Training AI to ask clinical questions as a new method to simplify physicians' work

MIT researchers have created a public database of more than 2,000 clinically relevant questions for physicians to help them more efficiently find information in a patient's medical record without spending a lot of time navigating. The database was created in collaboration with medical experts who are competent in the list of questions physicians ask when reviewing electronic medical records.

Now, the team is using this work to create a model that can automatically answer physician questions in the EHR. In the next phase, they will use their data set to train a machine learning model that can automatically generate thousands or millions of good clinical questions. They will then be used to train a new model to automatically answer questions.

The research work is to be presented soon at the annual conference of the North American Chapter of the Association for Computational Linguistics.