Dead Zone of Accountability: Why Social Claims in Machine Learning Research Should Be Articulated and Defended

TL;DR

This paper examines the disconnect between social claims made by ML research and the actual impacts, proposing accountability mechanisms to address the claim-reality gap and improve social responsibility.

Contribution

It introduces the 'dead zone of accountability' concept, diagnosing resistances to accountability and proposing collaborative research agendas for social claim accountability in ML.

Findings

Identification of cognitive and structural resistances to accountability

Introduction of the 'dead zone of accountability' lens

Proposals for collaborative research to enhance accountability

Abstract

Many Machine Learning research studies use language that describes potential social benefits or technical affordances of new methods and technologies. Such language, which we call "social claims", can help garner substantial resources and influence for those involved in ML research and technology production. However, there exists a gap between social claims and reality (the claim-reality gap): ML methods often fail to deliver the claimed functionality or social impacts. This paper investigates the claim-reality gap and makes a normative argument for developing accountability mechanisms for it. In making the argument, we make three contributions. First, we show why the symptom - absence of social claim accountability - is problematic. Second, we coin dead zone of accountability - a lens that scholars and practitioners can use to identify opportunities for new forms of accountability. We apply this lens to the claim-reality gap and provide a diagnosis by identifying cognitive and structural resistances to accountability in the claim-reality gap. Finally, we offer a prescription - two potential collaborative research agendas that can help create the condition for social claim accountability.