PhilSci Archive

Understanding (and) Machine Learning's Black Box Explanation Problems

Boge, Florian J. (2025) Understanding (and) Machine Learning's Black Box Explanation Problems. [Preprint]

[img] Text
UndestandingBBExplProblemPre.pdf

Download (558kB)

Abstract

Machine learning (ML) is a major scientific success. Yet, ML models are notoriously considered black boxes, where this black boxness may refer to details of the ML model itself or details concerning its outcomes. Hence, there is a flourishing field of "eXplainable Artificial Intelligence" (XAI), providing means for rendering several aspects of ML more transparent. However, given their tremendous success,
why would we even want to explain black boxed ML models with XAI? I here suggest that, in order to answer this question, we first need to distinguish between proximate and ultimate aims in using XAI: While the proximate aim may be uniformly
to provide instruments for explaining aspects of ML to relevant stakeholders, the ultimate aim varies with the context of deployment. Furthermore, I argue that in
science, the ultimate aim is the understanding of scientific phenomena. I then sketch three paths along which understanding of phenomena may be gained by means of
ML and XAI. In a coda, I address the possibility of gaining understanding from ML directly, without explanations and XAI.


Export/Citation: EndNote | BibTeX | Dublin Core | ASCII/Text Citation (Chicago) | HTML Citation | OpenURL
Social Networking:
Share |

Item Type: Preprint
Creators:
CreatorsEmailORCID
Boge, Florian J.florian-johannes.boge@udo.edu0000-0002-1030-3393
Keywords: Machine Learning; Artificial Intelligence; XAI; understanding; aims of science
Subjects: General Issues > Data
Specific Sciences > Artificial Intelligence
Specific Sciences > Artificial Intelligence > Machine Learning
Depositing User: Prof. Dr. Florian Boge
Date Deposited: 18 Aug 2025 12:54
Last Modified: 18 Aug 2025 12:54
Item ID: 26251
Subjects: General Issues > Data
Specific Sciences > Artificial Intelligence
Specific Sciences > Artificial Intelligence > Machine Learning
Date: 2025
URI: https://philsci-archive.pitt.edu/id/eprint/26251

Monthly Views for the past 3 years

Monthly Downloads for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item