Google has created artificial intelligence to decipher the language of dolphins. The model called DolphinGemma will be tested in real conditions this summer. Her main goal is to help scientists better understand the complex vocalizations of marine mammals and determine whether they have features similar to human speech. About this informs Popular Science.
Dolphins have long been considered one of the most intelligent animals on the planet. They show the ability to learn, cooperate, recognize themselves in the mirror and demonstrate complex social behavior. Their language—a system of various sounds, including whistles, clicks, and even “croaking”-like sounds—has remained largely unsolved until today.
The main task of researchers is to understand exactly how dolphins use sounds in social interactions. Some of the vocalizations have already been associated with specific behavior. For example, dolphins emit individual “name” whistles to recognize each other, and sounds that scientists describe as “croaks” usually accompany conflict situations. However, to find out whether their communication really has the characteristics of language, a deeper analysis is needed.
In this context, DolphinGemma comes to the rescue, a model built on top of Google’s Gemma open source AI models, which use the same architecture as the commercial Gemini models. DolphinGemma uses SoundStream technology, which allows you to convert dolphin sounds into a format suitable for processing by artificial intelligence. The model analyzes the sound messages of the dolphins and predicts the next “token” — an acoustic unit that can be meaningful in their communication. Her work resembles the principle of operation of LLMs, which generate text in response to a request.
DolphinGemma is expected to help identify complex sound patterns that can form the basis of a conditional “vocabulary” of communication between humans and dolphins. Google notes that without the involvement of AI, such an analysis would take decades.
Another advantage of DolphinGemma is its optimization for Pixel smartphones. For several years now, the WDP team has been using the CHAT (Cetacean Hearing Augmentation Telemetry) device in the field, assembled on the basis of the Pixel 6. This system not only records dolphin sounds, but also allows the reproduction of synthetic vocalizations – potential “words” that can facilitate the establishment of contact between researchers and dolphins.
