Scientists have known for decades that dolphins “talk” to each other using clicks, whistles, and pulses. Today, Google announced the DolphinGemma artificial intelligence (AI) model that was developed to help scientists understand dolphin conversations.
DolphinGemma was developed through a collaboration between Google, Georgia Tech, and the Wild Dolphin Project. In addition to interpreting animal conversations, it can also translate human language into dolphin language to enable communication between the two mammal species.
This AI model is based on the Gemma model, which was developed specifically for scientific research. It has around 400 million parameters using data collected by the Wild Dolphin Project since 1985. The DolphinGemma model is small enough that it can be run using a Pixel 6 in the field by researchers.
After the model was developed, the CHAT (Cetacean Hearing Augmentation Telemetry) system was developed by Georgia Tech to collect the sounds and then try to understand whether they are conversations. Researchers are now producing specific sounds using CHAT for specific objects in the hope that dolphins can learn new vocabulary that didn't exist before. In the near future, the Pixel 6 devices currently in use will be replaced by the more powerful Pixel 9.
The ability to communicate with different species is science fiction that previously only existed in TV series. In the SeaQuest DSV series, for example, humans can already communicate with dolphins using a device mounted on their heads. We have also successfully taught chimpanzees to use sign language, but not using the chimpanzee's own language. Perhaps in the near future, science fiction will become reality with the help of AI.