Animal Translator? There's An AI For That

Welcome to this week’s Deep-Fried Dive with Fry Guy! In these long-form articles, Fry Guy conducts in-depth analyses of cutting-edge artificial intelligence (AI) developments and developers. Today, Fry Guy dives into AI’s growing ability to help us understand animal communication. We hope you enjoy!

*Notice: We do not receive any monetary compensation from the people and projects we feature in the Sunday Deep-Fried Dives with Fry Guy. We explore these projects and developers solely to showcase interesting and cutting-edge AI developments and uses.*

Have you ever wished you could chat with your cat or ask a whale what it’s singing about? For generations, the idea of talking with animals has been a staple of fairy tales and Dr. Dolittle-style adventures. Thanks to advances in AI, that far-fetched dream is inching closer to reality. Scientists are deploying machine learning algorithms to decode animal communication, from the deep songs of whales to the dances of honeybees.

In today’s exploration, we’ll look at how AI is learning to speak whale, dolphin, bird, bat, bee, and even elephant. We will also have some fun imagining a future where we can talk with the animals around us.

DECIPHERING ANIMAL COMMUNICATION OF ALL KINDS

Ocean animals like whales and dolphins have vastly different lives from ours, yet they are highly social and vocal. Sperm whales, for example, communicate in Morse-code-like clicks called codas. An interdisciplinary team called the Project Cetacean Translation Initiative (CETI) is applying advanced machine learning to thousands of hours of sperm whale recordings. Their goal is bold: train a computer to “learn to speak whale.” In its first phase, CETI is using an enormous dataset of whale sounds and behaviors to decipher the meanings of different clicks​. By recognizing patterns far too subtle for a human ear, the AI can start to decode whale communication—identifying which whale is “speaking” and possibly what the context is.

Researchers have already made some breakthroughs. In one proof of concept, marine biologist Shane Gero’s team fed recordings to a neural network, and it learned to identify individual whales from their voices with 99% accuracy. That’s like picking out a single whale’s “voice” from a crowded ocean. The next steps are even more exciting: scientists hope to generate synthetic whale calls to talk back. The ultimate test is to see if real whales respond to these AI-generated messages as if another whale produced them—a sort of Turing test for interspecies communication. If a whale “talks” to an AI and can’t tell it’s a machine, we’ve effectively built a rudimentary translator! Dolphins are also on the radar; with their complex repertoire of whistles and clicks, they are natural candidates for AI analysis. In fact, an early attempt by researchers involved a handheld device that translated a few invented whistles back and forth with wild dolphins—a very primitive dolphin translator. Modern AI could take this much further by analyzing dolphin speech at scale to find repeating patterns that could be likened to words or grammar.

Not all animal chatter is friendly. Sometimes it’s downright argumentative—at least with bats! In a fascinating (and rather adorable) study, scientists in Israel planted microphones in a cave of Egyptian fruit bats to decode their cacophony of squeaks. To the human ear, a bat colony just sounds like a bunch of shrill squeals, as if they’re all saying “get out of here!” nonstop. But machine learning unveiled a whole soap opera in those noises. Researchers trained an algorithm (originally developed for human voice recognition) on thousands of bat calls that had been linked to specific social interactions on video. The AI learned to tell individual bats apart by voice and even recognize the context of their squabbles. It turns out the bats were arguing, and about very specific things. The algorithm could discern if bats were fighting over food, a prime sleeping spot, or unwanted mating attempts, and it could even predict how the tiff would resolve. In fact, based on just the sound frequencies, the system identified the correct bat and the nature of the argument with around 70% accuracy—and could sometimes figure out who won the argument in the end based on sounds alone!

Whales, dolphins, and bats are just a few of the animals that AI is being used to explore. Other experiments include decoding the language of bees, elephants, birds, and more!

HOW AI LEARNS TO SPEAK ANIMAL

At this point, you might be wondering: How exactly do these AI models figure out what a whale or a bat is saying? The process is a mix of high-tech pattern recognition and old-fashioned behavioral science. Machine learning algorithms, especially modern deep learning models, are extremely good at finding patterns in big data. You feed them hours of animal sounds, and they learn the statistical structures—clustering similar sounds together, distinguishing different call types, etc.

In the case of the bats, for example, researchers fed the algorithm labeled examples: they had tagged recordings where they knew which bat was vocalizing and what the context was (fighting over food, etc.), gleaned from video observation. The AI used this to train a classifier that could then take a new bat call and predict the likely context by comparing its features to the learned patterns. This is classic supervised learning—the same way a speech-recognition AI might be trained on labeled human audio clips.

But what about cases where we don’t know the meanings beforehand (which is most cases, frankly)? That’s where AI’s pattern-finding prowess truly shines. Researchers deploy unsupervised or semi-supervised learning to let the AI find structure in the noise on its own. A great example is the approach taken by the Earth Species Project: they are training large-scale models on massive unlabeled datasets of animal communications​. The idea is to let the AI discover categories of sounds and correlate them with things like time of day, group behavior, or environmental factors. For instance, if the AI notices that a certain dolphin whistle always occurs when a particular dolphin approaches its mother, that might be a clue it’s a contact call or greeting. Projects like Earth Species Project speculate that a similar method could align, say, elephant and human acoustic representations—essentially finding a common conceptual space that links an elephant rumble about water to a human word, “water.” It sounds futuristic, but early “foundation models” like NatureLM are aiming to be generalists that can be fine-tuned to any species. These models treat animal communication a bit like a foreign language to decode, but instead of translating Japanese to French, we might be translating whale sounds to English.

A SPECULATIVE FUTURE: CHATTING WITH CREATURES LARGE AND SMALL

So what might a future of AI-assisted interspecies communication look like? Of course, there is no way of knowing for sure, but we can have a bit of fun imagining it. Maybe in a decade or two, you’ll download “Google Translate: Animal Edition” on your phone. Heading out to snorkel with dolphins? The app might live-transcribe the dolphins’ whistles and tell you “they’re curious about your weird flippers.” You could speak into your device and it would emit an underwater whistle reply that hopefully means, “I come in peace!” Hiking in the woods, your smart earbuds might whisper: “FYI, that bird is warning others you’re coming—it thinks you’re a predator.” Imagine calling your dog on your lunch break, and through an AI translator your phone plays your voice as a series of friendly barks, while the dog’s excited yaps are translated back into a human voice saying, “Bring treats!”

On a more profound note, we might find ourselves forming relationships with wild animals in ways we never could before. Perhaps conservationists will use drones equipped with AI translators to literally talk to whales— maybe warning them of a coming danger or gently herding them away from an oil spill. Park rangers could communicate to an elephant family through low-frequency rumbles generated by a device, essentially saying, “There are people ahead, please turn back,” as a way to prevent human-elephant conflict. As one researcher named Sophie Bushwick put it, these AI tools are like a planetary hearing aid, finally allowing us to listen in on the conversations of the natural world.

Our practical interactions with animals might get interesting. We could negotiate peace treaties with raiding baboons by communicating in baboon terms that there’s an easier food source elsewhere. This raises big questions: How will humans handle the responsibility of being understood by other species? What if we hear things we’d rather not, like orcas telling us to quit overfishing their seas? Perhaps this might change the way we engage with the animals around us.

Now, much of this is speculation of course, but it’s rooted in some cool, real-world discoveries! So perhaps we won’t be having heart-to-hearts with hedgehogs quite yet, but we’re closer than ever to asking animals some questions and getting something like an answer. Maybe the animals have been talking this whole time, and we just didn’t know how to listen. With AI as our translator, the barks and chirps around us might start making a bit more sense.