Assessing the impact of cognitive biases in AI project development - Pôle Data, Models, Information, Decisions Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

Assessing the impact of cognitive biases in AI project development

Résumé

Biases are a major issue in the field of Artificial Intelligence (AI). They can come from the data, be algorithmic or cognitive. If the first two types of biases are studied in the literature, few works focus on the last type, even though the task of designing AI systems is conducive to the emergence of cognitive biases. To address this gap, we propose a study on the impact of cognitive biases during the development cycle of AI projects. Our study focuses on six cognitive biases selected for their impact on ideation and development processes: Conformity, Confirmation, Illusory correlation, Measurement, Presentation, and Normality. Our major contribution is the realization of a cognitive bias awareness tool, in the form of a mind map, for AI professionals that address the impact of cognitive biases at each stage of an AI project. This tool was evaluated through semi-structured interviews and Technology Acceptance Model (TAM) questionnaires. User testing shows that (i) the majority admitted to being more aware of cognitive biases in their work thanks to our tool, (ii) the mind map would improve the quality of their decisions, their confidence in their realization, and their satisfaction with the work done, which impact directly their performance and efficiency, (iii) the mind map was well received by the professionals, who appropriated it by planning how to integrate it into their current work process: for awareness-raising purposes for the onboarding process of new employees and to develop reflexes in their work to question their decision-making
Fichier principal
Vignette du fichier
2023-02-24-HCII2023-bernault_juan_chraibiK.pdf (1.74 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
licence : CC BY NC SA - Paternité - Pas d'utilisation commerciale - Partage selon les Conditions Initiales

Dates et versions

hal-04007898 , version 1 (28-02-2023)

Licence

Paternité - Pas d'utilisation commerciale - Partage selon les Conditions Initiales

Identifiants

  • HAL Id : hal-04007898 , version 1

Citer

Chloé Bernault, Sara Juan, Alexandra Delmas, Jean-Marc Andre, Marc Rodier, et al.. Assessing the impact of cognitive biases in AI project development. HCI 2023: 25th International Conference on Human-Computer Interaction, Jul 2023, Copenhague, Denmark. ⟨hal-04007898⟩
86 Consultations
302 Téléchargements

Partager

Gmail Facebook X LinkedIn More