The idea of speaking to animals has long been confined to science fiction and childhood dreams. But thanks to a breakthrough collaboration between marine biologists and AI experts, we might be closer to a real-life version of that fantasy, starting with dolphins.
A new AI model, DolphinGemma, could be the key to unlocking dolphin communication. Dr. Denise Herzing, founder of The Wild Dolphin Project, has been on this mission for over 40 years. Studying a specific pod of Atlantic spotted dolphins in the Bahamas, she’s built an extensive archive of dolphin sounds. With the help of Dr. Thad Starner, an AI pioneer from Google DeepMind and Georgia Tech, her research is now backed by powerful machine learning capable of interpreting complex audio patterns.
“The goal would be to one day speak Dolphin,” Herzing says.
For decades, researchers have relied on hydrophones to capture dolphin vocalizations, ranging from signature whistles used like names to echolocation clicks and social burst-pulse sounds. Analyzing these recordings using spectrograms has helped scientists spot patterns, but it’s been a time-consuming and imprecise process.
Now, DolphinGemma offers a much faster and more refined approach. The model builds on the same foundational technology that powers Google’s Gemini AI and is trained on a massive collection of dolphin audio from Herzing’s project. It converts these sounds into data tokens, identifying recurring structures and predicting what sounds might come next, similar to how your email app predicts the rest of your sentence. According to Herzing, feeding dolphin sounds into this system reveals patterns and subtleties that would otherwise go unnoticed by the human ear.
But decoding isn’t the only goal; communicating back is the next frontier. In the 1990s, Herzing’s team experimented with a keyboard designed for dolphin use, where artificial whistles were linked to symbols and toys. Although it sparked curiosity, it wasn’t interactive enough. So the team advanced their technology by developing underwater communication tools in collaboration with Georgia Tech. This led to the creation of the CHAT system,m an underwater wearable computer that detects and plays dolphin sounds and allows divers to engage in real-time interaction.
The CHAT system has evolved significantly. Using Google Pixel phones, the current version is smaller, more energy-efficient, and equipped with AI that can help researchers respond faster and more naturally to dolphin cues. The process involves associating a specific sound with an action, observing if a dolphin mimics the sound, and then quickly responding to reinforce that communication. The AI model makes the system smarter by helping anticipate and interpret these interactions with greater accuracy.
This field season, the Wild Dolphin Project will begin deploying DolphinGemma in real-time studies. Google plans to make the model open-source later this year, which means scientists around the world could adapt it to study other dolphin species, like bottlenose or spinner dolphins. That expansion could accelerate our understanding of cetacean communication globally.
Dr. Starner said, “If dolphins have language, then they probably also have culture.” Understanding what dolphins talk about, what matters to them, how they socialize, and what they teach one another could radically alter our perception of non-human intelligence.
