Managing the risks of inevitably biased visual artificial intelligence systems