NLM DIR Seminar Schedule

Abstract:

The promise of artificial intelligence (AI) in healthcare, from diagnosis to treatment optimization, is undeniable. However, as AI technologies like LLMs and medical imaging AI become integral to clinical practices, their inherent biases pose significant challenges. These biases can exacerbate healthcare disparities, making the pursuit of equity in AI applications not just a technical challenge but a moral imperative.
Our talk will cover two studies. The first one reveals biases in language models predicting healthcare outcomes, showing a tendency to replicate societal disparities in treatment recommendations and prognoses. The second one addresses fairness in medical imaging AI, introducing a causal fairness module that improves equity by adjusting for biases related to sensitive attributes without compromising diagnostic performance.
Addressing biases in AI is crucial for ensuring these technologies serve all patients fairly, regardless of their background. Our studies highlight the importance of continual assessment and adjustment of AI models to reflect ethical considerations alongside technical advancements.