Rapid progress has been made in artificial intelligence (AI) models for medical applications, especially over the past five years, with substantial efforts focusing on diagnosis from medical images. Most of the current AI models were developed using reference standards based on subjective evaluation. In this talk, we will discuss a different paradigm where AI development and evaluation focus on using outcomes as reference standards instead of subjective interpretation. We’ll use concrete examples of machine learning in healthcare research from Google to elaborate on this outcomes-oriented AI paradigm. Lastly, we will discuss the potential of using such technology to discover new medical knowledge.