Wild Dolphin Project uses AI with DeepMind to find patterns in dolphin sounds

Wild Dolphin Project uses AI with DeepMind to find patterns in dolphin sounds

Large language models already talk with people in dozens of languages, and Wired's report from WIRED World Fair asks whether the same approach could work with dolphins. That is the premise of research by Denise Herzing, founder and director of the Wild Dolphin Project. She first began thinking about how dolphins communicate on a 1985 trip to the Bahamas to study Atlantic spotted dolphins, after hearing their clicks and whistles.

Herzing told the WIRED World Fair audience that it has been a 40-year wait, but AI is finally letting her untangle what dolphins are saying. Her group has gathered countless hours of video and audio footage of dolphins, and it is now working with scientists at Georgia Tech and Google DeepMind to give structure to dolphins' language, though not to fully translate it yet. "With AI, we can find patterns," she said, adding: "This is a powerful new tool. I've personally been waiting for it."

The team has also used some of this work to attempt two-way communication. They rigged up an underwater, wearable computer that can play dolphin sounds, including sounds meant to indicate items such as a scarf and habitat such as sargassum, a seaweed. Two researchers would model behavior: one plays the scarf sound, for instance, and the other then plays with the scarf. The team then watches how the dolphins respond.

Telling communication from mimicry is one of the main challenges. Herzing said: "Even though we had these whistles and are labeling toys, there's a big difference between mimicking and understanding the function. That's hard to do in the wild." Working in a sanctuary or another more controlled environment might help researchers separate the two, she said.

What dolphins say to each other is still partly a mystery. The team has found patterns, but a two-way conversation remains a ways off. Humans have made decent headway in understanding dolphin social hierarchies and behavior, yet more work is needed before a real conversation is possible. In Herzing's words: "We might have the data and the structure, but what does it mean in dolphin culture? That's going to be an exploration on its own."

The article closes on a wider point. Talking with dolphins would be very cool, but trying to understand them could also teach people about themselves and their relationship with the planet. When Herzing asks people what dolphins would say to us, the most common answer is along the lines of "Why are you messing up our ocean?" That answer, she said, "tells you how we are thinking about the world."

Key facts

  • Denise Herzing, founder and director of the Wild Dolphin Project, says AI is helping her group find patterns in dolphin communication, after a 40-year wait since her 1985 Bahamas trip.
  • The group has countless hours of dolphin video and audio and is working with scientists at Georgia Tech and Google DeepMind to give structure to dolphins' language, not yet to fully translate it.
  • The team tried two-way communication with an underwater wearable computer that plays dolphin sounds for items like a scarf and habitat like sargassum.
  • Telling mimicry from understanding is hard in the wild; a sanctuary or more controlled setting might help, Herzing said.
  • A real two-way conversation remains a ways off, because what the patterns mean in dolphin culture still has to be worked out.

Why it matters

The idea is that the pattern-finding strength behind large language models, which already let people communicate in dozens of languages, could be pointed at animal communication. Herzing frames AI as a tool she has waited for since 1985: it lets her group find structure in a large archive of dolphin recordings. The work is about structure, not translation. The article says the team is giving structure to dolphins' language, if not fully translating it yet.

Who it affects

Mainly researchers who study animal communication, and the Wild Dolphin Project with its partners at Georgia Tech and Google DeepMind. The article also points at a broader audience: Herzing suggests that trying to understand dolphins could help people understand themselves and their relationship with the planet.

How to use it

There is nothing to download or buy here. The article describes a research method. Collect long audio and video records of animals, use AI to look for patterns, then test with a wearable underwater device that plays sounds tied to objects and habitat, with one researcher playing a sound and another acting on it while the dolphins' reaction is observed. Herzing's own suggestion for the next step is to work in a sanctuary or another controlled environment.

How solid is it

This is a conference talk reported by Wired, with Herzing as the main voice. The article reports no result from the two-way experiment: how the dolphins responded is not stated. It names no specific AI model or training method, and does not say what Georgia Tech or Google DeepMind contribute. It gives no quantity of data beyond "countless hours", and no timescale for a conversation. It does not say that any dolphin sounds have been translated or that any dolphin "words" have been decoded.

Risks and caveats

Herzing herself flags the main problem: there is a big difference between mimicking and understanding the function, and that is hard to establish in the wild. Even with data and structure, the meaning in dolphin culture is a separate exploration. The article says a two-way conversation remains a ways off. Its opening line is also hedged: it "may" not be a stretch to use language models to talk with dolphins.

“We might have the data and the structure, but what does it mean in dolphin culture?”

— Denise Herzing, founder and director of the Wild Dolphin Project, at WIRED World Fair