Within the frame of the AI Forensics project, Exposing.ai undertook a hybrid effort. On the one hand, it developed the Exposing.ai search engine for Flickr as part of the Forensic toolkit. This tool directly interfaces with Exposing.ai’s other component, a sustained sociotechnical investigation into biometric/surveillance datasets, their provenance, distribution, and use.

Exposing.ai (sociotechnical case studies)
Datasets

Investigations into 34 datasets used to train AI surveillance models

Read below
Exposing.ai (toolkit)
Search engine

9,231,981 Flickr photos from 28 image datasets

See in toolkit

Sociotechnical case studies

A pattern emerged in Exposing.ai’s investigations into systems developed for facial recognition technologies: the algorithms were being trained on hundreds of millions of images scraped from the internet, mostly from Flickr.com, where permissive and unprotected “Creative Commons” licenses were routinely and deliberately exploited for biometric data.

The project therefore features extensive sociotechnical investigations into a growing number of these datasets. Each dataset features a report, in quantitative and qualitative terms, that describes its genesis, broader statistics, and use in the training of biometric surveillance models.

The project thus helps tell a small part of the story of face recognition: what it is, how it became powerful, and how Creative Commons continues to play a major role in biometric data proliferation.

Explore:

Embedded from https://exposing.ai/datasets

Support

Exposing.ai is run by Adam Harvey and Jules LaPlace. In the frame of the AI Forensics project, it was institutionally hosted by HfG Karlsruhe’s KIM, they developed it on the basis of the previous MegaPixels project. It has also received support from the Weizenbaum Institut, Mozilla, the Surveillance Technology Oversight Project (S.T.O.P.), and the Copenhagen Business School's AI REUSE project.