Bloomberg: AI Crow Decoder Could Aid Poachers, Warn Ethicists

Biologist Vittorio Baglione spent years attaching tiny microphones to the tail feathers of 43 carrion crows. In 2018 he and research partner Daniela Canestrari opened the audio files to find thousands of hours of sound: the crows themselves, their chicks begging for food, even the call of their brood parasite, the great spotted cuckoo. To sort it, they borrowed a machine-learning model from the Earth Species Project, a nonprofit that bills itself as "the new frontier of interspecies understanding."

Baglione's favorite discovery was a soft burst the crows made when a buzzard attacked the nest, always followed by other birds arriving to defend it. He read it as something like "Hey! Come help!"

That, according to a Bloomberg Weekend feature by Morgan Meaker, is where the ethics get uncomfortable. The same tool that lets a scientist recognize a distress call gives a poacher the vocabulary to fake a reassurance. Baglione himself concedes his work could be co-opted by hunters to tell a crow, "Come, it's safe." Poachers have already used automated, looped recordings of bird calls to lure birds into traps.

Meaker's reporting frames a broader worry among bioethicists that AI decoders could give humans new ways to "manipulate, exploit, and harm animals," from luring crows to broadcasting synthetic whale songs across the ocean, echoing the sonar-style tactics once used to drive whales to the surface to be killed. Project CETI, which uses machine learning on the clicks and codas of Caribbean sperm whales, has itself warned that even routine recording and playback can stress the animals. Its founder David Gruber and colleagues at NYU's More than Human Life program have proposed a "PEPP" framework, for Prepare, Engage, Prevent and Protect, as principles for studying animal communication responsibly.

The AI Ethics desk here, which has logged 95 stories in the past 90 days, has been running variants of the same story most weeks this quarter. Meaker leaves the reader on the question Bloomberg posted on X alongside the piece: whether humans, having learned to listen, should answer back.