NEW YORK — This image shows the perspective of a fish created with artificial intelligence as it detects a virtual object at varying distances. The AI model, which appears in a study in today’s issue of Current Biology, explores how a species of real fish uses electric fields to scan the world around them. By capturing how this relatively simple sense could work, the research demonstrates the potential for artificial intelligence to help us understand our brain’s more complicated senses, such as vision.
Weakly electric fish scan the world by generating electric fields around themselves. Distortions in these fields reveal the size, distance, conductivity and other properties of nearby objects. This can help the fish detect if these objects are, say, rotting logs or juicy worms.
The scientists found the AI performed best if it first broadly scanned its entire surroundings to find out roughly where objects were and then focused in on those items.
"A good equivalent to this would be the children’s book series Where’s Waldo," said study first author Denis Turcu, PhD, formerly a graduate student co-advised by Larry Abbott, PhD, and Nathaniel Sawtell, PhD, and currently a postdoctoral fellow with the Allen Institute. "If you have a huge scene, it might be hard to find him, but if someone tells you to look in this little square over there, you'll find him much faster and more reliably."
Dr. Turcu says that the next step for this two-step model would be testing it using real fish.
