Aarhus University Seal / Aarhus Universitets segl

Datafication in HMC between representation and preferences: Designing gender neutrality in voice-controlled assistants

Publikation: Bidrag til bog/antologi/rapport/proceedingBidrag til bog/antologiForskningpeer review

  • Julie Mortensen, Aarhus Universitet, Danmark
  • Nicoline Siegfredsen, Aarhus Universitet, Danmark
  • Anja Bechmann
Datafication in Human-Machine Communication (HMC) most often relates to data used for training the AI of the system and is associated with problems of cultural and representational biases; either because historical data have been used to increase the amount of training data or because data is representing the normal rather than the marginal or abnormal behavior (Bechmann and Bowker 2017; Campolo and Crawford 2020; Crawford and Calo 2016; Henriksen and Bechmann 2020; Sweeney 2013). However, a less prominent dimension of datafication in HMC that distinguishes it from human-human communication is the ability to vary how the system is represented on the user interface and user preferences for such interfaces, potentially varying contextually depending on the task at hand. Focusing on voice-control as a dominant and growing user interface within HMC this chapter will discuss key theoretical, methodological and analytical challenges in datafying this layer by investigating gender neutrality in voice-controlled assistants. Using gender neutrality as a case study will allow the chapter to discuss fundamental issues of datafication of the interface between user representation and preferences. The chapter will discuss theoretically how to represent gender neutrality in voice- controlled assistants connecting to both gender and HMC literature, followed by a small quasi-experiment with 12 females, males and non-binary users testing to what extent user preferences match the theory in the field. The experiment only partially supports the theory, yet it allows for a discussion of key concepts within HMC (e.g. uncanny valley, contextual awareness) and methodological challenges of testing HMC.
TitelHandbook of Human-Machine Communication
StatusAccepteret/In press - 2021

Se relationer på Aarhus Universitet Citationsformater

ID: 194926246