An AI biologist
Screenshot_20250129-061706.png
About ten miles off the coast of California on a foggy October morning, a crane lifts a boxy yellow robot from the deck of the research vessel Rachel Carson and lowers it into the murky, metallic-gray waters of Monterey Bay. The remotely operated vehicle, packed with cameras and lights, remains tethered to the ship by a detachable cable, but the artificial intelligence has taken on a will of its own.
The robot, called a MiniROV, broadcasts images of a rarely seen jellyfish onto a wall of screens lining a cramped control room aboard the ship. Kakani Katija, a senior engineer at the Monterey Bay Aquarium Research Institute, lightly gripped a pair of levers, maneuvering the MiniROV closer to the translucent creature swimming in a marine blizzard of organic particles falling to the sea floor.
MBARI researchers working with the MiniROV aboard the RV Rachel Carson in late October. Photo: Rachel Bujalski/Bloomberg
Then Paul Roberts, a senior electrical engineer, presses a button on a laptop and announces, "We're starting to track the agent."
The "Agent" is an AI program built into the MiniROV's control algorithms. It's the first time it's been deployed in the ocean, allowing the robot to autonomously locate and track marine organisms. Katija lets go of the controls and smiles as the robot begins to chase the square-shaped jellyfish on its own, using thrusters to keep up with the animal's pace as it moves away.
To build a network of AI-powered ocean-surveillance robots, MBARI researchers are using citizen scientists to play a game that can train the machines much faster than a small group of researchers. The need for speed is clear: the ocean is the world's defense against climate change, and marine organisms play a vital role in recycling carbon from the atmosphere. Thanks to advances in robotics, underwater cameras, and sensor technology, researchers have amassed millions of images of sea life, but most of the creatures remain unknown. Indeed, classifying them could take years, limited by the availability of overworked taxonomies. The jellyfish that the MiniROV is tracking was first discovered in 1990, but it wasn't identified until 2003 as a new species, Stellamedusa ventana. The game for phones and tablets, called FathomVerse, fills a virtual ocean with images of sea creatures in their underwater habitats, stored in a vast database called FathomNet. Some of the photos feature marine animals whose identities have been verified by scientists. Others are AI-tagged organisms or have yet to be classified.
After players train as amateur marine biologists, they embark on missions by moving along ocean currents in search of pulsating dots that indicate where marine life has been recorded. Players tap the screen to see the animal and identify whether it is familiar or marked as unknown. The game then reveals whether their choices match the consensus of other players or if the creature remains undecided.
They earn points for correct classifications, as well as the number of organisms they distinguish. Players also receive bonus points if they correctly label a previously unidentified life form when a consensus is reached.
In the background, researchers check the players' consensus results and compare them with the AI classification. Chris Jackett, a research scientist at the Commonwealth Scientific and Industrial Research Organization of Australia, says games like FathomVerse could play an important role in training AI.
"The human ability to recognize patterns and identify unusual features remains unmatched, and having multiple observers looking at the same images helps create powerful training datasets," says Jackett, who works on AI detection and identification of corals and other marine species.
[attachment deleted by admin]