process of analysing and allocating the spatial and temporal (time-based) distribution of human activities in marine areas to achieve ecological, economic and social objectives that have been specified through a political process’.
Essentially, MSP assigns uses for different parts of the ocean that can allow offshore energy projects to be carried out after taking into consideration the multiple stakeholders involved and their respective goals.
It can therefore result in significant consequences for the use of the marine space for those various stakeholders and communities. If decisions are made tainted bias and a lack of transparency, certain marine communities or users could well be disadvantaged.
Co-authored by Jason Chuah, Professor of Commercial and Maritime Law at The City Law School and Professor Juei-Cheng Jao, College of Ocean Law & Policy, National Taiwan Ocean University, a recent study sheds much needed light on this growing issue.
It suggests that while beneficial, the downsides and potential risks associated with using AI need to be examined further, specifically the potential effects of introducing hidden biases into MSP.
Of course, AI systems hold many benefits over human-led marine planning. During planning, AI is readily able to identify underground structures such as faults, folds and traps, can detect potential hydrocarbon accumulations (i.e., striking oil) and significantly reduce the speed of interpretation time of large amounts of data from weeks to hours.
Nevertheless, the risk of certain biases must be accounted for when examining the reliability of these systems – the quality of their decision-making will only be as good as the quality of the data they receive.
Supported by the Taiwanese Ministry of Science and Technology, the study explored approaches that might be adopted so that AI can be used more effectively within MSP with reduced bias and increased accuracy and utility.
Professors Chuah and Jao stress the importance of human oversight when using AI generated ‘recommendations’ to reduce the risk of bias rather than permitting these models to entirely dictate decision making. Being transparent with the use of AI can also build trust with stakeholders and provide a better understanding of the science needed for a proper implementation of MSP.
For example, including local communities who possess ‘long-term ecological knowledge’ that AI models might miss, is one solution suggested from the research.
More broadly, ensuring that under-represented groups such as fishers, coastal communities, port authorities and conservation groups are consulted prior to adopting AI-generated plans could also enable greater data governance.
As many stakeholders do not currently have equal access to data, this may exclude them from decision making in a valuable manner, jeopardising the validity of the execution and implementation of MSP policies at all levels.
One particular approach that the researchers explored was that of the EU AI Act which was developed in 2024 to address the risks associated with the use of AI technologies more generally.
The Act uses a hierarchy of ‘risk’ associated with AI products classified as ‘unacceptable’, ‘high’, ‘limited’, and ‘minimal’ – determining whether products should be prohibited. While it was not formed with directly addressing bias in mind, the study authors used its analysis of risk as a starting point for trying to reduce bias in MSP, with their research concluding that it is yet to be clarified whether AI systems used in MSP fall within the high-risk category.
The study also examined ways to improve transparency and mitigate bias through regional and local governance of marine activities, including legal frameworks imposing obligations on companies involved in marine spatial planning (MSP) to minimise bias in data used by AI systems.
Reflecting upon the most important lessons to be learned from the study, Professor Chuah comments: “AI can definitely pick up offshore oil and gas patterns faster than any geologist — but it can’t spot its own blind spots. My concern is that if we let algorithms quietly redraw the marine spatial map, we’d better make sure fishers and coastal communities still hold the pen, not just the eraser.”