Examining the post-privacy world

In a recent paper, Wang and Kosinski claim that machine learning classifiers can accurately detect sexual orientation from facial images. They believe that due to the growing digitalisation of our lives and the rapid progress in artificial intelligence, we are inevitably headed towards a world in which privacy has been completely eroded, “the post-privacy world”. This thesis examines this post-privacy narrative, by questioning whether predictions by machine learning classifiers can violate one’s privacy, if we assume that the access account of privacy is correct. This thesis shows that predictions by machine learning classifiers could potentially violate privacy. First, to make the predictions, machine learning classifiers have to be trained, which is often done using data that is taken out of context, breaching contextual integrity and privacy in the process. In addition, the existence of a machine learning classifier that could uncover private information does not take away the reasonability of a claim to privacy with respect to this information. However, predictions by machine learning classifiers have acquired an unjust epistemic status. Due to this, the danger exists that predictions by machine learning classifiers are assumed to be privacy invasive, even when there is no strong evidence that they are.