A Google project called DolphinGemma uses AI to understand what dolphins say
Google is now leveraging its AI technology and the latest Pixel 9 smartphone to explore and decode the communication patterns of dolphins. With a particular interest in understanding what dolphins are trying to convey, the tech giant is working alongside The Wild Dolphin Project (WDP), a non-profit group dedicated to studying Atlantic spotted dolphins and their natural behaviors. The data collected by WDP has been used to train a specialized AI model named Dolphin Gemma, which builds upon Google’s open Gemma series and is optimized to function efficiently even on mobile devices. Impressively, Dolphin Gemma is capable of generating sounds that closely resemble actual dolphin vocalizations.
This ground-breaking study is a component of WDP’s larger plan to eventually develop a platform that can recognize dolphin sounds and produce interactive answers. Using the upgraded Pixel 9 phone, researchers plan to both synthesize dolphin-like sounds and listen to real-time vocalizations in order to craft contextually accurate replies. Previously, the team had been relying on the Pixel 6, but the move to the Pixel 9 provides the ability to simultaneously run advanced AI models and pattern-matching tools on a single device.
Switching to smartphones such as the Pixel 9 has proven to be highly beneficial for ocean fieldwork, as it reduces the need for bulky, specialized equipment. This shift not only makes the research setup more manageable but also ensures lower power consumption, smaller physical footprint, and increased reliability in unpredictable environments like the open sea.
Additionally, Dolphin Gemma supports the CHAT (Cetacean Hearing Augmentation Telemetry) system by enhancing predictive capabilities, helping researchers detect mimicry early in dolphin vocal patterns. As a result, human-dolphin interactions may be more timely and of higher quality.
Understanding the fundamental structure and potential meaning of dolphin communication is the ultimate goal. WDP has long observed how various vocal sounds are linked to specific social and behavioral scenarios, and now, with AI’s help, they hope to identify patterns that could point to a more language-like system beneath the surface.
