Can AI Be Harmful? A Conversation with MIT’s Dr. Marzyeh Ghassemi
NEJM AI Grand Rounds
00:00
The Performance Gap Between Black and White Patients
In a paper, the authors found that when you use contextual embeddings to process notes between different kinds of patients, there's a performance gap. They say it is extremely unlikely that we are going to remove all human biases from large language models. The field has changed over the last few years since they published this article but still faces challenges in terms of how to address these issues.
Transcript
Play full episode
