The fact that people have spent decades filming the ocean floor—sending down remotely operated vehicles and gathering thousands of hours of footage—and then mostly ignoring that footage is subtly unsettling. It is one of the most comprehensive records of a world we hardly comprehend, sitting on hard drives and archive servers, but only a small portion of it has ever been thoroughly investigated. It can take months to manually review the footage from a single expedition. Most of it just waits.
With Deep Vision, a two-year project that received £2 million from the Bezos Earth Fund earlier this month, Plymouth Marine Laboratory and the University of Plymouth are attempting to address that issue. The money is provided by the Earth Fund’s $100 million AI for Climate and Nature Grand Challenge, which links conservation science with machine learning resources and partners like AWS, Google.org, and NVIDIA. The individuals in charge appear to be aware of the importance of this vote of confidence.
The project’s principal investigator, Professor Kerry Howell, a professor of deep-sea ecology at the University of Plymouth, has about 20 years of experience in this area. She describes the situation with a certain matter-of-factness. “There is an urgent need to map its ecosystems as the deep-sea becomes more accessible for exploitation,” she stated in the announcement. Given what’s truly at risk—cold-water coral reefs, sponge fields, crinoid communities, ecosystems that take centuries to form and can be destroyed in a single trawl pass—the wording is restrained, almost understated.
In order to catalog what marine biologists refer to as vulnerable marine ecosystem indicator species—the organisms whose presence indicates a protected habitat below—the Deep Vision project will apply AI models to thousands of underwater photos collected from Atlantic expeditions. Researchers anticipate that the AI will create a comprehensive, standardized biodiversity map of the deep Atlantic at the scale of an entire ocean basin once it has been trained and put into use. That’s not a modest goal. Perhaps no such attempt has ever been made.
The larger team, which includes representatives from several nations and organizations, including the Institute of Marine Research Norway, the University of Galway, Bielefeld University, the University of Aveiro, and others, demonstrates how cooperative a project of this magnitude must be. Although the exact time needed to train the AI tools to a level of accuracy that satisfies the scientific community is still unknown, it is anticipated that tasks that would require months for human analysts could eventually be completed in a matter of days.
The legal context that surrounds this makes it more than just a research project. Priority conservation zones in international waters must be identified by nations and international organizations in accordance with the High Seas Treaty, which was adopted by the UN but is currently undergoing ratification procedures. Policymakers currently lack the maps necessary to accomplish that in a meaningful way. To provide them with those maps, Deep Vision is being developed. There seems to be a race between the law and science, and this project aims to ensure that science comes first.
It’s difficult to ignore how much the urgency has changed over the last few years when observing this from the outside. Interest in deep-sea mining is rising. International waters continue to have inadequate regulation of fishing pressure. Despite its remote location, the Atlantic floor is becoming more and more disputed. Plymouth’s scientists aren’t the only ones keeping an eye on the development of AI-assisted ocean mapping; organizations like NOAA have been discreetly monitoring it, knowing that whoever develops the best tools first will have a significant impact on conservation policy worldwide for years to come.
Deep Vision is presented as both a stand-alone project and a proof of concept. The team plans to take the strategy worldwide if the Atlantic mapping is successful. Two years, one ocean basin, and a ton of archived video for the time being. It’s a beginning.
