bloomberg.com
Talking to Animals Is AI’s Next Ethical Minefield
Morgan Meaker
11–13 minutes
Biologist Vittorio Baglione has spent his career obsessing over a bird that doesn’t want him around. “They don’t like you,” he says of the carrion crow, a cautious creature considered one of western Europe’s cleverest birds. “They don’t like your company.”
For the past 30 years, Baglione has pursued them anyway. As a young researcher, he studied the way carrion crows reared their chicks and patrolled their territory for threats. More recently, in 2018, he began trying to understand what they were saying, attaching tiny microphones to the tail feathers of 43 birds. But when Baglione and Daniela Canestrari, his wife and research partner, opened the audio files, they found thousands of hours of hard-to-define sound: the crows themselves, their chicks begging for food and even the call of their brood parasite, the great spotted cuckoo. Analyzing the vast dataset felt nearly impossible. That’s when they turned to artificial intelligence for help.
That decision catapulted the two biologists into a futuristic world. They borrowed a machine learning model from Earth Species Project, a group describing itself as “the new frontier of interspecies understanding.” The model could differentiate between sounds made by the bird wearing the microphone, its companion crows and other sources entirely. Baglione’s favorite call was the soft burst of sound the crows would make when their nest was under attack by a buzzard. It was always followed by the arrival of others to defend the nest, so he interpreted it as something like, “Hey! Come help!” Today, Baglione and Canestrari have a vast catalog of 150,000 recordings, one of the largest bird datasets in existence.
As rapid advances in AI promise to transform industries from healthcare to education and to shake up the world of work, researchers see the potential for humanity’s relationship with animals to change radically too. The technology could transform our understanding of other species — and put them at greater risk than ever before.
Technology vs. Animals
Humans have been trying to communicate with animals for thousands of years. Bone flutes discovered in Israel and dated at 12,000 years old were probably used to commune with falcons, say researchers. More recently, animal psychologist Irene Pepperberg carried out a 30-year experiment with Alex the African grey parrot, who died in 2007. British anthropologist Jane Goodall became famous, in part, for her claims to speak chimpanzee.
Today’s generation of animal interlocutors believe that simultaneous advances in AI and biologgers — small electronic tags that can be attached to different species to record their noises — are accelerating our understanding in ways not previously possible, potentially enabling humans to talk back. Private equity billionaire Jeremy Coller, a major donor in the field, suggests genuine two-way communication could be achieved within four years. “Because AI is so fast, I have absolute conviction we will crack the code by 2030,” he said in a speech in June.
Most researchers doing this work cite the same motivation: that if humans better understand the interior world of other species, they are less likely to shrug at climate change, endorse industrial farming or participate in habitat loss. Many in the field believe that if AI can be fed giant datasets of animal calls, it will eventually identify patterns that will demystify their meaning.
Long before AI, Irene Pepperberg achieved international acclaim for her 30-year experiment with Alex the parrot (pictured in the middle).Photographer: Rick Friedman/Corbis Historical/Getty Images
Yet human innovation also has a long track record of using technology against animals, says César Rodríguez-Garavito, director of the More-Than-Human Life (MOTH) Program at New York University. Twentieth century whalers used advances in sonar to drive panicked whales to the surface to be killed. Poachers have copied the tribal practice of mimicking bird calls during hunting, using automated and looped recordings to lure birds into traps. Baglione himself acknowledges that his research could be co-opted by hunters to communicate to a crow, “Come, it’s safe.”
“AI has potential for greater harm at scale,” says Rodríguez-Garavito. In a 2025 report, he and his co-authors presented a dystopian vision of how “nonhuman animal communication technologies (NACTs)” could expose animals to mobs of humans wanting to chat, leverage them for military purposes like sea mine detection or even be used to capture and kill them.
Concern is also growing about the harm humans can cause even by simply pursuing greater interspecies understanding. The promise of AI-enabled animal translation is already creating a rush for data, alarming animal-rights activists. Researchers are dropping hydrophones into water to listen to whales, attaching biologgers to wolves and designing backpack microphones for birds. It’s enough for some to call for a version of privacy rights for animals.
The AI Risk
There will never be an exact equivalent of Google Translate for animals. Where humans largely rely on words, animal language stretches across smell, color, movement and sonar. In 1973, Austrian zoologist Karl von Frisch won a Nobel Prize for his work documenting how bees direct each other to nectar using dance. For that reason, some researchers need video as well as audio to understand their animal subjects.
This summer, Baglione and Canestrari spent 14-hour days monitoring carrion crows in their own territory in the farmlands of northern Spain. From a small, camouflaged tent they recorded video and narrated for the AI’s benefit, documenting exactly what was happening around the nest. The hope was that this would enable algorithms to later establish correlations between sounds and their context. Before that, the biologists, both based at Spain’s University of León, spent nine months annotating 7,000 crow sounds to teach their AI model how to differentiate between the birds’ subtly different noises. “It costs a lot of effort,” Baglione says of trying to satisfy AI’s appetite for information. In three decades of research, Baglione’s fieldwork regime has never been so intense.
As the couple work toward turning their dataset into a type of crow dictionary, eventually they will have to test whether their assumptions about the calls’ meanings are correct. In the world of animal communications, the most common way to do this is by using a controversial technique called “playback” — the practice of broadcasting recorded animal calls to another animal.
A listening station off the coast of Dominica operated by Project CETI, an organization that uses machine learning and robotics to study sperm whale communication.Source: Project CETI
Like others, Baglione is worried about how playback experiments affect their subjects. “You can stress the animals. You can disrupt their social environment. You can pose conflicts between them, especially if you don’t really understand the meaning of the call,” he says. Notable examples include the zoologist Katy Payne’s admission on a 2007 podcast that she played the voice of a dead elephant to the elephant’s granddaughter, only to see her respond with “paroxysms of groaning and roaring.” Others have cautioned that playing recordings of an individual animal repeatedly to its group has the same effect as releasing a deepfake in its likeness, possibly prompting other members to ignore its genuine attempts to communicate.
“I’m worried about playback,” says Rodríguez-Garavito, noting that he doesn’t believe its advocates have done enough to prove it doesn’t distress their subjects. Existing ethical guidelines feel out of date in a world with AI, he adds. “No one really envisaged the intensive use of playback that AI could potentially enable.”
Among those working toward interspecies communication, there is a divide between those who say we should be exclusively listening to the animals and those who argue we should be striving to talk back. “It matters because if we’re talking back, we’re introducing ourselves into the animals’ lives,” says Kristin Andrews, a philosophy professor and specialist in animal minds at the CUNY Graduate Center in New York.
In South Africa, researchers have been carrying out playback experiments with wild African striped mice.Source: Léo Perrier and Nicolas Mathevon
In June, finalists for the Coller Doolittle Challenge, an annual $100,000 prize funded by Jeremy Coller, showed off photos and videos of their research at the award’s annual ceremony. Playback featured heavily. Wild mice were surrounded by speakers broadcasting sounds inaudible to humans. There was the chimpanzee who jumped, startled, when he heard the call of another chimpanzee played from a speaker hidden in the long grass. Then there was the zebra finch, jumping around frantically on its perch, ostensibly communicating with the cheeping animation projected onto a screen in its cage.
The prize is incentivizing an evolution of playback that is more automated and capable of interacting with animals. “One of the things the Coller Doolittle challenge is trying to do is the idea that there will be an automatic machine maybe that can have a kind of conversation with some animals,” explains Yossi Yovel, chair of the challenge and a zoologist at Tel Aviv University. (Through his foundation, Coller declined multiple requests to be interviewed for this story.)
Alongside the yearly prize for significant progress, the Challenge is also offering a $10 million reward for a project that enables an animal to communicate with researchers without recognizing it is communicating with humans — a criterion inspired by the Turing test. “We don’t know exactly what we want there,” says Yovel, whose own work involves sedating bats, putting them in an MRI machine and playing them recordings of other bats as they come back into consciousness. “The idea is to give a big prize to somebody who really goes far, maybe somebody who develops this fully autonomous device.”
Yovel is aware of the ethical concerns around playback and the direction interspecies communication research is taking. Yet he says that in the two years since its inception, the challenge has yet to receive any entries using research techniques he finds ethically troubling. “There’s more potential benefit than harm, and that’s what we should think about,” he says.
Back in Spain, Baglione doesn’t know where exactly his research will take him or what the long-term impact will be on his beloved birds. “Any advance will be great,” he says, “unless people will start using this knowledge against the animal.” Despite that concern, he plans to continue.
Andrews similarly believes conversation via playback will be necessary to really understand what and how animals think, not just whether the sounds they make refer to predators or food.
“I don’t think you’re going to do any decoding without engaging in two-way communication — or what’s going to be perceived as two-way communication from the perspective of the animals,” she says.
For that reason, Andrews suggests playback experiments should be limited to animals that enjoy human company, such as bottlenose dolphins that are free to swim away but voluntarily spend time with researchers. “They have chosen a bicultural life.”
