The creators of FaceApp, a photo filtering app which uses “neural networks” to manipulate people’s images, have apologised after it became apparent that the software was lightening people’s skin tones in the app’s ‘hot’ mode.
The company behind the app blamed “training set bias” for the program’s tendency to lighten skin-tone and narrow the noses of its users.
“We are deeply sorry for this unquestionably serious issue,” the company said in a statement to several media organisations.
The statement implies that the company fed predominantly European faces into the software during development.
“It is an unfortunate side-effect of the underlying neural network caused by the training set bias, not intended behaviour,” the company told media.
“To mitigate the issue, we have renamed the effect to exclude any positive connotation associated with it.”
Neural networks are computer systems designed to mimic the processes of the human brain and nervous system.
The filter has been renamed to a ‘spark’ emoji.
Technology companies are increasingly using forms of sophisticated artificial intelligence to analyse and process images.
The technology can be trained to recognise objects or elements within an image, and edit, remove or replace those elements automatically without disturbing background elements.
It’s not the first time a popular image editor has fallen into strife over the technology, Snapchat has been criticised for several filters which appeared to alter the race of users.
