In an effort to rein in illicit fishing, researchers have unveiled a new open-source AI model that can accurately identify what virtually all of the world’s seafaring vessels are doing, including whether a boat is potentially fishing illegally.

Atlantes is a 4.7M parameter transformer-based model trained on NVIDIA H100 Tensor Core GPUs and PyTorch. For training data, the Ai2 team used Automatic Identification System (AIS) messages—which nearly every vessel is required to broadcast—from January 2022 through June 2024.  To help fine-tune the model and improve its ability to identify a vessel’s activity, Ai2 hired 20 maritime experts to annotate location data encompassing more than 15 million location-related signals. 

Seattle-based Ai2 (the Allen Institute for AI) recently released a lightweight model named Atlantes to analyze more than five billion GPS signals a day emanating from the world’s nearly 600,000 ocean-going vessels. 

The model can predict with around 80% accuracy what virtually any vessel is doing at any moment. 

If a vessel appears to be illicitly fishing, the model, which is integrated into Ai2’s larger maritime monitoring platform, Skylight, can send alerts to coast guards, patrolling vessels, or other similar users.

Earlier this year, Argentina’s Navy interdicted—and later fined—a vessel illegally fishing in its exclusive economic zone after Skylight alerted it to the vessel’s suspicious behavior. 

The model alerts authorities within around 15-minutes of first detecting potentially illicit activity. That kind of precision and agility—considering 71% of the Earth’s surface, or 139 million square miles, is covered by oceans—is akin to routinely identifying a microscopic needle in a haystack, at scale.

The Skylight platform can process nearly two terabytes of information a day. That data includes radio frequency, satellite information—including publicly available data from NASA and the European Space Agency’s Sentinel-2—and, critically, the unique AIS location data nearly every ship broadcasts. 

Skylight is accessible on the web and also as an API-based service, and is offered for free to governments and conservationists . 

The stakes are high for addressing illegal, unreported, and unregulated (IUU) fishing, especially in poorer countries. 

Each year, IUU fishing represents as much as $23 billion in global fishery losses—or about 20% of the world’s fisheries catch, according to the Financial Transparency Coalition (FTC). 

Nearly half of all vessels engaged in IUU fishing operate in African waters, where local populations disproportionately rely on fishing for work, and as their primary source of protein, the FTC reported. 

“The purpose of what we’re doing is to help under-resourced nations protect their waters, their natural resources, and their livelihoods,” said Namrata Kolla, Ai2’s program lead for Skylight. “AI has made it possible for enforcement agencies to leapfrog into capabilities they traditionally couldn’t access, like identifying important behaviors within massive troves of datasets.”

Ai2’s ambitions for Atlantes don’t stop at the water’s edge. 

Going forward, Ai2 plans to adapt Atlantes and integrate it into another of its conservation platforms, EarthRanger. 

EarthRanger aggregates vast amounts of data from a variety of sources, including satellites, terrestrial camera traps, acoustic sensors, and animals—like elephants, rhinos, and wild dogs—fitted with tracking devices.

It offers conservation authorities a birds-eye view of relevant wildlife-related data overlaid on maps—including wildlife sanctuaries and national borders—and sends alerts if it recognizes worrying activity. 

In the near future, Ai2 plans to train its Atlantes model to predict elephant behavior, to help reduce the instances of human-elephant conflict. 

Jes Lefcourt, EarthRanger’s director, noted that human-wildlife conflict is now one of the biggest threats elephants face. Poachers still target elephants for their tusks. But increasing numbers of elephants die each year in clashes with farmers over crops.

Through its decades of work in the global conservation space, Ai2 has worked with partners to gather the world’s largest dataset of elephant movements. Today, it continuously tracks tens of thousands of elephants collared with tracking devices. Because elephants typically move in herds, this data represents about a third of the world’s elephant population.

“We’re going to use all of this elephant movement data, combined with historic data of where human-wildlife conflicts have occurred, and train the model to answer: ‘What does an elephant’s movements look like leading up to a human wildlife conflict event?’” Lefcourt said. 

Armed with that precise elephant data, conservationists and local authorities can then work with farmers and communities to interdict elephants before they engage with humans.

“Amazingly, the same infrastructure that’s used for classifying fishing vessels is the same infrastructure we’re going to use to predict elephant behavior, and try to help save elephants’ lives,” Lefcourt said. 

Read more about the Skylight and the Atlantes transformer model and check out additional information about EarthRanger’s real-time alerts.

AI Model Offers Conservationists New Tools to Protect Fisheries, Wildlife at Scale

In an effort to rein in illicit fishing, researchers have unveiled a new open-source AI model that can accurately identify what virtually all of the world’s seafaring vessels are doing, including whether a boat is potentially fishing illegally.

Atlantes is a 4.7M parameter transformer-based model trained on NVIDIA H100 Tensor Core GPUs and PyTorch. For training data, the Ai2 team used Automatic Identification System (AIS) messages—which nearly every vessel is required to broadcast—from January 2022 through June 2024.  To help fine-tune the model and improve its ability to identify a vessel’s activity, Ai2 hired 20 maritime experts to annotate location data encompassing more than 15 million location-related signals. 

Seattle-based Ai2 (the Allen Institute for AI) recently released a lightweight model named Atlantes to analyze more than five billion GPS signals a day emanating from the world’s nearly 600,000 ocean-going vessels. 

The model can predict with around 80% accuracy what virtually any vessel is doing at any moment. 

If a vessel appears to be illicitly fishing, the model, which is integrated into Ai2’s larger maritime monitoring platform, Skylight, can send alerts to coast guards, patrolling vessels, or other similar users.

Earlier this year, Argentina’s Navy interdicted—and later fined—a vessel illegally fishing in its exclusive economic zone after Skylight alerted it to the vessel’s suspicious behavior. 

The model alerts authorities within around 15-minutes of first detecting potentially illicit activity. That kind of precision and agility—considering 71% of the Earth’s surface, or 139 million square miles, is covered by oceans—is akin to routinely identifying a microscopic needle in a haystack, at scale.

The Skylight platform can process nearly two terabytes of information a day. That data includes radio frequency, satellite information—including publicly available data from NASA and the European Space Agency’s Sentinel-2—and, critically, the unique AIS location data nearly every ship broadcasts. 

Skylight is accessible on the web and also as an API-based service, and is offered for free to governments and conservationists . 

The stakes are high for addressing illegal, unreported, and unregulated (IUU) fishing, especially in poorer countries. 

Each year, IUU fishing represents as much as $23 billion in global fishery losses—or about 20% of the world’s fisheries catch, according to the Financial Transparency Coalition (FTC). 

Nearly half of all vessels engaged in IUU fishing operate in African waters, where local populations disproportionately rely on fishing for work, and as their primary source of protein, the FTC reported. 

“The purpose of what we’re doing is to help under-resourced nations protect their waters, their natural resources, and their livelihoods,” said Namrata Kolla, Ai2’s program lead for Skylight. “AI has made it possible for enforcement agencies to leapfrog into capabilities they traditionally couldn’t access, like identifying important behaviors within massive troves of datasets.”

Ai2’s ambitions for Atlantes don’t stop at the water’s edge. 

Going forward, Ai2 plans to adapt Atlantes and integrate it into another of its conservation platforms, EarthRanger. 

EarthRanger aggregates vast amounts of data from a variety of sources, including satellites, terrestrial camera traps, acoustic sensors, and animals—like elephants, rhinos, and wild dogs—fitted with tracking devices.

It offers conservation authorities a birds-eye view of relevant wildlife-related data overlaid on maps—including wildlife sanctuaries and national borders—and sends alerts if it recognizes worrying activity. 

In the near future, Ai2 plans to train its Atlantes model to predict elephant behavior, to help reduce the instances of human-elephant conflict. 

Jes Lefcourt, EarthRanger’s director, noted that human-wildlife conflict is now one of the biggest threats elephants face. Poachers still target elephants for their tusks. But increasing numbers of elephants die each year in clashes with farmers over crops.

Through its decades of work in the global conservation space, Ai2 has worked with partners to gather the world’s largest dataset of elephant movements. Today, it continuously tracks tens of thousands of elephants collared with tracking devices. Because elephants typically move in herds, this data represents about a third of the world’s elephant population.

“We’re going to use all of this elephant movement data, combined with historic data of where human-wildlife conflicts have occurred, and train the model to answer: ‘What does an elephant’s movements look like leading up to a human wildlife conflict event?’” Lefcourt said. 

Armed with that precise elephant data, conservationists and local authorities can then work with farmers and communities to interdict elephants before they engage with humans.

“Amazingly, the same infrastructure that’s used for classifying fishing vessels is the same infrastructure we’re going to use to predict elephant behavior, and try to help save elephants’ lives,” Lefcourt said. 

Read more about the Skylight and the Atlantes transformer model and check out additional information about EarthRanger’s real-time alerts.