Using babies, Israeli study reveals why AI fails to truly understand animal calls

SEPTEMBER 26, 2026 07:40Humans have always wanted to talk with animals.Dog owners try to understand their pets. At the same time, parents make great efforts to communicate with their babies before they start to produce words.King Solomon is said to have done it, and so did Adam and Eve.Others who accomplished it were a variety of fictional figures – Dr. Dolittle (Hugh Lofting’s physician who learned animal language from his parrot); Tarzan (Edgar Rice Burroughs’s iconic lord of the jungle who communicated with various jungle beasts); and Mowgli (Rudyard Kipling’s The Jungle Book hero who grew up with wolves).“Talking with other creatures would not only make it possible to empathize with animals, learn about other levels of consciousness, and enrich our understanding of life. It would also have immense practical value,” suggested Prof. Yosef Yovel, who has worked in the zoology department of Tel Aviv University (TAU) for the last 15 years.Tel Aviv University (credit: TEL AVIV UNIVERSITY)“Farmers might benefit from a machine that allows them to ask a cow about its health, though it might turn them into vegetarians,” he continued.“Such a machine would also come in handy once we land on a planet inhabited by some simple alien life form, while if advanced aliens ever make it to Earth, they will probably already have such a machine.”Even as artificial intelligence masters human language, a groundbreaking study he and his team conducted reveals a fundamental flaw in how AI decodes the animal kingdom: it hears the volume and pitch, but entirely misses the point.The researchers have now discovered why AI is getting animal and human toddler communication all wrong.For years, scientists have turned to AI to unlock the secrets of bird calls, bat chatter, and whale songs, but the new research published in Current Biology under the title “The challenge of decoding animal communication using AI” suggests our high-tech translators might be looking at the problem completely backward.The TAU experts explained that while AI is a powerful tool, it needs a paradigm shift – moving away from raw audio analysis toward understanding neural and perceptual meaning.The study was conducted by Mor Taub, Inbal Arnon, Amiyaal Ilany, Mirjam Knörnschild, and Yoav Ram, and it was supervised by Yovel.The team also included scientists from the Hebrew University of Jerusalem, the University of Edinburgh, the Natural History Museum-Leibniz Institute for Evolution and Biodiversity Science, and Humboldt University (both in Berlin).Two sounds that look completely different on a spectrogram can carry the same message to a receiver, while the same acoustic sound can mean entirely different things depending on the context – proving that communication relies heavily on the receiver’s perception, not just the physical properties of the sound wave.A spectrogram is a visual representation of the spectrum of frequencies of a signal as it changes over time. When applied to audio, it is essentially a picture of sound often referred to as a sonograph, voiceprint, or spectral display.A standard 2D spectrogram uses three dimensions of data mapped onto a flat image.Misleading communication systemThe researchers found that acoustically similar sounds do not necessarily carry similar meanings, while sounds that appear different may convey the same information to the receiver.Therefore, classifying sounds by their acoustic similarity, as most studies do, may create a misleading picture of the communication system and the meaning of the messages it conveys, Yovel added.They discovered that humans listening to a toddler’s sequences perceive increasing urgency, apparently from subtle acoustic modifications. None of the tested AI models captured the ordering of those vocalizations according to urgency.“In recent years, there has been growing excitement about the possibility of using AI to decode animal communication, but our study shows that these promises should be treated with caution,” said Yovel, who earned his bachelor’s degree in physics and biology and his master’s degree in neurobiology from TAU, followed by a PhD in biology from Germany’s University of Tübingen.“Identifying acoustic patterns is not necessarily the same as deciphering meaning: to understand what an animal is ‘saying,’ we need to know how the animal receiving the message perceives it and responds to it,” he added.“The path toward truly deciphering animal communication will require a combination of AI, behavioral observations, experiments, and research into the nervous system. Artificial intelligence is a powerful tool, but it is no substitute for the perspective of the animal itself.”To investigate the problem, the team used a unique communication system – the vocalizations of human toddlers who have not yet fully developed speech.Unlike animal vocalizations, in this case the researchers could try to know how the humans to whom the vocalizations are directed interpret them.The recordings included vocalizations made in three contexts: distress, calling to a specific person (the mother or the father), and asking for food.The researchers analyzed the recordings using a classical acoustic method and two state-of-the-art deep neural networks: one trained on animal vocalizations and another trained on adult human speech.The models were asked to group the vocalizations according to their characteristics.The results showed that the deep neural networks performed better than the classical acoustic method, but even they failed to classify the toddlers’ vocalizations by their meaning.In some cases, they grouped together vocalizations carrying different messages; in others, they separated different vocalizations intended to convey the same message.The models also failed to identify how a sequence of vocalizations expressed increasing urgency, a distinction that the human ear perceives naturally.According to the researchers, reliably deciphering animal communication will require combining AI tools with behavioral observations, playback experiments, and sometimes measurements of brain activity.PROF. YOSEF YOVEL (credit: TAU)Using contextEvery species has its own unique perceptual world, and understanding what animals are “saying” therefore requires more than analyzing sound alone: it also requires examining how they hear the sound and respond to it.“We cannot connect specific signals with meaning. The basic idea is that AI models will find the largest differences between the signals, but these are not necessarily the important differences for the brain,” Yovel explained.“‏We can take an example from the stickleback fish during the breeding season. Males have a red abdomen and are very aggressive. The red color is, for them, the most important signal for detecting another male. They don’t care about the shape of the fish or how it moves.“Researchers showed long ago that if they take a sphere and color code it in red, the fish will perceive it as another fish. AI models would never detect this unless we specifically ask them to pay attention to color.“This is exactly the point with baby and other animal vocalizations. AI will look for the most significant acoustic difference, but these are not necessarily equivalent to the important differences for the brain.“The human brain is using context. AI language models can learn to do this too with human language, but it’s far more difficult with non-human language.”When asked why they chose pre-linguistic toddlers rather than an animal species whose communication they already studied, Yovel explained that “the reason was that we wanted to know something about the meaning, so we studied.”“I thought I knew something about the meaning of their vocalizations. With bats, I needed to guess what they received, but my brain understood human toddler vocalizations,” he continued.“I can’t be sure, but when my toddler calls for me, I am more certain what he wants. The sample size was small, but we only intended to use this as a case to highlight and exemplify the problem. We also used a data set of real human words and showed that the result was completely different.”Asked if a toddler could produce acoustically similar sounds for completely different purposes depending on the context, he agreed: “This is exactly one of the big challenges in decoding animal communication because animals will do this as well.”There has been enormous excitement about AI potentially decoding whales, bats, birds, and other animals.“Nearly all of these studies are finding patterns in animal communication. That’s the first step, but we are very far from understanding meaning or even knowing how to do that exactly.”Follow us on Google