Paper short abstract:

This paper examines the development of facial recognition for bears, analyzing how bear facial recognition it was identified as a tractable technological problem, the consequences of its deployment for ecological research, and production of species divides in facial recognition technology.

Paper long abstract:

Scientists are developing an ever expanding “internet of animals,” a collection of tools that includes machine learning, AI, and GPS-telemetry. These technologies have changed the way animal life is tracked and quantified, fundamentally transforming scientific understandings of wildlife. These new technologies include the use of machine learning for animal facial recognition. While many big tech companies are growing more circumspect about the use of facial recognition for humans, interest in animal facial recognition is surging. This paper examines the development of facial recognition for bears, termed “FaceNet for bears” by its developers (a reference to Google’s human facial recognition program). It analyzes how bear facial recognition was identified as a tractable technological problem for computer programmers and ecologists and the consequences of its deployment for ecological research. Bear facial recognition is made possible by developments in camera technology, the popularity of bears on social media, and the artificial neural networks making image recognition possible. It is meant to assist ecologists in studying individual bears and changes in overall population size and health, improving on human ability to recognize and process data about individual bears. This paper examines how bear facial id is like and unlike human facial recognition. Machine learning and AI for conservation are often siloed and treated as if the ethical concerns that apply to human-directed technology are irrelevant; this paper examines how divisions between humans and other life forms are created in the design of technologies and when and how these divides are transgressed or breakdown.