The harbor at Moss Landing, California, a tiny fishing and research village tucked between Monterey Bay and the Elkhorn Slough wetlands, is not very striking. The R/V David Packard appears more industrial than exploratory, the kind of vessel that emphasizes patient work rather than adventure, as it rests at its dock next to commercial fishing boats.
A few hundred meters away, engineers and biologists at the Monterey Bay Aquarium Research Institute are accomplishing something that oceanographic organizations with much larger budgets and more well-known names haven’t been able to do: they are methodically mapping the deep ocean at scale, discovering thousands of species that were previously unknown to science, and developing the AI tools to process what they find—all without the kind of significant expedition fanfare that usually accompanies announcements of that magnitude.
The methodology used by MBARI reverses nearly every aspect of conventional deep-sea research for the majority of the 20th century. Ships, nets, and tangible specimens were the foundation of traditional oceanography. These tools were used to extract organisms from the deep, preserve them in jars, and transport them to laboratories on land where taxonomists could study them under light. It was acknowledged as a necessary expense of conducting science that the practice caused disruption to both the individual species and the environments from which they were removed.
Given what autonomous underwater vehicles and high-resolution imagery could accomplish in its place, MBARI made the early decision that it was likewise an unnecessary expense. With laser scanners and stereo-imaging cameras that create three-dimensional representations of deep-sea life exactly where it lives and exactly as it acts, their AUVs descend without a research team on board or a support ship on standby. The animals remain. The information appears. Nothing is netted by anyone.
The institute’s idea transitions from intriguing to truly astonishing when it comes to the video annotation challenge. No human crew could evaluate the footage produced by an AUV doing months-long seafloor sweeps in a reasonable amount of time. To directly address this, MBARI developed the Video Annotation and Reference System, or VARS. From material that would otherwise remain on a hard drive for years before anybody could access it, the AI system is trained to recognize and prioritize organisms in deep-sea video, identifying unidentified species, classifying behavior, and creating a searchable database.
A public game called FathomVerse, which was created to allow casual users to identify deep-sea species and generate labeled training data through gameplay, provided some of the training data for VARS. This source is atypical by academic research standards. People playing a phone game about ocean animals while lounging on their couches may have contributed to the creation of the most extensive record of deep-sea biodiversity currently under construction. It’s not a critique. It describes a genuinely innovative solution to an actual data challenge.
Rather than being merely incidental, MBARI’s relative obscurity is intriguing because it shows how public exposure and scientific influence can occasionally be unrelated. Operating out of a little coastal town on the central coast of California, this institute has recorded species that no human has ever seen before. These are organisms that have evolved to such harsh environments that their very existence broadens our understanding of where life can exist.
The seamounts that MBARI’s AUVs have mapped are geological formations that rise kilometers above the abyssal plain. They are home to ecosystems that predate the industrial age by millions of years and were essentially undetectable to science until recently. Robotic survey data from those structures is influencing conservation decisions that will impact policy for decades and altering ecological models.
Looking at what MBARI has created in Moss Landing, there’s a sense that the model reflects something that the larger ocean research community is only now beginning to take in: that the most scientifically fruitful method of deep-ocean exploration might not entail humans entering the ocean at all, at least not in the ways that produce dramatic footage for documentaries.
While it’s a more subdued form of exploration than the crewed submersibles that continue to captivate the public’s attention, autonomous systems operating for months, AI processing the results at scale, and public engagement tools training the machine learning are all working much more quickly.
