Selective ‘Machine Unlearning’ could help scrub privacy-eroding algorithms and even delete the inferences gained from the processed data