Google has recently unveiled an AI model called DolphinGemma, marking a significant step forward in the quest to understand and potentially communicate with dolphins. This initiative, announced on National Dolphin Day, is a collaborative effort between Google, the Wild Dolphin Project (WDP), and researchers at Georgia Tech. DolphinGemma is designed to analyze the complex vocalizations of dolphins, including clicks, whistles, and burst pulses, with the ultimate goal of deciphering their communication patterns.
The WDP has been diligently collecting data on dolphin communication for nearly four decades, establishing the world's longest-running underwater dolphin research project since 1985. This extensive dataset, meticulously pairing underwater video and audio recordings with individual dolphin identities, life histories, and observed behaviors, serves as the foundation for training DolphinGemma. The AI model, backed by Google's Gemini technology, has been trained on this vast, labeled dataset to identify recurring sound patterns, clusters, and sequences within dolphin vocalizations.
DolphinGemma operates by running dolphin sounds through Google's SoundStream tokenizer, which identifies patterns and sequences. This process uncovers hidden structures and potential meanings within the dolphins' natural communication, a task that previously required immense human effort. The model can also predict the subsequent sounds a dolphin may make, much like how large language models for human language predict the next word in a sentence. One of the key advantages of DolphinGemma is its size. With approximately 400 million parameters, the model is compact enough to run on Pixel smartphones. This allows researchers to analyze dolphin sounds in real-time in the field, reducing costs and the need for custom hardware. The WDP is already using Pixel phones to record and analyze data underwater and plans to upgrade to a Pixel 9 this summer.
Beyond analyzing natural communication, the WDP, in partnership with Georgia Tech, is exploring potential two-way interaction with dolphins using the Cetacean Hearing Augmentation Telemetry (CHAT) system. CHAT is an underwater computer designed to establish a simpler, shared vocabulary with dolphins. It works by associating novel, synthetic whistles with specific objects that dolphins enjoy, such as sargassum, seagrass, or scarves. Researchers hope that dolphins will learn to mimic these whistles to request the items.
Google plans to release DolphinGemma as an open-source model this summer, allowing researchers worldwide to adapt it for studying other cetacean species, such as bottlenose or spinner dolphins. By providing tools like DolphinGemma, Google aims to empower researchers to mine their acoustic datasets, accelerate the search for patterns, and collectively deepen our understanding of these intelligent marine mammals.
While DolphinGemma represents a significant technological advancement, ethical considerations are also being taken into account. Researchers are mindful of potential disturbances to dolphins and are working to minimize any negative impacts on their natural behavior. The use of AI in animal communication research raises broader questions about animal rights, consent, and the responsible use of technology.