Elephants are among the most communicatively complex animals on Earth. They gesture, trumpet, and produce deep rumbling sounds, some pitched so low that human ears cannot detect them at all. For decades, researchers have recorded these vocalizations in the wild. The harder question has always been: what do they actually mean? A study published on October 1 in the journal Scientific Reports takes a significant step toward an answer, using machine learning to analyze not just individual elephant calls, but the coordinated choruses that groups produce together.
902 Recordings and What the Algorithm Found
The research drew on audio collected between 1986 and 2022 at Amboseli National Park in Kenya, covering wild African savanna elephant family groups. The dataset comprised 902 recordings of rumbles, both solo calls and chorused ones, captured across six distinct social situations. These included “coo” rumbles directed by females toward calves, “contact” rumbles exchanged between separated individuals, “let’s-go” rumbles used to initiate group movement, “cadenced” rumbles produced during eating, resting, or slow movement, “little-greeting” rumbles between closely bonded elephants, and “greeting” rumbles during reunions.
Researchers trained a machine learning algorithm to classify elephant activity based on these recordings. The model performed better than random chance, though it still made a notable number of errors. That imperfect result is worth pausing on. The algorithm was not reading a clean signal. It was attempting to interpret layered, overlapping vocalizations produced by multiple animals simultaneously, a task that would be practically impossible for human analysts working through audio manually at any meaningful scale.
Choruses as Coordinated Messages, Not Just Noise
The study’s central finding is that elephant choruses are not simply the acoustic result of several animals calling at once. The joint vocalizations appear to carry information as a unit, information that cannot always be extracted from any single rumble within the group. In certain contexts, the individual calls within a chorus became more acoustically similar to one another over time. This phenomenon is called “vocal convergence.”
Sara Keen, a behavioral ecologist at the Earth Species Project, a nonprofit focused on using AI to decode animal communication, described vocal convergence to New Scientist as analogous to what happens in human conversation, where people gradually begin to mirror each other’s speech patterns or accents. In the elephant data, vocal convergence appeared specifically during cadenced, contact, and greeting rumbles. The researchers suggest it may help elephant families coordinate group behaviors, though they note that further research is needed to confirm this interpretation.
Andrew McLean, an independent animal behavior researcher and member of the International Union for Conservation of Nature’s Asian Elephant Specialist Group, described the findings as “fascinating,” while noting they are “not really surprising, given how intensely social elephants are.” That framing is useful. The results do not overturn existing knowledge about elephant intelligence. They extend it, and they do so using tools that allow researchers to process data at a scale and speed that was not previously available.
What This Reveals About AI’s Role in Understanding the Natural World
Here is what most coverage of studies like this one tends to underemphasize: the significance is not just about elephants. It is about a methodological shift in how science approaches animal communication.
Machine learning algorithms can identify patterns across hundreds or thousands of recordings in the time it would take a human researcher to manually review a fraction of that material. This capacity is already being applied across species. Scientists have used similar tools to identify mosquito species, sex, and feeding status from sound alone, to identify what researchers describe as a sperm whale phonetic alphabet, to find that marmoset monkeys call one another by name, and to analyze crow vocalizations. African savanna elephants may also call one another by name, according to separate research referenced in the source material.
The common thread is pattern recognition at scale. Animal communication is dense, contextual, and often operates across sensory channels that humans do not naturally perceive, including infrasound, the low-frequency range where elephant rumbles travel through both air and ground. AI tools do not replace the expertise of behavioral ecologists and biologists. They handle the volume of data that human analysis alone cannot process, freeing researchers to focus on interpretation, hypothesis formation, and the design of follow-up studies.
Stephanie King, a behavioral biologist at the University of Bristol, put it directly: “If it’s used responsibly, AI can transform what we can do with the data we collect.”
That word “responsibly” carries weight. The algorithm in this study made mistakes. The researchers acknowledged them. That transparency matters because it reflects a realistic understanding of what these tools can and cannot do at this stage of development.
In Short
A machine learning model analyzed 902 recordings of African savanna elephant rumbles collected over nearly four decades, finding that group choruses carry coordinated information beyond what individual calls contain. A phenomenon called vocal convergence, where calls within a chorus become more acoustically similar, may play a role in group coordination. The study is part of a broader pattern: AI is becoming a core instrument in animal communication research, not because it understands animals, but because it can process the scale of data that understanding them requires.
Based on reporting from Smithsonian Magazine.