Beyond the Hype: Assessing AI for Ocean Sustainability

Sammanfattning

Artificial intelligence (AI) is increasingly positioned as a "game-changer" and “revolutionary” solution for addressing sustainability challenges. In the oceans specifically, AI applications are emerging rapidly, from machine learning for bycatch monitoring to satellite imagery for detecting marine pollutants. At the same time, the considerable risks of AI deployment are also starting to surface: AI infrastructure is energy- and resource-intensive, the pursuit of ever-larger models concentrates power among a small number of corporate actors, and AI development disproportionately benefits the Global North. This creates a fundamental tension in the AI-for-sustainability discourse — can AI drive deep, meaningful change while addressing the risks associated with its deployment? This study empirically investigates that question through an exploratory examination of 50 AI initiatives deployed in ocean sustainability contexts, assessing both where they intervene in the system and how they are governed. Drawing on Meadows' leverage points framework and a cluster analysis on governance dimensions, it identifies four archetypes: Open Transformers (n=16), Black Box Operators (n=21), Open but Gated (n=6), and Participatory but Opaque (n=7). The results indicate a pattern between governance archetype and depth of systemic intervention: initiatives with high transparency and participatory engagement (Open Transformers) operated at deeper, paradigm-level leverage points, while those with closed, opaque governance (Black Box Operators) concentrated at shallower levels. Accountability is the weakest governance dimension across all archetypes. These findings suggest that governance characteristics co-occur with deeper systemic intervention, pointing to the conditions under which AI might contribute to sustainability transformations.