Collective Memory

Collective Memory is a distributed network of contributors who document their surroundings (everyday life, environmental change, and events as they unfold) through a capture application that records provenance at the moment of observation. The corpus spans more than 18 million geolocated clips from 174 countries.

As co-founder I lead research and development across the arc of the project: the application as an instrument of field ethnography, methods for verification and curation, and the release of datasets under a model of contributor remuneration. The aim is an ethically sourced ground truth for machine learning, in which consent, origin and compensation are part of the record.

The project sits between several fields. For open-source investigation, it offers observation of known origin where coverage is limited, censored or contested, including sustained documentation in Gaza and Ukraine. For computer vision and geospatial research, it offers redundancy: the same places recorded by many people, devices and seasons, which supports geolocation, change detection and 3D reconstruction. For anthropology, it raises questions of representation: whose world is recorded, from which positions, and to what end.

The corpus, pipeline and tooling are proprietary; selected datasets are released openly on Hugging Face. I'm glad to discuss the work further in conversation.