AI Model May Help Us Understand What Dolphins Say

Google, an American multinational corporation and technology company, has developed a generative artificial intelligence (AI) model to speak to dolphins, using decades of data on the highly intelligent marine mammal.

The dolphin communication model DolphinGemma uses a Google-developed audio technology called SoundStream to tokenize dolphin vocalizations – clicks and whistles – and ultimately speak dolphin, even though “we do not know if animals have words,” says Denise Herzing from Wild Dolphin Project (WDP).

WDP has been studying dolphins since 1985, tracking a specific community of Atlantic spotted dolphins to analyze how the mammal vocalizes and how that can affect their social interactions. Decades of recording were used to connect some basic activities to specific sounds. The team created a massive, meticulously labelled data set perfect for analysis with generative AI. DolphinGemma has been trained on Atlantic spotted dolphin sounds, but Google suggests it should be feasible to fine-tune it for other cetacean species.