AI for all: bridging data gaps in machine learning and health

Artificial intelligence (AI) and its subset, machine learning, have tremendous potential to transform health care, medicine, and population health through improved diagnoses, treatments, and patient care. However, the effectiveness of these technologies hinges on the quality and diversity of the data used to train them. Many datasets currently used in machine learning are inherently biased and lack diversity, leading to inaccurate predictions that may perpetuate existing health disparities. This commentary highlights the challenges of biased datasets, the impact on marginalized communities, and the critical need for strategies to address these disparities throughout the research continuum. To overcome these challenges, it is essential to adopt more inclusive data collection practices, engage collaboratively with community stakeholders, and leverage innovative approaches like federated learning. These steps can help mitigate bias and enhance the accuracy and fairness of AI-assisted or informed health care solutions. By addressing systemic biases embedded across research phases, we can build a better foundation for AI to enhance diagnostic and treatment capabilities and move society closer to the goal where improved health and health care can be a fundamental right for all, and not just for some.