Google announced three major initiatives aimed at building equity into their health-related artificial intelligence work during their annual health event, The Check Up. These efforts seek to address disparities in healthcare outcomes that disproportionately affect people of color, women, rural communities, and other historically marginalized populations.
The company’s Chief Health Equity officer highlighted the importance of developing AI-powered health tools responsibly and equitably. The first initiative involves a new research paper providing a framework to identify and mitigate biases in medical large language models. This includes a set of adversarial testing datasets called “EquityMedQA” to help evaluate potential biases.
Secondly, Google introduced HEAL (Health Equity Assessment of Machine Learning performance), a framework designed to assess the likelihood of AI technology performing equitably across different subpopulations. The four-step process aims to prevent the deployment of AI models that could exacerbate existing health disparities.
Lastly, Google partnered with Stanford Medicine to create the Skin Condition Image Network (SCIN), a more representative dataset for dermatology. This open-access collection includes over 10,000 real-world images contributed by thousands of individuals, representing diverse skin tones, ages, genders, and conditions.
These initiatives address the underrepresentation of certain groups in medical datasets and AI models, which can lead to biased outcomes. By focusing on creating more inclusive tools and datasets, Google aims to contribute to a healthier future for everyone, regardless of background or location. The company emphasizes that this is an ongoing journey and plans to continue refining these approaches in collaboration with partners.
See “3 ways we are building equity into our health work” (March 19, 2024)
