Seeing / Saying 
text and images
Seeing / Saying examines the development of datasets used for machine vision and AI training. Amazon’s Mechanical Turk (MTurk) is a web service through which requesters can create human intelligence tasks (HITs) for workers to complete anonymously online. Through this platform AI developers task anonymous workers with annotating thousands of pictures, by “objectively” labeling the discrete objects they observe. To questions the presumptions of such utilities I created my own image identification HIT using personal family photos. Instead of asking them to refain from conjecture or emotional response I instead asked workers to infer what they thought was happening and to explain any personal responses they had to the images. Shown here are selected responses to my two questions paired with the photographs given. 

Summary of results from the HIT on MTURK