EU law states “the obligation of the administration to give reasons for its decisions”. This echoes a rule of law principle common to many constitutional traditions and conditional to fundamental rights related to powers interfering with personal freedom (for example, as concretised in the right of access, judicial review, nondiscrimination, self-determination). The increasing trend of asking algorithms to take decisions affecting personal rights (either in the private or public sphere) is deeply changing the legal reasoning on the limitation of powers. This paper would like to explore the existing global constitutional law principles on algorithmic decision-making. It inquires whether these principles are effective in tackling new challenges coming from deep-learning algorithms where “causation” is replaced by “correlation”, therefore when we use no humanly comprehensible reason for decisions.