Medical AI needs more datasets to improve accuracy and help clinicians, but collecting and labeling medical data poses privacy concerns. Researchers are exploring self-supervised and generative AI techniques to address these challenges by generating synthetic medical data without privacy issues. As language models and foundation models continue to grow in scale and capability, they may help advance medical AI by generating diagnostic reports, summarizing studies, and aiding discovery through analysis of vast amounts of data. However, privacy and bias remain important issues to address with further development of these techniques.