From the ridiculous to the sinister, Riitta Oittinen’s project, where craft meets technology, reminds us of the limitations and bias in computational ‘seeing’ systems. In 2014 Riitta experimented with Google image recognition using photos of handmade mini rugs – curious to see how accurately their images would be interpreted. The results were surprising. None returned other textile imagery and the searches revealed the system’s bias towards humans and household objects. It was also highly gendered. Reds returned female forms and blues male. Yellow and peachy colours also returned exposed human flesh (predominantly female) yet browns and blacks did not. Three years later, in 2017, she ran the experiment again to see whether Google’s algorithms had improved. Again, Riitta did not get any matches with her rugs.
The installation at the ODI presents her original rugs (made of new and recycled materials: pieces of cloth, candy and food wrappings, shoe laces, bits and pieces of decorative ribbon, nylon, paper, and cotton string, wool and other natural fibres) alongside the photographs and two sets of results from the searches, 2014 on the left and 2017 on the right. With current concerns of widespread image recognition and surveillance, it is intriguing that a simple rug appears to be a mystery to Google’s machine learning software. We are asked to consider what other imagery is unable to be recognised by these seemingly ‘intelligent’ systems.
Handmade mini rugs, digital prints
2018 at the Open Data Institute
Identifiable, Object, Processed, Shared
Riitta is a social historian and science journalist with a special interest in visual and vernacular culture. Her work is mainly documentary photography, but she also likes to play with the computer to create different kind…