'The Goal Would Be to Speak Dolphin': Google Develops AI to Decipher Communications Between Dolphins The AI model was trained on 40 years' worth of audio and video of dolphins communicating with each other.
By Sherin Shibu Edited by Melissa Malamut
Key Takeaways
- Google’s new AI model, DolphinGemma, attempts to decode the way dolphins communicate.
- DolphinGemma was trained on 40 years’ worth of data and can predict dolphin vocalizations.
- The AI model will be used in the field for the first time this season.
What are dolphins saying to each other? Google's new AI model attempts to understand the hidden language of dolphins, so humans can try to talk back.
Earlier this month, Google announced a new AI model called DolphinGemma, developed in partnership with the Georgia Institute of Technology and the Wild Dolphin Project, a nonprofit.
DolphinGemma is the first AI model that attempts to understand dolphin language. The AI was trained on 40 years' worth of audio and video of Atlantic spotted dolphins in the Bahamas. It absorbed decades of dolphin vocalizations with the end goal of identifying common patterns, structures, and even possible meanings behind dolphin communication.
Just like how an AI model predicts the next word in a typed sentence, Google's DolphinGemma AI model aims to use its training data to predict the next sound a dolphin makes based on observed patterns. It can also create new, made-up, AI-generated dolphin sounds.
Related: New Google Report Reveals the Hidden Cost of AI
The Wild Dolphin Project is starting to use DolphinGemma in the field this season for the first time to help researchers understand dolphin communication.
AI has the advantage of picking up on patterns that human beings might not recognize in dolphin audio and analyzing the data far more quickly than humans can. Dr. Denise Herzing, founder and research director of the Wild Dolphin Project, told Scientific American that it would take human beings 150 years to manually comb through the data and pull out the same patterns that DolphinGemma can pick up on today.
"Feeding dolphin sounds into an AI model like DolphinGemma will give us a really good look at if there are pattern subtleties that humans can't pick out," Herzing stated in an announcement video. "The goal would be to one day speak dolphin."
DolphinGemma will also come up with new, made-up dolphin-like sounds that the researchers will play in the water this season to see how the animals react to new vocalizations.
It works like this: A pair of researchers will swim next to a dolphin, playing the AI-generated sound and passing a food item that dolphins enjoy, like seagrass or sargassum, back and forth. If the dolphin mimics the AI-generated sound, the researchers will respond by giving the dolphin the treat.
Related: These Are AI's 'Most Obvious' Risks, According to Google's Former CEO
The research is limited to one population of dolphins in one area — other groups could vary in how they communicate with each other. Google says it plans to release DolphinGemma as an open-source AI model this summer, so that academics can use it to help study other dolphin species, like bottlenose or spinner dolphins.
"By providing tools like DolphinGemma, we hope to give researchers worldwide the tools to mine their own acoustic datasets, accelerate the search for patterns and collectively deepen our understanding of these intelligent marine mammals," Google wrote in a blog post.
AI is also being used to understand other animals. Late last year, the Earth Species Project announced an AI model called NatureLM, which can identify an animal's species, age, and state of distress based on audio. Meanwhile, Project CETI (Cetacean Translation Initiative) uses AI to study sperm whale communication.
