Interrogating AI Fairness and Bias in Dermatology and Beyond with Dr. Roxana Daneshjou
Jan 17, 2024
auto_awesome
Dr. Roxana Daneshjou, Assistant professor at Stanford specializing in dermatology, shares her journey in AI research. She discusses AI fairness and bias in dermatology and the challenges of integrating language models in healthcare. Dr. Daneshjou emphasizes the need for interdisciplinary collaboration and highlights the importance of addressing disparities in AI performance across diverse skin tones.
Addressing systemic bias in AI systems is crucial for healthcare fairness.
Representation and diversity in datasets are essential for improving fairness in dermatology AI.
Combining human expertise and AI capabilities can help reduce bias in large language models.
Deep dives
The importance of building fair AI systems in healthcare
Roxanna Dineshjou, an assistant professor of biomedical data science at Stanford University and a practicing dermatologist, emphasizes the need for building fair AI systems in healthcare. She discusses how biases in human data can be picked up by large language models, highlighting the importance of addressing systemic issues of bias in both humans and technology. While AI has the potential to exacerbate disparities, Roxanna stresses the need for interdisciplinary collaboration and the integration of domain-specific expertise to ensure fairness in AI in healthcare.
Examining algorithmic bias in dermatology AI
Roxanna Dineshjou's research focuses on AI fairness and bias in dermatology. In a study, she curated a diverse dataset of clinical images to evaluate how dermatology AI systems perform across different skin tones. The findings revealed significant differences in performance between white skin tones and brown/black skin tones. Roxanna emphasizes the need for representative datasets and benchmarks to address AI bias and improve fairness in dermatology AI.
Challenges of algorithmic bias in large language models
Roxanna Dineshjou discusses the potential bias in large language models (LLMs) and highlights a paper she co-authored on how LLMs can perpetuate race-based medicine. The study found that LLMs, when queried on medical topics involving race, often provided incorrect and biased responses. Roxanna emphasizes the importance of interdisciplinary collaboration and designing systems that reduce bias by combining human expertise and AI capabilities.
The role of clinicians in AI development
Roxanna Dineshjou asserts that the development and implementation of AI in medicine requires collaboration between clinicians and computer scientists. She emphasizes the need for domain-specific expertise in healthcare AI and the importance of clinicians understanding AI capabilities and limitations. Roxanna also highlights the potential of AI to assist clinicians in tasks such as documentation, allowing more focus on patient care and improving overall healthcare experiences.
Balancing academic career and parenting
Roxanna Dineshjou shares her experience juggling an academic career and raising small children. She highlights the need for support from institutions to accommodate working parents and underscores the importance of flexibility and work-life balance. Roxanna emphasizes the value of transparency and open communication with trainees about the challenges and realities of being a working parent in academia.
In this episode of the AI Grand Rounds podcast, Dr. Roxana Daneshjou shares her journey from a childhood influenced by early exposure to science to her current role as an assistant professor at Stanford. Her path includes a critical shift during medical school, where her interest in computational methods and human genomics led her to pursue both an M.D. and a Ph.D. Her specialization in dermatology was driven by its visual nature and the opportunity to form long-term relationships with patients. Dr. Daneshjou emphasizes the importance of AI in addressing fairness and bias in dermatology, discussing her research on disparities in AI performance across diverse skin tones. The podcast also delves into broader issues of AI in health care, discussing the potential and challenges of integrating large language models into medical practice, and highlighting the need for interdisciplinary collaboration between clinicians and computer scientists in AI development. Dr. Daneshjou’s optimism for the future centers on the new generation of medical professionals who are increasingly concerned about fairness and equity in AI.