California's COVID-19 vaccine distribution algorithm used ZIP codes instead of census tracts to allocate additional vaccine supply, potentially excluding over 2 million people in vulnerable communities, many of them communities of color, from receiving needed additional vaccine doses.
In January 2021, California decided to use the Healthy Places Index to identify areas needing additional COVID-19 vaccine supply, targeting the bottom 25% of communities by health outcomes. The state initially planned to use census tracts but later announced that Blue Shield would build an algorithm allocating vaccines based on ZIP codes rather than census tracts. This algorithmic decision had significant consequences because ZIP codes represent larger geographical areas containing both low-income and wealthy communities, while census tracts are specifically designed for socioeconomic comparisons. Analysis showed that using ZIP codes could exclude 469 census tracts where over 2 million Californians live from receiving additional vaccine supply. These excluded communities are disproportionately communities of color, with Latinx people making up 53% of potentially omitted areas compared to 38% statewide, and Black people comprising 8% versus 6% statewide. The affected areas span from Susanville in Lassen County to Chinatown in San Francisco to El Cajon near San Diego. The state justified using ZIP codes for operational simplicity in tracking vaccine delivery, but critics argued this choice undermines equity goals and could have life-and-death consequences for vulnerable populations.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Unequal treatment of individuals or groups by AI, often based on race, gender, or other sensitive characteristics, resulting in unfair outcomes and unfair representation of those groups.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed