Skip to Main content Skip to Navigation
Conference papers

Interprétabilité et explicabilité pour l’apprentissage machine : entre modèles descriptifs, modèles prédictifs et modèles causaux. Une nécessaire clarification épistémologique

Abstract : The lack of validation and of explainability of some Machine Learning (ML) models involves operational, legal and ethical issues. One of the main objectives of our research project is to provide ethical explanations of the outputs produced by a ML based application, considered as a black box. The first step of this project, presented in this article, is to underline the epistemic differences between the validation of an ML model and of a causal mathematical model. ML is based on statistical correlations between input - which could have a high dimension - and output parameters without building causality links between them, unlike most mathematical models in science and engineering. This absence of causality is the major drawback of ML, making difficult the validation and the explanation of some ML methods, generally the most efficient ones. Our scientific contribution is to highlight, in this context, the epistemic distinctions between the different functions of a model, on the one hand and between the function and the use of a model, required to build explanation. Our current work in the evaluation of the quality of an explanation, which could be more persuasive than informative and consequently generates ethical problems, is reported in the last part of the article.
Complete list of metadatas

Cited literature [26 references]  Display  Hide  Download

https://hal.sorbonne-universite.fr/hal-02184519
Contributor : Christophe Denis <>
Submitted on : Friday, September 27, 2019 - 4:07:04 PM
Last modification on : Tuesday, July 28, 2020 - 4:18:02 PM
Long-term archiving on: : Monday, February 10, 2020 - 4:32:16 AM

File

CNIA_2019-Denis&Varenne.pdf
Publisher files allowed on an open archive

Identifiers

  • HAL Id : hal-02184519, version 1

Citation

Christophe Denis, Franck Varenne. Interprétabilité et explicabilité pour l’apprentissage machine : entre modèles descriptifs, modèles prédictifs et modèles causaux. Une nécessaire clarification épistémologique. National (French) Conference on Artificial Intelligence (CNIA) - Artificial Intelligence Platform (PFIA), Jul 2019, Toulouse, France. pp.60-68. ⟨hal-02184519⟩

Share

Metrics

Record views

924

Files downloads

693