Scientist wins $100,000 for decoding birdsong
Dr. Julie Elie of UC Berkeley's Theunissen Lab won the 2026 Coller-Dolittle Prize, a $100,000 award for progress toward two-way animal communication, after using machine learning to decode zebra finch vocalizations into 11 core call-types covering meanings like distress, hunger, and greeting. Elie's team validated the findings by testing the birds themselves: zebra finches correctly recognized individual companions from single calls and re-sorted the call classifications by meaning rather than raw acoustics, confirming the calls function as an identity-plus-intent code. The prize, run by the Jeremy Coller Foundation and Tel Aviv University, carries a $10 million grand prize for any team that achieves full two-way communication with an animal. Elie's peer-reviewed findings appeared in the journal Science.
For AI/ML practitioners, this result is a concrete example of applying classification methods to a genuinely messy real-world acoustic dataset, then closing the loop by validating the model's output against the study subjects themselves rather than trusting the classifier alone.
What happened
Dr. Julie Elie of the Theunissen Lab at UC Berkeley won the 2026 Coller-Dolittle Prize, a $100,000 annual award from the Jeremy Coller Foundation and Tel Aviv University for progress on two-way interspecies communication. Working with more than a decade of zebra finch recordings, Elie used machine learning to sort the birds' vocalizations into 11 core call-types tied to specific meanings, such as distress, hunger, aggression, and greeting, and to detect individual vocal signatures layered on top of each call type. The findings were peer-reviewed and published in the journal Science (DOI: 10.1126/science.ads8482).
Technical context
According to Scientific American, Elie built a classifier that could sort calls using acoustic features alone, but the model struggled to separate calls with similar sounds and different meanings, such as aggression versus distress. She addressed this by folding in behavioral context, the situation the bird was in when it called, alongside the raw acoustics. Elie then validated the resulting taxonomy behaviorally: birds were tested on whether they could identify individual companions from a single call type, and whether they agreed with Elie's call classifications. In both tests, she reports, the birds performed above chance and effectively confirmed the model's groupings.
For practitioners
The pattern here, pairing an acoustic/ML classifier with independent behavioral validation rather than trusting model output alone, generalizes to any bioacoustic, sensor, or signal-classification pipeline where ground truth cannot be directly observed. Elie described the tradeoff plainly: machine learning helps capture acoustic differences, but context about the animal's behavior is what resolves the ambiguous cases the acoustic model alone could not.
What to watch
The prize sets a Turing-test-style bar beyond this result: a $10 million grand prize (or a $500,000 cash alternative) awaits a team that achieves genuine two-way communication with an animal, a much higher threshold than one-way call classification. Three other 2026 finalist teams, working on African striped mice, chimpanzees, and bonobos, used comparable AI-based approaches, suggesting a broader trend of machine learning applied to animal-communication research worth tracking for anyone working in bioacoustics or signal processing.
Key Points
- 1Elie used machine learning to sort over a decade of zebra finch recordings into 11 call-types, then had the birds themselves validate the classifications behaviorally.
- 2Zebra finches sort calls by meaning rather than acoustic similarity, and use individual vocal signatures that let companions recognize who is calling.
- 3The $10 million Coller-Dolittle grand prize rewards genuine two-way communication with an animal, a much higher bar that AI-assisted bioacoustics research is only beginning to approach.
Scoring Rationale
A niche but genuinely ML-driven research result: peer-reviewed in Science, with a real machine-learning classification methodology and multi-source verification including the official prize institution. Impact on mainstream AI/ML practice is minimal, keeping it in the minor tier despite the stronger sourcing found this audit (up from 4.2).
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems