Challenges and Prospects of Deploying AI and Machine Learning for Clinical Diagnosis in African Healthcare

Abstract

The integration of artificial intelligence (AI), machine learning (ML), and robotics into clinical diagnosis has become prevalent. For example, ML-driven image recognition has demonstrated remarkable efficacy, prompting clinicians to rely increasingly on these technologies for “accurate” medical diagnoses and prognoses of diseases. Although these advancements have exhibited their relevance and effectiveness in medically advanced regions of the Global North and selected areas in the Global South, the question arises as to their viability within the healthcare landscape of Africa, given contextual variations. In this paper, I delve into the potential efficiency of deploying these technologies within African healthcare, aiming to address these contextual concerns. Employing a phenomenological methodology, I demonstrate that the deployment of these technologies might inadvertently introduce biases and
discrimination against Africans. This stems from the inherent nature of the data used to develop these technologies, primarily sourced from healthcare experiences in designing nations, coupled with the pervasive algorithmic biases prevalent in contemporary ML systems. I call for a paradigm shift in AI and ML development. I propose that African nations should proactively engage in the design of healthcare AI and ML technologies that are attuned to distinct African conditions, prevalent medical conditions, and prognostic methodologies. Key prerequisites include the establishment of robust infrastructure for efficient data collection and storage of electronic healthcare records and capturing the intricacies of day-to-day healthcare encounters across the African continent.