We built a workshop tool to facilitate lessons on bias in AI labelling for Meta's policy arm, TTC Labs.

Labelling is an important step in (supervised) machine learning, but it is susceptible to bias. In collaboration with Craig Walker, we built a tool for TTC Labs that asked users to label three images within a workshop. The workshop was designed to demonstrate the need for diverse people, diverse labels and diverse images training AI.

Workshop participants are asked to categorise and label the features on three images. These labels were then used to generate an avatar from hundreds of image assets. If consent was given the results were collected in Airtable. These results were then available to further workshops to allow bias to be examined. Airtable provided an easy to use API for submitting results to a database and the perfect interface for periodic export by the design team.