Qualification and Quantification in Machine Learning. From Explanation to Explication

Abstract

Moving beyond the conundrum of explanation, usually portrayed as a trade-off against accuracy, this article traces the recent emergence of explainable AI to the legal “right to an explanation”, situating the need for an explanation in the underlying rule of law principle of contestability. Instead of going down the rabbit hole of causal or logical explanations, the article then revisits the Methodenstreit, whose outcome has resulted in the quantifiability of anything and everything, thus hiding the qualification that necessarily precedes any and all quantification. Finally, the paper proposes to use the quantification that is inherent in machine learning to identify individual decisions that resist quantification and require situated inquiry and qualitative research. For this, the paper explores Clifford Geertz’s notion of explication as a conceptual tool focused on discernment and judgment rather than calculation and reckoning.

Keywords

• gdpr
• right to an explanation
• explainable machine learning
• methodenstreit
• qualculation
• proxies
• explication