Risk in AI-mediated medical translation

Abstract Concerns about the future of AI implementation, particularly with the explosion of generative AI practices, are the result of the high impact AI is having in all areas of society and the need for debate and reflection about the role of technology in human practices. This paper addresses the medical translation field and the risks associated with the use and integration of AI technologies. To do so, it takes an interdisciplinary perspective that includes the human-centered AI (HCAI) paradigm in translation studies (e.g., Jiménez- Crespo, 2023; O’Brien, 2023), responsible AI (Arrieta et al., 2020), and AI for Social Good (Hager et al., 2019). More specifically, it reflects on areas where human agency is key at the lexical level in AI-mediated translation processes. In order to achieve this purpose, this paper reviews the notion of risk from the perspective of the translation of medical texts and their users, with emphasis in the multimodal forms of communication, which are in continuous growth and are often at the center of non-supervised automatic machine translated practices. This is illustrated with examples from a corpus analysis of human and AI solutions of English-Spanish translations of multimodal texts on mental health, an extremely sensitive and high-stakes domain where existing biases and stigma demand special attention. The focus of the analysis is on aspects such as the comparison of professional and AI translators’ solutions when dealing with terminological variation, interference, metaphor, cultural adequacy or multidimensionality. These key areas illustrate the importance of the human role in rendering appropriate solutions for different users’ profiles, including the role of creativity, an introspective human-specific skill, in promoting critical thinking and avoiding the bias and stigma of information on mental illness. Ultimately, the results highlight the need for a closer collaboration between technology and the humanities. This collaboration is needed to guarantee ethical practices in AI as well as to develop AI literacy in Translation Studies, and it should include the analysis of high-stakes areas in specific domains and the detection of risk and ways to tackle it.