Analyzing Technosolutionism in Synthetic Face Data Companies

As long as facial recognition technologies (FRT) have existed, they have struggled with gender and racial bias–and thus, also with barriers in critique and regulation, leading to limited product efficacy and growth. Such biases exist due to training datasets lacking demographic diversity, which leads to misclassification of underrepresented groups in various applications of FRT in society [4]. Robust training datasets, however, require appropriate protocols for obtaining and handling human data, and are thus costly and time-consuming to create [1] [3]. In response, FRT companies have turned to image-based synthetic data generation methods as a proposed solution, claiming it enhances fairness while sidestepping ethical concerns related to data privacy and consent [1] [3] [8]. Though synthetic data is framed as a superior, even moral, alternative to real human data, Whitney and Norman (2024) argue that these claims are often without sufficient scrutiny of ethical and technical limitations, “appearing to resolve valid criticism about a dataset's distribution and representation … in a way that is superficial” [9].