Scientists Use AI To Decode Animal Communication, Finding Whale And Crow Calls More Complex Than Once Thought

Researchers around the world are increasingly turning to artificial intelligence to decode the communication systems of other species, using machine learning models capable of processing vast quantities of audio and behavioral data to uncover patterns that had long gone unnoticed in animal calls.

The Earth Species Project, a nonprofit organization dedicated to using AI to promote interspecies understanding, has developed a generalizable machine learning model that can be applied across different species, while also collaborating with researchers around the world to build custom models tailored to specific animals. In northern Spain, the organization's AI tools are helping scientists understand how a population of cooperative-breeding crows communicate within family groups, according to Mongabay's coverage of the project. The same technology has separately been deployed to study how orcas communicate with one another and how underwater noise pollution affects those communications.

Vittorio Baglione and Daniela Canestrari, researchers who have spent decades studying carrion crows, have used the technology to examine how the birds coordinate complex, cooperative behavior during breeding season, when entire extended families become involved in raising chicks. The technology has also proven useful for practical field research purposes, helping scientists synchronize voice notes recorded in the field with corresponding whale audio recordings or drone footage.

The Earth Species Project's flagship AI system, known as NatureLM-Audio, takes an unconventional approach to training, incorporating human language and music into the model before it begins analyzing animal sounds. Aza Raskin, co-founder and president of the Earth Species Project, explained the reasoning behind that method.

"Teaching the model human language and music first helps it then understand animal communication," Raskin said.

Raskin said the results of that training approach have already exceeded expectations in certain respects.

"Already, it can correctly identify species by name, even when the model has never heard that species before," Raskin said.

Raskin has expressed confidence that with sufficient additional data, these AI models could eventually move beyond simply decoding existing animal communication, potentially advancing toward predicting what animals are likely to communicate next, and even drafting artificial responses intended to interact with them directly.

A separate research effort, Project CETI, has focused specifically on sperm whale vocalizations using machine learning to study how the whales' distinctive clicking patterns are organized and whether specific patterns tend to emerge during particular social situations. Researcher Shane Gero, who has studied sperm whale communication extensively, has described the variation found within these vocalization patterns as resembling a kind of "sperm whale phonetic alphabet," suggesting the whales may be using structured combinations of sounds to convey more complex information than previously assumed.

The broader scientific effort to apply AI to animal communication traces back to a landmark 2023 paper published in the journal Science, in which researchers including Christian Rutz, Michael Bronstein, Aza Raskin, Sonja Vernes, Katherine Zacarian and Damián Blasi outlined how machine learning techniques could be systematically applied to decode communication across species. That foundational work has since spurred a wave of follow-up research applying similar methods to specific animal groups, including primates, whales and birds, some of which researchers now believe may exhibit vocal complexity approaching that of human language in certain respects.

Recent studies published this year have continued expanding the scope of this research. One 2026 paper examined potential bias in "black box" AI models used to evaluate whether elephants use name-like calls to refer to specific individuals, underscoring the growing scientific scrutiny being applied to ensure AI-driven conclusions about animal communication rest on solid methodological ground rather than pattern-matching artifacts. Separately, a tool called TweetyBERT has been developed to automatically parse birdsong through self-supervised machine learning techniques, allowing researchers to analyze vocal patterns across large recorded datasets without requiring extensive manual labeling beforehand.

Despite the rapid pace of progress, researchers involved in this work have been careful to distinguish between identifying structural patterns within animal sounds and definitively proving what those patterns actually mean. Kristin Andrews, a philosopher who has examined the ethical dimensions of this research, has raised questions about the broader implications of using machine learning to interpret and potentially communicate with non-human species, exploring what responsibilities researchers may hold as these tools become more capable.

Scientists caution that animal communication systems remain fundamentally different from human language in important ways. While humans rely on grammar and symbolic structures capable of expressing complex, abstract and hypothetical ideas, most animal communication systems studied to date appear to lack that same open-ended structure, even as researchers continue finding unexpected layers of organization and context-dependence within specific species' vocalizations.

Researchers have generally framed the ultimate goal of this work not as building a literal universal translator capable of converting animal sounds directly into human language, but rather as developing rigorous scientific tools to test specific hypotheses about how different species communicate on their own terms. Proponents of the approach argue that these AI-assisted methods are creating what amounts to a provisional, evolving reference guide to non-human communication across the planet, built on accumulated evidence rather than the kind of casual, anecdotal interpretation that has historically characterized much of the public fascination with animal communication.

The stakes of this research extend beyond pure scientific curiosity. Advocates argue that a deeper, evidence-based understanding of how animals communicate could meaningfully inform conservation efforts, helping researchers better monitor wildlife populations, assess the health and social dynamics of animal groups, and evaluate the impact of human-generated noise pollution on species that rely heavily on acoustic communication, such as whales navigating increasingly busy ocean shipping lanes.

As these AI models continue improving and expanding to additional species, researchers involved in the field have emphasized the importance of maintaining scientific discipline and rigorous evidence standards even as public interest in the prospect of "talking" to animals continues to grow. For now, the technology's most significant contributions remain focused on uncovering previously hidden structure within animal communication systems, rather than achieving anything resembling full, two-way conversation between humans and other species, a goal researchers say remains firmly in the realm of long-term ambition rather than near-term scientific reality.