A Step Toward Dialogue With Animals: What AI Already Can Do

Artificial intelligence is increasingly entering a field that until recently was considered an exclusively human privilege – understanding language. But today scientists are increasingly asking: what if language is not unique to humans? And if animals have it, can modern neural networks finally decode it?

Over the past few years, several major scientific projects have reported significant progress in analyzing animal communication. Researchers are already using AI to study signals from mice, chimpanzees, bonobos, dolphins, sperm whales, birds, and even cuttlefish. Many scientists no longer hide that the ultimate goal of these studies is far more ambitious than simple sound classification. It is about creating full two-way communication between humans and animals.

One of the most impressive projects was the study of African striped mice in the semi-desert of the Karoo in southern Africa. A team led by Professor Nicolas Mathevon from the University of Saint-Etienne installed a network of 23 high-sensitivity microphones around several mouse colonies. Over twelve days, more than 122,000 ultrasonic signals were recorded – sounds that the human ear cannot hear. The volume of data was so large that without artificial intelligence it would have taken years to process. The neural network discovered something previously considered almost impossible to prove. It turned out that each mouse colony has its own acoustic identity, and individual animals have unique vocal signatures that allow others to recognize specific individuals.

When researchers played recordings of foreign animals through speakers, mice reacted differently. Unknown sounds caused alertness, neighbor signals triggered anxiety, while sounds from their own group were mostly ignored. This became one of the first pieces of evidence that even such small mammals are capable of distinguishing complex social signals.

Equally remarkable results were obtained in the study of chimpanzees. For more than twenty years, Professor Catherine Crockford has been collecting behavioral recordings of wild primates in Taï National Park in Côte d’Ivoire. Today the database includes around 20,000 hours of observations covering several generations of animals.

Using machine learning methods, scientists concluded that chimpanzees are capable of combining a limited set of basic sounds to create new meanings. This is essentially a form of proto-syntax. Some sound combinations are used only in specific situations and convey information that is not present in the individual elements of the signal.

Even further progress has been made in studies of bonobos in the tropical forests of Congo. Analysis of hundreds of contextual parameters revealed signals used exclusively for social conflict resolution. In essence, some sound combinations resemble primitive calls for reconciliation and de-escalation of aggression.

Particular interest is drawn to marine mammals. The CETI (Cetacean Translation Initiative) project uses robotics, underwater drones, and AI systems to decode sperm whale communication. Researchers have already identified complex sequences of clicks that show structures remotely similar to elements of human speech. In parallel, the DolphinGemma project uses language models to analyze dolphin whistles. Some signals function like individual names or call signs. Each dolphin has its own unique whistle by which it is recognized by others in the group.

The Earth Species Project aims to go even further. The organization is building universal machine learning models that simultaneously analyze communication across birds, whales, monkeys, and other animals, searching for shared principles of information transfer regardless of species.

A particularly interesting experiment involved zebra finches. Researcher Julie Elie did not just classify bird signals but validated results through behavioral experiments. Birds were trained to select specific sounds in exchange for food. It turned out that they more often confused signals with similar meaning rather than similar sound. For linguists, this became one of the most intriguing findings in recent years, as it resembles principles of human language.

However, as research progresses, more questions arise. The main issue is that neural networks are excellent at finding statistical patterns, but do not necessarily understand their meaning.

Modern AI can detect that a certain signal is consistently associated with a specific behavior. But this does not necessarily mean understanding the meaning of the signal itself. The system may simply be capturing correlation rather than decoding communication in the human sense.

Many experts warn about the risk of anthropomorphism – the tendency to see familiar linguistic structures where they may not exist. In other words, we may interpret complex statistical patterns as real language simply because we strongly want to find something human-like in animals.

Ethical concerns are also significant. Some scientists fear that attempts to initiate dialogue with wild animals could disrupt natural social structures. If researchers begin reproducing signals that animals interpret as warnings, mating calls, or alarm signals, the consequences could be unpredictable.

Nevertheless, interest in the topic continues to grow. An additional stimulus is the Dolittle Prize, which offers awards of up to 10 million dollars for creating a system capable of enabling real interaction between humans and animals.

It is important to understand that half a century ago such projects were almost impossible. In the 1960s, famous dolphin researcher John Lilly also attempted to decode animal communication but faced the limitations of computational power at the time. Today the situation has changed dramatically. Modern neural networks can analyze millions of signals simultaneously and detect patterns that were previously completely hidden from humans.

Therefore, the key question is no longer “do animals have language?” but rather “how close are we to understanding it?” For the first time in human history, we may truly be opening a door to a world that has always existed around us but remained inaccessible to understanding.

And if one day technology allows full dialogue with animals, humanity will face an unexpected problem. For decades we have dreamed of learning to speak with other species. But are we ready to hear what they might say in return?