How researchers are using self-supervised learning, a branch of AI that has proven effective for handling human language, to better understand animal sounds
New kinds of artificial intelligence are enabling scientists to better understand the sounds of the animal world, from whale songs to mouse squeaks Tweets: @mims , @drescotet , @mims , and @2morrowknight Tweets: Christopher Mims / @mims : Thanks to everyone who weighed in, I couldn't in the end resist how iconic John Krause was able to make this illustration, & how well it fit the subject, so we went with the RCA reference https://www.wsj.com/... https://twitter.com/... Miguel Angel Escotet / @drescotet : Alexa for Animals: AI Is Teaching Us How Creatures Communicate. New kinds of artificial intelligence are enabling scientists to better understand the sounds of the animal world, from whale songs to mouse squeaks • #Psychology #AI #animalpsychology • https://www.wsj.com/... • https://twitter.com/... Christopher Mims / @mims : With “Alexa for Animals”, which uses AI to parse the “speech” of animals, scientists are asking a profound new question: Is the best way to probe one alien intelligence to use another? https://www.wsj.com/... Sean Gardner / @2morrowknight : Researchers are using #AI to parse the “speech” of animals, enabling scientists to create systems that, for example, detect and monitor whale songs to alert nearby ships so they can avoid collisions. https://www.wsj.com/... #ML #DeepLearning #NLP #IoT #animals @WSJ #DYK
Context & Ripple Effects
This story lands mid-arc. Google's earlier machine-learning work on ocean data and whale-song identification showed that audio at ecological scale was parseable, and Project CETI's effort to decode sperm-whale vocalizations off Dominica framed animal communication as an explicit AI problem rather than a field-recording sideline.
What changed with the WSJ piece is the method: self-supervised learning, the approach that proved itself on human language, lets models learn structure from raw unlabeled sound — removing the hand-labeling bottleneck that kept bioacoustics small. Within months, the Times was profiling decoding work spanning fruit bats, crows, whales, and naked mole rats, with eventual plans to converse with marine animals.
First-order effects
- Research groups like Project CETI can now train on raw recordings of whale vocalizations without labeled examples, cutting the per-species cost of building acoustic models.
- Bioacoustics pipelines shift from bespoke signal-processing toward general-purpose architectures borrowed directly from human-language research.
Second-order effects
- As decoding results accumulate across species, funders and major labs begin treating animal communication as a tractable AI benchmark, pulling talent and compute away from purely text-centric NLP agendas.
Third-order effects
- The transfer pattern extends past animals: Meta's lab applying comparable decoding methods to EEG readings points to self-supervised representation learning becoming the default instrument for any natural signal — animal, neural, or otherwise — collapsing what were separate scientific silos into one modeling practice.
The trend: Language-model techniques are migrating out of human speech and text into biology, converting animal communication and brain activity into machine-learning problems.