Abstract
With large language models, smart surveillance watches and automated content moderation, AI entered a new phase. However, the increasing attention these solutions have gained is not only attached to their benefits. The adverse impacts of algorithms raised discussions in scientific literature and media outlets. The design of AI algorithms can perpetuated biases and prejudices from their creators, amplifying oppression on a massive scale and potentially affecting millions in mere moments. Historically marginalized groups bear the brunt of these harmful consequences. In the context of facial recognition, gender minorities, including non-binary and transgender individuals, face unwarranted scrutiny as applications attempt to infer their identity based on cis-heteronormative standards. In this paper, we report the results of a systematic mapping study centred on the key issues that transgender individuals encounter when dealing with facial recognition applications. A total of 24 primary studies provided us with 4 main problems and 20 underlying causes, which we represented via Map of Problems and Ishikawa Diagram techniques. This diagnostic assessment enabled us to derive a set of guidelines to be followed by requirements engineers and software developers to ensure more ethically inclusive AI solutions. Besides, we believe it supports future research and practice in Requirements Engineering (RE) field, providing valuable insights for enriching the early requirements phase.
