Investigating the cultural adaptability of Generative AI in Type-2 Diabetes heal
Abstract:
Healthcare is an umbrella term capturing various (medical) attempts to strengthen and/or restore physical or mental bodily needs with the goal of improving health. It is an essential part of life, and good healthcare is often correlated with good quality of life. While it is mostly preventative, some healthcare is remedial and this can be life-long. Diseases like Type-2 diabetes (T2D) require life-long healthcare that includes significant changes to the patient’s lifestyle. Technological healthcare interventions have long proven to be effective in the short and long run, and Generative Artificial Intelligence (GenAI) models show potential in providing tailor-made healthcare solutions to each patient according to their individual needs. We examine the use of such models in generating stories and images. T2D patients can use these as part of their behavioural change tools as they develop and/or maintain new lifestyles to manage the disease. Two GenAI models are used, Dolphin-Mistral to generate stories, and DreamShaper XL to generate the accompanying images that patients can use in their Diabetes management. We present a method for these models to capture and interpret contextual and cultural prompts. The generated stories and images are measured for their contextual and cultural appropriateness. We find that while there are a lot of challenges, these models are able to produce contextual and culturally appropriate material, even for low-resource contexts and cultures which don’t contribute much to the models’ training.