Continuing with the impact of AI at the community and population level, in this paper the authors review how artificial intelligence can influence health equity, examining both its potential to reduce disparities and the risks that poorly designed models may exacerbate existing inequalities. The paper highlights how AI systems increasingly analyze diverse data sources—including clinical, laboratory, sociodemographic, administrative, and community-level information—to support prevention, diagnosis, and treatment decisions. While these capabilities may enable more personalized and context-aware care, the review emphasizes that biased or incomplete datasets can lead to inequitable outcomes, particularly for racial and ethnic minorities, individuals with disabilities, and low-income populations who are often underrepresented in training data. The authors describe key sources of bias such as limited data diversity, reliance on proxy variables, and population heterogeneity, all of which can affect model performance across subgroups and influence downstream clinical decisions.
From a public-health and healthcare-system perspective, the implications are substantial. AI has the potential to improve equity by enabling targeted interventions, enhancing clinical decision support, and tailoring care to the needs of underserved communities. However, the review underscores that without careful design, evaluation, and governance, AI tools may perpetuate or even worsen disparities. The authors highlight strategies to promote equitable AI, including assessing performance across diverse populations, incorporating community engagement throughout development, and embedding equity considerations across the entire AI lifecycle. They also stress that meaningful stakeholder involvement and transparent validation processes are essential to build trust and ensure fair deployment. Overall, the paper positions AI as a powerful but double-edged tool—capable of advancing health equity when responsibly implemented, but also capable of reinforcing structural inequities if bias mitigation and inclusive design are not prioritized.
