In Conversation with
Rocío Joo
Using satellite data and machine learning, Global Fishing Watch is working to reveal human rights abuses in the fishing industry—and help authorities take action
The open ocean remains one of the most challenging places on Earth for human rights enforcement. Covering more than 70 percent of the planet’s surface, its sheer immensity can render it invisible to the human eye — a data void where transparency and accountability are far from reach.
“It’s notoriously difficult to know what happens at sea,” says Rocio Joo, a senior data scientist with Global Fishing Watch’s research and innovation team. “And that is especially true aboard fishing vessels operating far from shore.”
Sitting at the intersection of artificial intelligence (AI) and human rights, Joo’s role as human rights researcher at Global Fishing Watch has taken on increasing importance amid a growing wave of evidence pointing to human rights violations at sea. Indeed, it is estimated that some 128,000 fishers are trapped in forced labor aboard fishing vessels scattered across the planet. Stuck in isolated waters, these fishers work in intensely hazardous and remote conditions, and are subjected to exploitation that can range from forced labor and debt bondage to human trafficking.
The big challenge, notes Joo, is in figuring out which vessels are most at risk for these violations. Commercial fishing vessels can remain at sea for months, even years, operating beyond the reach of authorities and far from public scrutiny. This, in turn, leaves much of the exploitation unseen and undocumented.
“Without a clear picture of what’s happening on the ocean, we simply cannot effectively manage maritime activities or take smart action to protect the ocean itself and those who earn their livelihoods from it,” Joo adds. “Our work at Global Fishing Watch is to harness satellite data and advanced AI algorithms so we can craft a more complete picture of the scale and type of human activity occurring out on the ocean. Only in this way can we clamp down on the human rights violations that are occurring.”
We spoke with Rocio Joo to learn more about how Global Fishing Watch’s machine learning algorithms and network models can help predict and understand a vessel’s risk of forced labor involvement and contribute to the fight for human rights at sea.
How is Global Fishing Watch leveraging artificial intelligence to detect potential human trafficking on fishing vessels?
As part of our human rights work, we have developed—and are continuously refining—a machine learning algorithm to identify fishing vessels likely associated with forced labor based on movement patterns and vessel characteristics. The model is trained on reported cases of forced labor. Some of these involve human trafficking, but that is not always the case and certainly not the focus of the model.
A second type of model that we are developing is concentrated on the networks of encounters at sea related specifically to forced labor. We know forced labor on a vessel can be enabled by other vessels through the provision of food and fuel or by receiving catch and the transfer of crew. Indeed, this last element could include potential incidents of human trafficking. To craft our model, we use large volumes of data on encounters at sea and apply techniques inspired by social networks—like statistical models called Stochastic Block Models and centrality metrics— to study the connections between vessels that might be participating, either directly or indirectly, in forced labor. By exposing these hidden connections, we want to shine a light on the broader system that enables labor abuses and also provide authorities and civil society with a clearer picture of the different actors involved.
What specific data points or patterns does AI analyze to flag suspicious activity at sea?
Movement patterns alone may not serve as definitive proof of illegal activity, but they can help raise red flags for suspicious behavior. For example, an unusually high number of fishing hours per day could signal excessive working hours. When combined with extended trips lasting months or even a year, large gaps in automatic identification system (AIS) transmissions and minimal port visits, these patterns could suggest a heightened risk of labor abuse.
At-sea encounters further enable vessels to extend their time offshore and away from the scrutiny of authorities. In these instances, at-sea transfers of food, fuel or catch allow vessels to avoid port calls, reducing oversight and increasing the risk of labor violations and exploitation. Moreover, interactions with vessels already reported for forced labor could indicate crew transfers, raising the possibility of human trafficking. Additional risk indicators, such as type of fishery (e.g. fishing gear) or other vessel characteristics, are also factored into these patterns.
While none of these factors alone constitute proof of labor abuses or human rights violations, they provide critical intelligence for identifying possible high-risk vessels and bolstering transparency and oversight at sea.
Can AI-driven monitoring tools differentiate between normal fishing operations and signs of forced labor or human trafficking?
Yes, but there are some caveats. It is important to remember that AI and machine learning are only as effective as the data that fuels them. To train our model to detect forced labor at sea, we rely on reported cases, teaching the system to recognize vessel movement patterns associated with labor abuse. As a check on accuracy, we compared our model’s predictions with a confidential list of vessels certified under the International Labour Organization’s C188 Work in Fishing Convention. Since there’s no absolute guarantee that any vessel is entirely free of forced labor, we couldn’t use this list for training. But we did use it to test the model’s consistency and the results were promising: our latest version predicted more than 90 percent of certified vessels as negative for forced labor and more than 90 percent of reported forced labor vessels as positive. This suggests the model is effective at identifying high-risk cases—but only for patterns similar to those it has been trained on. And that’s why expanding our dataset with more diverse and verified cases is critical: the more examples we feed the system, the better it becomes at uncovering hidden abuses on the high seas.
How does AI-enhanced vessel tracking support law enforcement and policymakers in combating human rights abuses in the fishing industry?
AI-powered vessel tracking has the potential to revolutionize monitoring, control and surveillance efforts, offering a crucial tool in the fight against human rights and labor violations at sea. By analyzing vessel movements and identifying suspicious patterns, our models can help port and coast guard inspectors prioritize which ships to examine—an essential advantage in an industry where enforcement resources may be stretched thin. Rather than relying on random checks, authorities can use data-driven insights to focus on high-risk vessels, improving efficiency and accountability. Beyond enforcement, governments can also strengthen protections for workers at sea by aligning with international legal frameworks and implementing national regulations that leverage these advanced tracking tools to monitor vessel behavior more effectively.
What challenges remain in using AI for human trafficking detection, and how can technology continue to evolve to address them?
The effectiveness of AI-driven methods hinges largely on the quality and diversity of the data used to train the model. If we rely too heavily on cases from a single fishery, the model may become highly accurate for that specific context but unreliable elsewhere. We work to mitigate this bias using fairness metrics, but even these assessments are limited by the data we can collect. The more cases we gather, the more refined and informative the model becomes.
Another challenge lies in the perception of AI predictions as absolute truth. AI and quantitative tools are designed to enhance transparency at sea and guide authorities toward areas that require closer scrutiny. If a model flags a vessel as high-risk for forced labor, that should prompt inspections, document reviews and verification of crew contracts, health insurance and declarations on the number of crew shifts. For instance, long periods at sea with extensive fishing hours could be a red flag unless multiple crew shifts are in place to distribute the workload.
Right now, one of the greatest obstacles is the near-total absence of human rights and labor inspections in the fisheries sector. Governments are failing to treat this issue with the urgency it demands, and port inspectors often lack the training needed to conduct labor-related inspections. Tackling forced labor at sea requires a coordinated, multi-sector and multinational approach engaging labor and fisheries departments, flag states, port states, coastal states and the home countries of crew members. Without this level of collaboration, real progress will remain out of reach.
Where do you hope to see Global Fishing Watch technology in a few years vis-a-vis detecting human rights abuses?
At Global Fishing Watch, we’re working hard to integrate a labor insights tab into our Vessel Viewer tool. My hope is that it can one day be included in the various capacity building activities we conduct with countries around the world. Crucially, I would also like to see it used by governments in their efforts to inspect vessels; the seafood industry in their efforts to increase scrutiny in their supply chain; labor unions as they mobilize to support fishing crews; journalists and civil society actors as they expose labor abuses and inform the public.
Looking further ahead, I hope to see more inspections, fewer cases of exploitation, and ultimately, a world where tools like these are no longer necessary because labor conditions at sea will have finally met the standard of decency they deserve.
