Biometric Mirror

Abstract

Facial analysis applications are increasingly being applied to inform decision-making processes. However, as global reports of unfairness emerge, governments, academia and industry have recognized the ethical limitations and societal implications of this technology. Alongside initiatives that aim to formulate ethical frameworks, we believe that the public should be invited to participate in the debate. In this paper, we discuss Biometric Mirror, a case study that explored opinions about the ethics of an emerging technology. The interactive application distinguished demographic and psychometric information from people's facial photos and presented speculative scenarios with potential consequences based on their results. We analyzed the interactions with Biometric Mirror and media reports covering the study. Our findings demonstrate the nature of public opinion about the technology's possibilities, reliability, and privacy implications. Our study indicates an opportunity for case study-based digital ethics research, and we provide practical guidelines for designing future studies.

Machine Learning Processes As Sources of Ambiguity: Insights from AI Art

“…Documentation of the ML process were sometimes presented in metatext accompanying the work, such as in the websites for Machine Bias, Learning to See, POSTcard Landscapes from Lanzarote, Butcher's Son, ImageNet Roulette, in transitu, Unsupervised and Biometric Mirror. Further explanations are frequently offered elsewhere in texts written by the artists such as essays [28,88], academic papers [5,43,81,82,91] and a PhD dissertation [3].…”

“…Biometric Mirror (Fig. 3, top), created by Microsoft Research Centre for Social Natural User Interfaces [91], presents itself as a system designed to "stimulate individual reflection on the ethical application of artificial intelligence". The system invites people to have their faces photographed through a webcam and analyzed by a psychometric system, which classifies their faces on a range of dimensions from relatively overt traits such as age, gender, and ethnicity to more diagnostic concepts such as aggressiveness, weirdness, and emotional instability.…”

“…Year Technology Types of Ambiguity Tactics Butcher's Son [52] 2017 Image synthesis Information 1, 3 Learning to See [4] 2017 Image synthesis Information & Context 1, 2, 5 Machine Bias [94] 2018 Image synthesis Information & Relationship 1, 2, 4, 10 Biometric Mirror [91] 2018 Facial analysis Information, Context & Relationship 2, 3, 4, 5, 8, 10 ImageNet Roulette [69] 2019…”

“…Fig.3. From top to bottom: Biometric Mirror[91],POSTcard Landscapes from Lanzarote[43] and ImageNet Roulette[69] …”

Privacy-Enhancing Technology and Everyday Augmented Reality

“…At its most basic level, this data will unlock the ability to pseudo-anonymously identify and track nearby individuals whenever they are within sensing distance of the AR user. Backed by social media platforms, other publicly available biometric data sets and aided by cloud computing platforms, our capacity to break the veil of public anonymity will never be greater [6,7]; a prospect exacerbated by the public's limited awareness of where/how such data is used [159]. Consequentially, applications will have all the necessary data to strip bystanders of their anonymity.…”