Google Is Training AI to Speak Dolphin

Scientists are making strides toward two-way communication with dolphins, using artificial intelligence to decode and replicate dolphin vocalizations. Google, the Georgia Institute of Technology, and the Wild Dolphin Project (WDP) have developed DolphinGemma, the first large language model (LLM) trained on 40 years of Atlantic spotted dolphin data. This AI generates dolphin-like sounds, including complex “burst pulses,” previously unreproducible by traditional methods. Researchers hope the LLM can identify patterns in dolphin communication and possibly create “words” for objects, testing dolphin responses using specialized technology called CHAT.

However, experts caution that mimicked behaviors may not indicate true understanding—dolphins may be reacting to rewards rather than grasping language. Critics also question whether dolphins possess a language structure comparable to humans. Still, projects like this offer new insight into dolphin cognition and communication. If dolphins respond meaningfully, it could reshape how we perceive and empathize with them, encouraging stronger efforts toward their conservation.