AI Is Rewiring Fish Farming, But the Biggest Obstacles Are Human, Not Technical

Artificial intelligence is quietly transforming the world’s fastest-growing food sector. Aquaculture now supplies more aquatic food for human consumption than wild-capture fisheries, and a sweeping new review argues that the technology’s future will hinge less on clever algorithms than on people, ethics, and governance. The synthesis, published in Frontiers in Aquaculture, analyzed 220 peer-reviewed publications from 2015 to 2025 and maps out how machine learning, computer vision, and the Internet of Things are reshaping everything from feeding schedules to disease surveillance across the global blue economy.

The numbers behind the transformation are striking. AI-enabled sensor systems have been credited with reducing fish mortality by up to 40 percent and boosting yields by 15 to 50 percent. Computer vision models built on deep learning architectures such as U-Net and Mask R-CNN now estimate biomass in shrimp and Nile tilapia with prediction accuracies exceeding 90 percent, replacing laborious manual sampling with cameras that watch fish around the clock. In turbid water where optical systems fail, acoustic and sonar platforms fused with multiple-object-tracking algorithms maintain roughly 90 percent accuracy for counting fish and detecting leftover feed, a feat that would have been unthinkable a decade ago.

Predictive modeling has become the analytical backbone of precision aquaculture. Random Forest algorithms trained on sparse, noisy sensor data have achieved near-perfect accuracy in forecasting dissolved oxygen, supporting more than 6,000 automated corrective interventions while keeping survival rates above 90 percent. Hybrid CNN-LSTM networks extend this precision to multi-parameter forecasting of temperature, ammonia, and nitrite in recirculating systems. Reinforcement learning controllers, including Q-learning agents, develop adaptive feeding policies that minimize waste and ammonia accumulation, while mechanistic growth models coupled with intelligent controllers have improved growth rates by 13 to 47 percent. Model predictive control frameworks, meanwhile, cut mortality by up to 27 percent compared with conventional controllers, and economic variants have achieved 41 percent improvements in cost efficiency by balancing biomass growth against energy expenditure.

Disease management is undergoing perhaps the most dramatic shift, moving from reactive treatment to predictive prevention. Convolutional neural networks paired with RGB, fluorescence, and ultrasonic imaging can now recognize pathogens such as Ichthyophthirius multifiliis and Flavobacterium columnare with precision rates above 95 percent, compressing detection times from days to minutes. Geospatial machine learning models combine weather, water-quality, and farm-management records to forecast outbreaks of white spot syndrome virus and tilapia lake virus in Southeast Asia before they escalate. In the Philippines, AI-supported advisory platforms have reportedly contributed to reductions of up to 30 percent in prophylactic antibiotic use among shrimp and tilapia farms, a development with major implications for the fight against antimicrobial resistance under a One Health framework linking aquatic, human, and environmental health.

Yet the review’s central message is a cautionary one: technology alone is not enough. Adoption remains constrained by affordability, limited digital literacy, inadequate infrastructure, and stubborn data interoperability barriers. Most farm-level data are siloed in proprietary systems or fragmented across government and private repositories, undermining model transferability. Labeling errors, dataset bias, and class imbalance continue to erode model reliability, and phytoplankton detection studies still suffer from limited class representation, hampering accurate forecasting of harmful algal blooms. The review emphasizes that models trained in controlled settings frequently lose accuracy when deployed across different species, farming systems, and environmental contexts, making open benchmark datasets and standardized evaluation protocols essential.

The digital divide is starkly geographic. High-income economies such as Japan, Singapore, and South Korea are advancing toward AI-integrated digital twins and real-time production analytics, supported by robust broadband networks and skilled labor. By contrast, developing Southeast Asian nations, including the Philippines, Indonesia, and Vietnam, show strong potential but face structural barriers of inconsistent connectivity, limited data infrastructure, and gaps in digital expertise. The review also stresses that AI adoption is highly system-dependent: recirculating aquaculture systems, with their controlled environments and dense sensor data, suit advanced deployment, while the pond-based systems that dominate smallholder farming in Asia face far greater constraints from turbidity, irregular power, and limited automation.

Beyond the farm gate, AI is reshaping supply chains and spatial planning. Deep learning systems detect biofouling and structural defects in aquaculture nets, Bayesian networks forecast food-safety vulnerabilities with roughly 81 percent accuracy, and blockchain-AI architectures promise lot-level traceability in an industry where mislabeling affects nearly 20 percent of global seafood products. On the water, satellite constellations such as Sentinel-2 and Landsat 8, combined with machine learning, now forecast harmful algal blooms and climate hazards with near-real-time accuracy, while GIS-based carrying-capacity models help prevent over-intensification of coastal zones. In Japan, such models identified 74 percent of southern Hokkaido’s shallow marine zones as optimal for kelp cultivation.

Perhaps the most provocative dimension of the review is its treatment of ethics and genomic sovereignty. As environmental DNA, multi-omics datasets, and AI-assisted breeding programs multiply, questions of who owns genetic information, who controls access, and who benefits from resulting innovations have moved to the fore. The author argues that genomic datasets derived from endemic species in biodiversity-rich countries are strategic assets, and that responsible AI-enabled genomics requires recognition of Indigenous biodiversity rights, prior informed consent, and equitable benefit-sharing consistent with the Nagoya Protocol. Algorithmic bias, cybersecurity, privacy, and data ownership more broadly demand adaptive governance mechanisms that keep pace with the technology itself.

Human-centered artificial intelligence emerges as the connective thread. Explainable AI techniques such as SHAP and LIME can reveal why a model predicted a dissolved-oxygen crash or recommended a particular ration, building the trust that farmers need before ceding control to automated systems. The review warns that excessive automation risks diminishing human situational awareness and professional agency, and it proposes an integrated AI-Sustainability-Governance Framework in which innovation is embedded within social, ethical, and policy-responsive systems aligned with the FAO’s Blue Transformation agenda and the OECD AI Principles.

The practical pathway the review outlines is deliberately incremental. Farms should begin with a digital readiness assessment, prioritize high-impact problems such as feed waste and disease risk, pilot tools at small scale, validate them under real conditions including seasonal variability and sensor drift, and scale only when technical reliability, economic value, and governance safeguards are demonstrated. For smallholders, the entry points are low-cost sensors, mobile advisory systems, and cooperative data-sharing; for commercial operations, digital twins and automated feeding control. The author’s conclusion is unambiguous: the long-term sustainability of AI in aquaculture will depend not on algorithms or computing power alone, but on the co-evolution of governance structures, human capacity, and environmental stewardship, ensuring that the blue economy’s digital revolution leaves no farmer behind.

Subject of Research: Artificial intelligence applications, ethics, and governance in sustainable aquaculture

Article Title: Artificial intelligence in aquaculture: human-centered innovation, ethical governance, and data foundations for sustainable blue growth

Article References: Cortes, J. R. (2026). Artificial intelligence in aquaculture: human-centered innovation, ethical governance, and data foundations for sustainable blue growth. Frontiers in Aquaculture, 5, Article 1907758. https://doi.org/10.3389/faquc.2026.1907758

Image Credits: AI Generated

DOI: 10.3389/faquc.2026.1907758

Keywords: artificial intelligence, aquaculture, machine learning, computer vision, explainable AI, blue transformation, fish farming, digital twins, antimicrobial resistance, data governance, sustainability, Internet of Things