In a fascinating intersection of artificial intelligence and marine biology, Google has embarked on an ambitious project to decode and potentially communicate with dolphin language. This innovative initiative combines Google's AI capabilities with specialized hardware to help researchers better understand one of the planet's most intelligent species.
The DolphinGemma AI Model
Google has developed a specialized AI model called DolphinGemma, designed specifically to analyze and interpret dolphin vocalizations. Based on Google's existing Gemma AI models, DolphinGemma has been trained on an extensive dataset of dolphin sounds collected over nearly four decades by the Wild Dolphin Project (WDP). The model processes sequences of dolphin sounds to identify patterns and structure, ultimately predicting likely subsequent sounds in a sequence—much like how language models work with human text.
Key Components of the Dolphin Communication Project:
- AI Model: DolphinGemma (400 million parameters)
- Hardware: Pixel phones (currently Pixel 6, upgrading to Pixel 9 in 2025)
- Research Partners: Wild Dolphin Project, Georgia Institute of Technology
- Technology: SoundStream tokenizer, CHAT (Cetacean Hearing Augmentation Telemetry)
- Data Source: 40 years of dolphin sound recordings from the Wild Dolphin Project
Dolphin Language Complexity
Dolphins communicate through a variety of distinct sounds, including whistles, squawks, and clicking buzzes. Each of these sound types corresponds to different contexts and behaviors. For instance, dolphins use signature whistles that function similarly to names, helping mothers locate their calves. Click buzzes often accompany courtship or shark-chasing behaviors, while burst-pulse squawks typically occur during confrontations. The structure and patterns in these vocalizations suggest a sophisticated communication system that researchers are eager to decode.
Pixel Phones as Field Research Tools
What makes this project particularly innovative is the use of Google's Pixel smartphones as the primary recording and processing devices. Field researchers are using Pixel phones to capture dolphin sounds underwater in real-time. The phones run the recordings through Google's SoundStream tokenizer, allowing the sounds to be fed directly into the DolphinGemma model as they're being recorded. Currently, researchers are using the Pixel 6, with plans to upgrade to the Pixel 9 for the 2025 summer research season.
CHAT System for Two-Way Communication
Beyond just listening and analyzing, the project aims to establish rudimentary two-way communication with dolphins. The Wild Dolphin Project has partnered with Georgia Tech to develop CHAT (Cetacean Hearing Augmentation Telemetry), a system that creates synthetic whistles associated with specific objects dolphins enjoy, such as sargassum, seagrass, or scarves used by researchers. The hope is that dolphins will mimic these sounds to request these items, creating a simple form of interspecies communication.
Practical Benefits of Mobile Technology
Google's approach offers significant practical advantages for field research. By optimizing DolphinGemma to run directly on Pixel devices, the project dramatically reduces the need for specialized hardware, improves system reliability, lowers power consumption, and cuts costs—all crucial factors for researchers working in remote marine environments. The 400 million-parameter model is specifically sized to function efficiently on the mobile devices used in the field.
Open Science Approach
Recognizing the potential broader impact of this research, Google plans to share DolphinGemma as an open model this summer. While the AI was primarily trained on Atlantic spotted dolphin sounds, Google notes that it still has utility for studying other species like bottlenose or spinner dolphins. This open approach aims to accelerate research worldwide and foster a deeper understanding of these highly intelligent marine mammals.
Future Directions
The ultimate goal of the project extends beyond just understanding dolphin communication patterns. Researchers hope to eventually construct sounds that dolphins can understand and respond to, potentially establishing a rudimentary form of cross-species communication. The upcoming Pixel 9-based CHAT system will be able to run both deep learning and template matching simultaneously, potentially enabling more fluid interactions with dolphins during the 2025 research season.
