AI’s Next Trick? Speaking Dolphin

Denise Herzing, founder and director of the Wild Dolphin Project, told the audience at WIRED World Fair that artificial intelligence has finally given her a way to work through roughly 40 years of dolphin recordings. Herzing first began thinking about marine mammal communication during a 1985 trip to the Bahamas, where she studied Atlantic spotted dolphins and heard their characteristic clicks and whistles.

Her research group has amassed countless hours of video and audio of dolphins, and it is now collaborating with scientists at Georgia Tech and Google DeepMind to impose structure on dolphin language. Herzing framed AI as the missing ingredient after decades of manual effort: "With AI, we can find patterns," she said, calling it "a powerful new tool" that she had personally been waiting for. The work aims to organize the signal in the recordings, though Herzing was explicit that full translation has not yet been achieved.

The team has also attempted two-way communication by building an underwater wearable computer capable of playing recorded dolphin sounds. In demonstrations, the device was used to signal specific items and habitats — for example, a sound standing for a scarf and another for sargassum, a type of seaweed. Two researchers would model the behavior, one playing the scarf sound while the other interacted with the object, and then the group would observe how the dolphins responded.

A central obstacle is separating mimicry from genuine understanding. "Even though we had these whistles and are labeling toys, there's a big difference between mimicking and understanding the function," Herzing said, noting that distinction is hard to establish in the wild. She suggested that working in a sanctuary or another more controlled environment might help researchers tell the two apart.

Even with patterns in hand, a real conversation with dolphins remains distant. Herzing's team has made progress on dolphin social hierarchies and behavior, but meaning is a separate problem: "We might have the data and the structure, but what does it mean in dolphin culture?" she asked, calling that "an exploration on its own." Looking ahead, the open questions are whether AI-derived patterns hold up as semantic structure rather than statistical correlation, whether controlled settings can resolve the mimicry problem, and what a two-way exchange would even require. Herzing also noted that when she asks people what dolphins would say to humans, the most common answer runs along the lines of, "Why are you messing up our ocean?"

Herzing framed the effort as ultimately instructive about humans as much as dolphins — a way to examine our own relationship with the planet through another species' communication.