An in-depth look at Project CETI, a research effort that seeks to use AI to understand the vocalizations of sperm whales off the coast of Dominica
Craig Welch / National Geographic : Tweets: @davideagleman , @liv_boeree , @dabeard , and @natgeo Tweets: David Eagleman / @davideagleman : Too cool. I retain hope that within our lifetimes we humans will be able to decode what our cetacean cousins are chatting about. https://www.nationalgeographic.com/ ... Liv Boeree / @liv_boeree : This would be so unbelievably epic https://www.nationalgeographic.com/ ... David Beard / @dabeard : For 40 minutes, the two whales were having a chat. The biologist had spent 13 years recording whales, but he never overheard anything like this. Now a new project matches whale clicks with behavior to decode their language. https://www.nationalgeographic.com/ ... @CraigAWelch @Brian_Skerry @natgeo : In what may be the largest interspecies communication effort in history, scientists will try to decipher what sperm whales are saying to one another https://www.nationalgeographic.com/ ...
Context & Ripple Effects
Project CETI marks the moment cetacean communication research gets a dedicated AI program: a team of biologists and machine-learning researchers stationed off Dominica to record sperm whales at scale, betting that the same techniques that cracked human language can crack theirs. The bet looked speculative at launch but has aged well — by 2024 researchers reported unlocking a kind of sperm whale "alphabet" using machine learning.
The project also sits inside a broader turn in the field: techniques proven on human language, notably self-supervised learning, migrated into bioacoustics, and Google followed with its own DolphinGemma model for spotted dolphins — turning what CETI started into a race with corporate entrants.
First-order effects
- Biologists studying sperm whales shift from years-long manual recording to ML-driven analysis of large vocalization datasets, making the Dominica population one of the most intensively instrumented animal groups anywhere.
- CETI's interdisciplinary roster — linguists, roboticists, AI researchers alongside marine biologists — establishes the template that other animal-communication projects now copy.
Second-order effects
- Corporate players move in: Google building DolphinGemma shows Big Tech treating interspecies decoding as a model-building problem it can fund at scale, raising the compute bar beyond academic labs.
- The approach spreads across species — fruit bats, crows, and naked mole rats are now targets of the same ML toolkit, pulling funding and talent away from traditional observational ethology toward data-intensive programs.
Third-order effects
- If decoding matures into actual two-way exchange, the practical payoff likely lands first in conservation tooling — real-time detection systems like Whale Safe already alert ships to whales in shipping lanes, and decoded vocalizations would give such systems richer signal to work with.
- Interspecies communication risks becoming a contested domain: who owns the models, the recordings, and any 'translations' is unresolved, setting up questions about access and stewardship of nonhuman communication data that regulators have not yet touched.
The trend: Animal-communication research is consolidating around large-scale machine learning — from CETI's sperm whale corpus to Google's dolphin LLM — moving decoding from niche science toward an industrialized, model-driven field.