How scientists are using ML to decode communication between fruit bats, crows, whales, and naked mole rats, eventually planning to converse with marine animals
Scientists are using machine learning to eavesdrop on naked mole rats, fruit bats, crows and whales — and to communicate back. Tweets: @dodaistewart , @nickkristof , @natalieben , @jcalpickard , @justinhendrix , @tuviae , and @justinhendrix Tweets: @dodaistewart : “If there was a big event that happened a week ago, how would we know that they're still communicating about it? Do whales do mathematics?” https://www.nytimes.com/... Nicholas Kristof / @nickkristof : Fascinating article about machine learning to understand the sophistication of animal languages and calls. Who knew that mole rats not only speak, but have multiple dialects? And some day, machine learning may enable us to speak mole or whale: https://www.nytimes.com/... Natalie Bennett / @natalieben : How little we know of complex world we're fast trashing Bats may vary vocalizations depending on relationship to & knowledge of the offender, way people might use different tones when addressing different audiences. #Biodiversity https://www.nytimes.com/... Justin Pickard / @jcalpickard : Yes, good. ‘Hidden in this everyday exchange is a wealth of social information, Dr. Barker and her colleagues discovered when they used machine-learning algorithms to analyze 36,000 soft chirps recorded in seven mole rat colonies.’ https://www.nytimes.com/... Justin Hendrix / @justinhendrix : “These experiments may also raise ethical issues, experts acknowledge. ‘If you find patterns in animals that allow you to understand their communication, that opens the door to manipulating their communications,’ Mr. Mustill said.” https://www.nytimes.com/... Tuvia Elbaum / @tuviae : A really interesting ML use case - I'm all here for Can you imagine ?? 🤯 https://twitter.com/... Justin Hendrix / @justinhendrix : “Several years ago, researchers at the University of Washington used machine learning to develop software, called DeepSqueak, that can automatically detect, analyze and categorize the ultrasonic vocalizations of rodents.” https://www.nytimes.com/...
Context & Ripple Effects
This piece sits mid-arc in a research lineage the related coverage traces clearly: Project CETI had already staked out AI-based decoding of sperm whale vocalizations off Dominica in 2021, building on earlier work applying machine learning to ocean data including Google-assisted identification of whale song. What changed by August 2022 is scope — the same techniques moving beyond cetaceans to fruit bats, crows, and naked mole rats, where Dr. Barker's team mined 36,000 soft chirps across seven colonies to identify distinct dialects.
First-order effects
- The stated endgame shifts from passive eavesdropping to two-way exchange with marine animals, putting Project CETI-style efforts at the center of an emerging conversational-bioacoustics agenda rather than pure classification work.
- Tooling built for one species generalizes fast: the University of Washington's DeepSqueak, which detects ultrasonic rodent vocalizations, shows lab-grade ML analyzers becoming reusable infrastructure across the mole-rat, bat, and crow studies.
Second-order effects
- Experts in the coverage warn that decoded communication enables manipulation of animals' signals, forcing research groups to confront ethics protocols before any playback or response experiments scale up.
- The methodological bet is on transfer from human-language AI: the self-supervised learning approaches proven on human language become the contested resource, concentrating advantage in labs with both large labeled vocalization datasets and NLP expertise.
Third-order effects
- If the pattern holds through results like the later machine-learning-derived sperm whale alphabet, interspecies communication matures from novelty into a structured research field with its own governance questions around consent, interference, and who may speak for a species.
- Vocalization datasets and decoding models become strategic scientific assets in their own right, analogous to how foundation-model training corpora concentrate power — a dynamic echoed in coverage of labs studying LLMs themselves as objects of study.
The trend: Animal-communication research is consolidating around language-model techniques borrowed from human NLP, moving from one-way decoding toward governed two-way exchange with wild species.