Twitter investigates its image cropping algorithm for racial bias
Twitter investigates its image cropping algorithm for racial bias
Twitter is investigating complaints of racial bias in its image cropping algorithm. The algorithm is used to resize and reframe images, so that they focus on what Twitter’s systems consider to be the most important parts of an image.
Users’ experiments, however, reveal that the algorithm frequently prioritises the face of a white person over a black person in its image crops when both faces appear in the same image.
Concerns about the risks of bias in artificial intelligence technologies led IBM to announce in June that it will no longer offer general purpose facial recognition software, while in August, the Court of Appeal in R (Bridges) v The Chief Constable of South Wales Police [2020] EWCA Civ 1058, [2020] All ER (D) 26 recognised that such software can sometimes have an inbuilt bias that needs to be tested for.
The issues reported in Twitter’s image processing technology again bring to the fore the potential risks of unlawful discrimination and bias that can arise when artificial intelligence systems are improperly designed, developed or tested. For further analysis of this issue refer to my recent article, ‘Automating bias?’.