Gulls Become Test Pilots for AI-Driven Behavioural Research
On-board machine learning and multi-sensor loggers enable seabirds to initiate and record playback trials over land and sea.
A long-standing challenge in behavioural research
For decades, audio playback experiments have been a cornerstone of field ornithology, helping scientists explore how birds respond to sounds such as predator calls, alarm notes, or courtship songs. But traditional methods require researchers to be close enough to observe responses or to set up static devices in fixed locations, restricting the range of situations that can be studied.
A Japanese research team has now unveiled an innovative solution – a bio-logger worn by the bird itself that uses artificial intelligence to detect a target behaviour in real time and then automatically play a sound and record the response. The study, published in Ecology and Evolution, demonstrates the concept with Black-tailed Gulls Larus crassirostris on Kabushima Island, Japan.
How the system works
The custom-built bio-logger combines multiple sensors, including an accelerometer, GPS, camera, microphone and speaker. A lightweight on-board computer runs a machine-learning model trained to recognise when a bird is flying. Once this behaviour is detected for several seconds, and the bird is confirmed to be away from the breeding colony, the device begins recording video and other sensor data. After a minute’s pre-playback recording, it plays either a Peregrine Falcon Falco peregrinus call or white noise, then continues recording for another minute to capture the bird’s reaction.
This “AI on Animals” approach removes the need for human observers or fixed playback points, enabling experiments in situations previously impossible to study – such as birds foraging far from shore or moving unpredictably through the air.
Field trials with gullsTen male Black-tailed Gulls were fitted with waterproofed devices weighing under 4.31% of their body mass. Across eight successful deployments, the bio-loggers carried out 46 playback experiments, all while the birds were flying outside the colony. In 20 other cases, playback was correctly cancelled when the bird stopped flying, saving battery life. The machine-learning behaviour recognition performed well, with an F1-score of 0.91 for identifying flight.
Analyses of acceleration, GPS speed changes, and head movement in the video footage revealed that some individuals responded clearly to the sounds – not only predator calls but also white noise – suggesting a startle effect from the unusual source of the sound.
Potential and next steps
The authors note that sound played from directly on a bird’s back may be perceived differently from natural sources, potentially explaining the mixed responses. They suggest refining playback types and volumes – for example, using conspecific alarm calls – to elicit more ecologically relevant reactions.
Looking ahead, the technology could be adapted to detect other behaviours, stimulate different senses, or even run interactive experiments that adjust playback in response to the bird’s real-time behaviour. Such tools could also find applications in conservation, for example by testing non-lethal deterrents to reduce human–wildlife conflict.
A proof of concept with broad implications
While the current study focused on a single seabird species and a specific behaviour, the autonomous playback bio-logger opens up new possibilities for studying wildlife responses in the wild, untethered to human presence or static infrastructure. As the researchers put it, it offers “a variety of playback experiments to be tested in various situations” – potentially transforming how we investigate animal communication, decision-making, and behavioural ecology.
August 2025
Share this story
