Can Machine Learning be Moral?

TL;DR

This paper examines the ethical implications of machine learning, proposing that supervised learning is the only ethically defensible approach when evaluated through a relational ethics lens.

Contribution

It introduces a relational ethics perspective to evaluate ML practices and argues that supervised learning uniquely aligns with ethical accountability.

Findings

Supervised learning is identified as the most ethically defensible ML method.

Relational ethics provides a novel framework for assessing ML practices.

The paper challenges the ethical acceptability of other ML methodologies.

Abstract

The ethics of Machine Learning has become an unavoidable topic in the AI Community. The deployment of machine learning systems in multiple social contexts has resulted in a closer ethical scrutiny of the design, development, and application of these systems. The AI/ML community has come to terms with the imperative to think about the ethical implications of machine learning, not only as a product but also as a practice (Birhane, 2021; Shen et al. 2021). The critical question that is troubling many debates is what can constitute an ethically accountable machine learning system. In this paper we explore possibilities for ethical evaluation of machine learning methodologies. We scrutinize techniques, methods and technical practices in machine learning from a relational ethics perspective, taking into consideration how machine learning systems are part of the world and how they relate to different forms of agency. Taking a page from Phil Agre (1997) we use the notion of a critical technical practice as a means of analysis of machine learning approaches. Our radical proposal is that supervised learning appears to be the only machine learning method that is ethically defensible.