Abstract
The question that interests me here is whether human values, such as privacy or non-discrimination, can be translated into computational metrics and how such translation affects the protection offered by text-driven norms (such as law). This question raises the issue of how the assumptions of computer science, notably those concerning computability, interact with the assumptions of human language and text, notably those concerning the fundamental ambiguity of meaning. This, in turn, relates to the issue of temporality, that seems to play a different role in text and computation; whereas a machine learning algorithm cannot be trained on future data, human language seems inherently focused on integrating anticipation and remembrance. The latter should not be reduced to a prediction based on mathematical extrapolation of historical data, but rather in terms of imagination and its constitutive impact on the world we share (understood in terms of institutional facts). This will, finally, lead me to a juxtaposition between the affordances of computational systems and those of text, arguing that both generate different types of normativity.