Machine learning has the potential to replace up to 80% of doctors by being more accurate and objective than the average physician. Diagnostic errors cause over 40,500 adult patient deaths in ICUs annually in the US, higher than deaths from HIV or firearms injuries. Doctors are subject to cognitive biases that can influence diagnosis, but machine learning could help by running diagnostic tests, generating hypotheses based on patterns in data, and providing a "second opinion" to reduce errors from biases. While human interaction and empathy are important, machine learning may be able to assist doctors in reducing diagnostic errors.