How researchers are using self-supervised learning, a branch of AI that has proven effective for handling human language, to better understand animal sounds
Context & Ripple Effects
The Wall Street Journal piece is the earliest marker in a research arc that ran through 2022–2024: applying self-supervised learning — the technique behind modern language models — to animal vocalizations. Five months later, the New York Times reported scientists decoding communication in fruit bats, crows, whales, and naked mole rats, with some eventually aiming to converse with marine animals.
The significance goes beyond bioacoustics. Quanta Magazine reported computational neuroscientists finding self-supervised models correspond more closely to brain function than supervised ones, and the same learning-from-unlabeled-signals logic later showed up in Meta's AI lab decoding EEG readings and in models trained on infant headcam footage to study language acquisition. Animal sound is one instance of a broader move: AI learning structure from raw, unlabeled signals.
First-order effects
- Researchers studying animal sounds gain a method that works without labeled datasets — the bottleneck that previously limited bioacoustics to species where humans had already annotated calls.
Second-order effects
- The same self-supervised toolkit spreads across adjacent signal domains: the NYT's animal-communication work, EEG decoding at Meta's AI lab, and infant language-acquisition studies all reuse the approach, turning bioacoustics into a proving ground for cognition research more broadly.
Third-order effects
- If self-supervised models keep tracking brain function better than supervised ones — as computational neuroscientists told Quanta — the field's center of gravity shifts toward learning from raw sensory streams, with animal communication, neural signals, and child development studied under one methodological umbrella.
The trend: AI research is extending self-supervised learning from human language to every raw signal domain — animal vocalizations, brain activity, infant speech — as a unified method for decoding natural communication.