Skip to content
CrowLingo

Frontier · Sub-page

Ethics — the floor, not the ceiling.

Crows are recognized individuals. They remember. Every interaction we run is part of their experience of humans — permanently, for years. The rules below are the absolute floor.

AI narration · Frontier · Ethics floor

The ethical constraints we hold to: no playback experiments without IACUC review, no nest-proximity recording, no city-precise location publication, no commercial use of behavioral data, no claim of translation. We name these explicitly because the AI-assisted bioacoustic space has a recurring failure mode — the field moves fast, the popular framing outruns the science, the actual animals catch the harm. The Earth Species Project's launch framing for NatureLM-audio is the cleanest public version of these constraints. CrowLingo's editorial floor is the same. If a feature on the v1-through-v5 roadmap would violate one of these, it doesn't ship.

AI interpretation, not translation.
Vertical editorial card titled Six Rules for Listening Back with six numbered ethical rules for crow playback.
IG · 08 · SIX · RULES
The six rules in one card. Print it. Tape it to the recorder. They apply at v0 and at v5; they don't expire.

The six rules

  1. No playback near active nests in breeding season. Reproductive stress is the highest-stakes animal-welfare cost a playback study can incur. The default is "no"; the exceptions are vanishingly rare and require ethics-board justification.
  2. Never simulate alarm calls in public habitat. Alarm calls recruit. A simulated alarm in habitat triggers a real recruitment response in animals who cannot distinguish stimulus from event. The cost is wasted vigilance budget, disrupted foraging, and cumulative habituation that compromises future genuine signals.
  3. Limit sessions to under 90 seconds per location per week. Habituation and individual recognition together mean a site you visit for playback this week is a contaminated site for the next 2–4 weeks. Rotate sites; log the visit calendar.
  4. Always observe response; stop on distress. Distress signals are well-documented (sustained alarm calls, recruitment of multiple individuals, sudden silence + extended fixation, direct flight at speaker or observer). Any of them = halt. The halt is itself a result.
  5. Publish raw recordings, methods, and outcomes openly. Closed-data behavioral work on animals reproduces the pre-replication crisis of psychology, but with the added cost that we're imposing experiments on subjects who can't consent. Open publication is the price of doing it at all.
  6. When in doubt, listen more, broadcast less. Most questions in crow vocal communication can be answered with observational audio + behavioral data. Playback is a last-resort intervention, not a default methodology.

What CrowLingo will not do

The four constraints below are project-wide policy. They apply to v0, v1, v5. They are not aspirational.

  • No alarm/distress playback files in trivially redistributable formats. Audio that's easy to broadcast at scale by a non-researcher is a tool for harassment. We host such audio only as analysis-quality source for vetted collaborators, never as MP3 downloads.
  • No location data finer than city-level. Crow family groups can be located precisely from geo-tagged audio. We coarsen all location metadata to the city level before publication.
  • No endorsement of commercial "crow translator" apps. No such product currently rises above entertainment. Several mislead users about the state of the field. We don't link to them; we don't list them in the library.
  • No anthropomorphism."The crow said" is shorthand we don't use. "The crow produced a call of type X in context Y" is the boring, correct phrasing — and the boring phrasing is what protects the claim.

How to flag a concern

If you see CrowLingo content (or playback protocols published here) violating this floor, the right contact is contact@kymatalabs.com. Replies within seven days. Errata posted on the page in question.

Frequently asked

What people ask about this.

Is it ethical to do audio playback experiments with wild crows or near active nests?
Playback near active nests is effectively prohibited — reproductive stress is the highest-stakes welfare cost for wild birds, and the default position across the corvid research community is no playback during the breeding season near nest sites, with vanishingly rare exceptions requiring formal ethics-board justification and documented research-question necessity. Outside nesting contexts, playback can be ethically defensible if conducted as a pre-registered experimental protocol with observer, time-bounded sessions, immediate halt on distress signals, and disclosed reporting. CrowLingo's editorial floor declines all wild-crow playback in product features; the atlas is record-and-observe, not call-and-respond. The Demartsev et al. carrion-crow wearable-logger methodology represents the modern non-playback approach to acoustic-context research.
Why should AI bioacoustic systems never simulate or reproduce crow alarm calls?
Alarm calls function as recruitment signals — a vocalization recruiting conspecifics to mob a perceived threat. A simulated alarm triggers real recruitment from crows who cannot distinguish a synthesized stimulus from an actual predator event, producing wasted vigilance behavior, disrupted foraging, cumulative habituation that erodes the call's biological function, and stress on the entire local population. The risk compounds across deployments: an AI app that plays alarm calls 'just to listen to crows' in a residential neighborhood produces measurable population-level harm. This is the most-cited ethical floor across modern corvid research and the field's most-consistent position; CrowLingo's atlas reproduces the principle in product design (no playback features anywhere on the site).