Dolphins appear to have names.
Now, an AI is listening to everything else they say.
It has studied decades of clicks, whistles and burst-pulse squeals.
Already, it is finding patterns nobody expected.
Could a machine finally open a door humans have knocked on for decades? What is hidden in that chatter?
To follow the story, start with the people doing the listening.
The video right below this article shows how AI is being used to analyse dolphin sounds, and where marine biologists urge caution. Read on to the end, because one finding about dolphin “names” changes how you hear all of it.
Forty Years of Recordings Fed to One Machine

It begins in the Bahamas, in 1985.
Mount Kailash: The Sacred Peak No One Has Ever Climbed
There, Dr Denise Herzing founded the Wild Dolphin Project.
Since then, her team has followed wild Atlantic spotted dolphins season after season.
Over time, they matched sounds to individual dolphins, life histories and behaviour.
In April 2025, Google announced Dolphin-Gemma, developed with Georgia Tech and the project.
The model has roughly 400 million parameters. It is small enough to run on a Pixel phone in the field.
Its job sounds simple. It predicts which sound comes next, much like a chatbot predicts the next word.
Whistles That Might Be Names, Alarms & Questions

Meanwhile, in Sarasota Bay, another team is listening.
Biologist Laela Sayigh studies about 170 resident bottlenose dolphins there. Each one has a signature whistle, which researchers compare to a name.
Then came a surprise. The team found 22 shared whistle types, and one made dolphins swim away, like an alarm.
Another seemed to spark curiosity. A young male even answered 148 playbacks of his mother’s whistle with his own signature whistle.
Were those names, alarms and questions? Or something closer to words?
What happens when an AI starts sorting those whistles for us?
Patterns Are Found, but Meaning Is Still Missing
Here, the careful scientists step in.
DolphinGemma predicts sounds. It does not claim to translate them.
Researcher Christian Rutz argues that machine learning cannot extract meaning on its own, as Gizmodo reports. Context from behavioural ecology is essential.
The Sarasota results also varied. Age, sex, mood or social context may shape how each dolphin reacts.
Even the coverage of that study states plainly that this is not language in the human sense.
For now, the pattern is real. What it means is still being decided.

If dolphins really do call each other by name, what is the first thing you would ask them?