Intelligence articielle, mégadonnées et gouvernance

Abstract : Algorithmic data processings, presented by their promoters as immune to the irrational mistakes of the human spirit, would be out of reach of its understanding and could not see their results challenged. In view of the alleged benefits, great is the temptation to base the management of human societies on them. However, like any artifact, these data processings are shaped by human biases, which impact their results. These biases concern equally the definition of the model, the selection of the data chosen to train the inductive processings, and the configurations to which these processings converge. As such, their self-configuration phase must be dissociated from their production phase, otherwise it would not be possible to reproduce the results, or even explain their operation in a transparent manner. The inherently conservative nature of these processings, essentially designed to reproduce the past in the future, condemns their use beyond tasks limited to a very constrained problem space. The time of algorithmic governance has not come yet, provided that human beings do not back away from their responsibility.
Document type :
Journal articles
Liste complète des métadonnées

https://hal.inria.fr/hal-02012785
Contributor : François Pellegrini <>
Submitted on : Saturday, February 9, 2019 - 1:08:32 PM
Last modification on : Tuesday, February 12, 2019 - 9:44:47 AM

File

megadonnees-gouvernance_201611...
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02012785, version 1

Citation

François Pellegrini. Intelligence articielle, mégadonnées et gouvernance. Revue Lamy Droit de l'immatériel, Lamy (imprimé) / Wolters Kluwer édition életronique 2018, pp.56-59. ⟨hal-02012785⟩

Share

Metrics

Record views

37

Files downloads

53