The digitalisation of medicine is both an opportunity and a risk for patients. The
democratisation of the use of LLMs such as Chat GPT makes it easier for individuals to find
medical information online. However, these digital tools need to be monitored, both in their
development and their deployment, against the demonstrated risks of providing false
information and propagating bias. There is an abundant literature on algorithmic biases.
Scholars have highlighted the spread of biases related to gender, origin or based on other
legally protected characteristics. Less attention has been paid to linguistic bias. Yet the
interactions between culture and medicine are numerous. Failure to consider cultural factors
including language or dialect can lead to inequitable health predictions for the populations
concerned. Beyond biases, the acceptability of digital medicine should imply that systems are
developed with regard to the cultural and linguistic norms of the target population. The
presentation will contribute to deconstructing the myth of neutral AI by establishing the link
between linguistic bias and health inequity on the one hand, and the myth of AI "neutralization"
on the other hand, by advocating for an active integration of cultural - especially linguisticfactors
in algorithmic systems in order to build adapted thus sustainable medical AI.