Artificial Intelligence is transforming industries around the world, but one of its most jaw-dropping applications is happening beneath the waves. In a groundbreaking development, U.S. tech giant Google has unveiled DolphinGemma, a large language model (LLM) designed to facilitate communication between humans and dolphins.

This AI-powered system is currently undergoing testing to determine whether it can replicate and respond to the complex vocal behavior of dolphins. If successful, DolphinGemma could mark the culmination of over 40 years of research and conservation efforts focused on understanding dolphin communication.

Dolphins are widely recognized as some of the most intelligent creatures in the ocean. Their intricate social structures and vocalizations have fascinated scientists for decades. The Wild Dolphin Project (WDP) has been at the forefront of this research, carefully documenting the behaviors and sounds of a single community of Atlantic dolphins in the Bahamas since the 1980s. Researchers have cataloged various sound patterns linked to courtship, arguments, and even signature “names” used between dolphins.

Historically, the idea of human-cetacean communication has seemed more like science fiction than science fact. The challenge has always been the complexity of dolphin vocalizations—whistles, clicks, and burst pulses that are difficult to decode, let alone replicate. That began to change with the advent of powerful AI models capable of analyzing vast datasets and making nuanced predictions.

In partnership with Google and the Georgia Institute of Technology, WDP began exploring whether LLMs—the same type of technology powering Google’s Gemini and tools like ChatGPT—could bridge the communication gap. The result is DolphinGemma, an AI model with around 400 million parameters, tailored specifically to understand and predict dolphin vocalizations.

DolphinGemma works by receiving underwater audio inputs and analyzing them to forecast the next likely sound in a sequence. This predictive approach mimics how LLMs handle human language. The system operates in conjunction with the Cetacean Hearing Augmentation Telemetry (CHAT) interface, installed on modified Google Pixel smartphones. While CHAT doesn’t fully translate dolphin language, it facilitates a simplified, shared mode of interaction—a foundational step toward real-time interspecies communication.

Although full “conversations” with dolphins remain a long-term goal, DolphinGemma represents an extraordinary leap forward. By building a rudimentary vocabulary shared between species, the project opens up new possibilities for understanding marine life, enhancing conservation efforts, and rethinking the boundaries of AI applications.

With continued development, DolphinGemma could change how we interact with the natural world—turning age-old dreams of talking to animals into a scientific reality.

By Impact Lab