Dolphins have long been considered some of the most intelligent animals on Earth, with their impact also recognized in health improvement as one of the therapeutic methods. And now, in the age of technology, unprecedented opportunities and potential are opening up to understand the communication between these remarkable animals. This path is being opened by artificial intelligence (AI) and Google’s application of it.

Google Takes First Steps in Decoding Dolphin Communication

Just recently, on April 14, Google announced that its newly developed DolphinGemma large language model will assist scientists who study the communication of these creatures and aim to uncover the nature of how these animals interact.

In collaboration with researchers at Georgia Tech and the fieldwork team of the Wild Dolphin Project (WDP), Google is developing a core AI model trained to learn the structure of these animals’ vocalizations and to generate new sequences of sounds that resemble those used by dolphins.

The Capabilities of the DolphinGemma Model

The AI model introduced by Google utilizes the company’s specific audio technology – SoundStream token generators to represent animal sounds. These captured sounds are then processed through the model’s architecture.

Essentially, it is a medium-sized model with around 400 million parameters, operating directly on Pixel phones used by the WDP.

Google’s model is based on the advanced open model collection called Gemma, whose technologies and research are similar to the Gemini models.

The model is trained on a 40-year dataset of Atlantic spotted dolphin vocalizations. Over time, the AI model will be capable of predicting the most likely next sounds in a sequence, like how large language models can now predict the next word in a sentence.

But That’s Not All

In partnership with the mentioned institutions, Google is also developing a device called CHAT – Cetacean Hearing Augmentation Telemetry. This is a wearable technology that will allow researchers to generate AI-created dolphin-like sounds associated with specific objects animals’ favour.

The principle behind the device is that two divers, wearing the technology, swim near the animal, and an object paired with a generated sound is transmitted as a signal. If the animal eventually starts mimicking the specific sound, the researcher will reward it with the corresponding object. The hope is that these creatures will learn to imitate whistles and request particular items.

Example of a portable CHAT device. Image source: Georgia Tech College of Computing

Contribution to Dolphin Research

Google has developed, and will continue to improve, a unique technology that will undoubtedly have a significant impact on the field of animal studies. For now, it is emphasized that it is important to ensure this becomes a natural integration rather than a form of conditioning, where animals learn to act in exchange for rewards.

There is still work to be done before it functions reliably, and Google reports that this summer they plan to release DolphinGemma as an open-source model, so that researchers can test it with various cetacean species.

Final Word

AI and tech giants once again demonstrate that technological progress is ongoing and being applied to meaningful and important global research efforts. Although this technology is still in its early stages, we can already begin to imagine what the future of decoding dolphin language might look like.

If you are interested in this topic, we suggest you check our articles:

• AI Climate Solution: Will We Finally Solve the Ongoing Battle Against Climate Change?
• AI Weather Forecasting: New Dawn of Meteorology?
• How are AI Solutions for Climate Change Making an Impact?
  

Sources: Google, Smithsonian magazine