Attitudes and Folk Theories of Data Subjects on Transparency and Accuracy in Emotion Recognition

Abstract

The growth of technologies promising to infer emotions raises political and ethical concerns, including concerns regarding their accuracy and transparency. A marginalized perspective in these conversations is that of data subjects potentially affected by emotion recognition. Taking social media as one emotion recognition deployment context, we conducted interviews with data subjects (i.e., social media users) to investigate their notions about accuracy and transparency in emotion recognition and interrogate stated attitudes towards these notions and related folk theories. We find that data subjects see accurate inferences as uncomfortable and as threatening their agency, pointing to privacy and ambiguity as desired design principles for social media platforms. While some participants argued that contemporary emotion recognition must be accurate, others raised concerns about possibilities for contesting the technology and called for better transparency. Furthermore, some challenged the technology altogether, highlighting that emotions are complex, relational, performative, and situated. In interpreting our findings, we identify new folk theories about accuracy and meaningful transparency in emotion recognition. Overall, our analysis shows an unsatisfactory status quo for data subjects that is shaped by power imbalances and a lack of reflexivity and democratic deliberation within platform governance.

Emotion AI atWork: Implications forWorkplace Surveillance, Emotional Labor, and Emotional Privacy

“…To ensure fair treatment of workers, emotion AI technologies should be used fairly and ethically [143]. Fair and ethical use of emotion AI may include commitments by actors deploying emotion AI systems that it is meaningfully consented to [23]; that its (potentially biased, unreliable, and inaccurate [14,38,110,120]) information is transparent and contestable [31,66,143]; and that its use does not widen power asymmetries, such as those already present between workers and their employers [4,31]. Yet, so far, the use of emotion AI in workplaces remains largely unconstrained and unregulated, and in the modern US workplace, the growing adoption of emotion AI-enabled workplace surveillance is predicted to become the new norm [157].…”

Constructing Certainty in Machine Learning: On the performativity of testing and its hold on the future

“…For example, emotion recognition algorithms are highly contested as emotions are situated, personal, relational, and complex and thereby cannot be simply measured as neatly observable categories (Grill and Andalibi 2022;Boehner et al 2007), but some companies and researchers still claim that through algorithmic classification of facial expressions, emotions can be automatically and universally recognized (Stark 2019a). Since folk theories suggest that facial expressions in images often correspond to assumed universal emotions, some advocates of the technology were able to convince others of referring to facial expression classification as emotion recognition.…”

“…Other important examples include the term "ground truth" which implies that usually human-labeled data represent some absolute truth about the world (Jaton 2017), "data-driven," which suggests that human judgment does not matter, or the term prediction, which insinuates that ML algorithms are able to forecast the future reliably (Chun 2021). This issue also extends to how basic research problems in the field are named, e.g., emotion recognition wrongly implies that actual emotions are recognized by algorithms (Grill and Andalibi 2022). Thus, there is a need for the careful renaming of concepts and terms to improve descriptions of the capabilities and limits of algorithms for different audiences.…”

“…This achievement to reinvent accuracy led to a self-fulfilling prophecy as funding was focused on inertial guidance, making it seem, in retrospect, as though it was the best and inevitable choice. This example illustrates how accuracy is imagined and understood matters in the construction of facts (Grill and Andalibi 2022) and thereby is a form of discursive power shaping what is considered true. Consequently, already naming a quantitative metric for quality in ML "accuracy" can be also understood as an act of power to shape how algorithms and their accuracy is imagined.…”

“…Companies also perform efficacy and accuracy beyond testing through demonstrations that often also just highlight cherry-picked results that appear promising to others or they use strong phrasing and promises to be convincing. This kind of messaging is particularly effective with those that already hold techno-deterministic beliefs about how technologies such as AI are inevitable and will create a better future (Lindtner 2020; Avle et al 2020;Grill and Andalibi 2022).…”