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 conditions for social claim accountability.

Two goals—improving replicability and accountability of Machine Learning research respectively—have accrued much attention from the AI ethics and the Machine Learning community. Despite sharing the measures of improving transparency, the two goals are discussed in different registers: replicability registers with scientific reasoning whereas accountability registers with ethical reasoning. Given the existing challenge of the responsibility gap—holding Machine Learning scientists accountable for Machine Learning harms due to them being far from sites of application—this paper posits that reconceptualizing replicability can help bridge the gap. Through a shift from model performance replicability to claim replicability, Machine Learning scientists can be held accountable for producing non-replicable claims that are prone to eliciting harm due to misuse and misinterpretation. In this paper, I make the following contributions. First, I define and distinguish two forms of replicability for ML research that can aid constructive conversations around replicability. Second, I formulate an argument for claim-replicability’s advantage over model performance replicability in justifying assigning accountability to Machine Learning scientists for producing non-replicable claims and show how it enacts a sense of responsibility that is actionable. In addition, I characterize the implementation of claim replicability as more of a social project than a technical one by discussing its competing epistemological principles and practical implications on Circulating Reference, Interpretative Labor, and research communication.

[working abstract] This study is motivated by two concerns. First, while discussions of AI hype abound across academia, industry, and civic research community, definitions of hype remain fragmented. Second, the deployment of Machine Learning–based tools in policing has drawn critical attention for issues such as privacy, inequity, lack of accountability. Through document analysis of communication artifacts across multiple communities for four policing tools, this study makes three contributions. First, it offers a novel definition of hype that foregrounds the relational dynamics of technological discourse. Second, it demonstrates how this relational conception of hype provides an analytic frame for diagnosing shortcomings in the communicative dimensions of ML-based policing tools. Finally, it identifies three problematic discursive practices in the communication of AI-based policing tools and suggests how rectifying them can advance responsible communication.

[working abstract] In machine learning (ML) research, broad social claims about potential societal benefits or harms are often made without clear articulation of scope, supporting evidence, or context, limiting both transparency and foresight for application harms. This paper proposes a path toward new forms of ML accountability by focusing on the concept of social claim replicability: the ability to systematically document, organize, and evaluate the evidence supporting social claims. First, we developed a social claim typology with a list of commonly cited supporting evidence to guide the articulation and substantiation of claims. Second, we proposed evidence diagram for systematizing supporting evidence, thereby improving clarity and rigor of research communication. By integrating these approaches, we aim to create conditions for establishing social claim accountability in ML research.