Using Identification with AR Face Filters to Predict Explicit and Implicit Gender Bias

Sun, 01 Oct 2023 00:00:00 +0000

What was tested. The study separates the filter's gender presentation from the participant's reported identification, then examines their relationship with implicit association, androcentrism, and explicit gender-norm agreement.

The diagram makes the article’s central point visible: AR self-transformation is not treated as a single intervention. Its effects must be understood in relation to the particular filter, the person using it, and the outcome being measured.

AR identity, face filters and societal biases

Wed, 20 Apr 2022 08:59:34 +0000

Marie Skłodowska-Curie postdoctoral research at IMT Atlantique and Lab-STICC, examining the psychological consequences of everyday AR self-transformations.

Postdoctoral research project · 2021-2023 ARAVIDEN Augmented Reality, avatars, identification, and the social consequences of changing one's own appearance.

At a glance

Social-media face filters are one of the most widespread forms of Augmented Reality (AR). They alter a person’s visible appearance while preserving their physical setting and their own body - a form of self-representation that differs fundamentally from replacing the user with a full virtual avatar in Virtual Reality (VR).

ARAVIDEN investigates embodied identification: the extent to which people experience an augmented version of themselves as part of who they are. The project examines how this identification changes self-perception, attitudes toward social groups, and cognitive performance - while treating these effects as context-dependent rather than assuming that an AR transformation will automatically create empathy or reduce prejudice.

Research questions

Identification in AR - How do people identify with face filters that modify their age, gender presentation, or social role while their physical body and environment remain visible?
Social attitudes - Which combinations of user characteristics, transformation, and identification are associated with implicit and explicit attitudes toward gender and age groups?
Performance and self-representation - Can brief AR cues, such as virtual scientific attire, affect engagement or performance on cognitive tasks?
Responsible design - How can researchers and designers avoid treating stereotypical appearance changes or short-term filter interactions as universal bias-reduction interventions?

Research approach

The project uses controlled, mixed-design user studies with Snapchat face filters. Participants interact with no-change and transformed versions of their face, complete behavioural or cognitive tasks, and report their degree of embodied identification. The studies combine validated implicit measures, including Implicit Association Tests (IATs), with explicit-attitude questionnaires and performance measures.

This approach deliberately separates identity cues that are often conflated. In particular, the age study distinguishes virtual aging from the visual cues associated with a scientist role, allowing their individual and combined effects to be examined.

Transform the visible selfFilters independently altered apparent age and scientist-role cues while keeping the user's real surroundings visible.

Measure what changesFilter interactions were paired with attitude measures, identification questionnaires, and a timed cognitive task.

Key findings

Gender study · 119 analysed participants

Identification was not a universal bias intervention

Gender-manipulated filters did not reliably change implicit gender associations or androcentrism. Identification was associated with explicit gender-norm agreement in patterns that depended on both the participant and the filter.

Age and role study · 105 analysed participants

Virtual attire mattered more for performance than age bias

Brief virtual aging and scientist-role cues did not reliably reduce age bias. Higher identification with scientist attire was, however, modestly associated with mental-arithmetic performance.

Project lead

Marie Skłodowska-Curie postdoctoral fellow · 2021-2023

Marie Jarrell

Marie led ARAVIDEN at the intersection of human-centred computing, immersive media, and social psychology. Her work used familiar social-media AR to test how people relate to altered versions of themselves - and what follows from that identification.

Scientific collaboration with Étienne Peillard, IMT Atlantique and Lab-STICC.

Pictures

News

Funding

European Union · Horizon 2020This project received funding through a Marie Skłodowska-Curie Actions grant (No. 899546).

Social XR | Étienne Peillard