Evaluation of the sensitivity of cognitive biases in the design of artificial intelligence - Archive ouverte HAL Access content directly
Conference Papers Year : 2021

Evaluation of the sensitivity of cognitive biases in the design of artificial intelligence

Abstract

The reduction of algorithmic biases is a major issue in the field of artificial intelligence. Despite the diversity of sources and temporalities at the origin of these biases, the current solutions to achieve ethical results are mostly technical. The consideration of human factors and more particularly cognitive biases remains incomplete. Nevertheless, the task of designing artificial intelligence systems is conducive to the emergence of cognitive biases. The aim of our study is to test the awareness of individuals who design artificial intelligence systems of the impact of their cognitive biases in their productions. The study focuses on conformity bias, confirmation bias and illusory correlation bias. The first results of this pre-experimentation show that these individuals believe that their decisions are subject to the cognitive biases of conformity, illusory correlation and confirmation.
Fichier principal
Vignette du fichier
RJCIA_2021_paper_22.pdf (356.43 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03298746 , version 1 (23-07-2021)

Identifiers

  • HAL Id : hal-03298746 , version 1

Cite

M Cazes, N Franiatte, A Delmas, J-M André, M Rodier, et al.. Evaluation of the sensitivity of cognitive biases in the design of artificial intelligence. Rencontres des Jeunes Chercheurs en Intelligence Artificielle (RJCIA'21) Plate-Forme Intelligence Artificielle (PFIA'21), Jul 2021, Bordeaux, France. pp.30-37. ⟨hal-03298746⟩
247 View
361 Download

Share

Gmail Facebook Twitter LinkedIn More