IBD Salle 15
AMU - AMSE
5-9 boulevard Maurice Bourdet
Feriel Kandil: feriel.kandil[at]univ-amu.fr
Miriam Teschl: miriam.teschl[at]ehess.fr
Algorithmic prediction is very different from the idea of prediction established in modern society since the 18th century. Whereas in the modern view the future is open and unknowable because it does not yet exit and depends on present actions and expectations, predictive algorithms promise to know the future in advance. Machine learning, although using almost exactly the same tools as statistics, nonetheless, has some resemblance to the magical and divinatory mentality of pre-modern societies. Like divination, algorithmic prediction does not address averages and general trends, but attempts to give precise indications about the future of a single event or individual on which it directly intervenes. This is the source of the power and of the specific liabilities of algorithmic prediction, connected to risks of pre-emption and overfitting. In these cases, the prediction, even if correct, may prove ineffective or even harmful.