A genealogy of the ambiguous use of digital memory in machine learning and artificial intelligence

The aim of this paper is to discuss the epistemological risk that we face by building representations of social phenomena that exclude contingency and unpredictability from the frame. Datification is the new representation tool within social sciences. The objectives of data science are the classification of what have happened and the prevision of what will happen. The anticipation of the future is not confined to irrelevant marketing areas, it includes very sensible situations in which social and human rights are involved. Humans can assume the responsibility of their perspective, humans can change their mind, they can assess the variability and multiplicity of preferences and find negotiated political solutions to common problems, though their judgements are fallible. Humans possess the ‘sense of possibility’ introduced by Musil that, according to Bernard Stiegler, we can interpret as a “thought within entropy”. This algorithmic attitude towards future risks to reduce the adaptability of human beings. If we accept to base every judgement on a calculation, even when circumstances do not allow a complete and exhaustive measurement of all element of the situation, we risk misunderstanding the context, being incapable of reacting appropriately when facing external, unpredictable conditions. We might not face the radicality of contingency.