Four equity considerations for the use of artificial intelligence in public health

Maxwell J. Smith, Renata Axler, Sally Bean, Frank Rudzicz, James ShawBulletin 2020 · DOI

TL;DR

This paper argues that the integration of Artificial Intelligence (AI) in public health requires careful consideration of equity. It proposes four key dimensions – the digital divide, algorithmic bias, plurality of values, and fair decision-making – to ensure AI benefits all populations and doesn't exacerbate existing health disparities.

What Problem Does This Paper Address?

The rapid advancement and adoption of AI technologies in healthcare, particularly in public health, present a dual potential: to either improve or worsen existing health inequities. Public health initiatives often target entire populations rather than individuals, making the impact of AI on collective well-being and the distribution of benefits and burdens particularly significant. Unlike previous technological shifts, AI's reliance on big data and machine learning introduces novel pathways through which health inequities can emerge or be amplified. The paper addresses the critical need to proactively identify and understand these unique equity considerations before widespread implementation, moving beyond general discussions of AI ethics to a public health-specific framework.

Methods

The paper does not present a novel algorithmic method or empirical study. Instead, it offers a conceptual framework. The authors' "method" involves a critical analysis of the potential equity implications of AI in public health, drawing on existing literature and ethical guidance. They propose a categorization of these implications into four distinct, yet interconnected, considerations. This approach aims to provide a starting point for researchers, policymakers, and practitioners to systematically evaluate and address equity concerns.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • AI's Unique Equity Landscape: The authors emphasize that AI's impact on health inequities is not simply a continuation of past trends seen with other technologies. The nature of big data and machine learning creates new challenges and opportunities for equity.
  • Public Health's Collective Focus: Because public health often deals with populations, AI's influence can be amplified, affecting entire communities rather than just individuals. This necessitates a broader equity lens.
  • Two Core Aims for Equitable AI: The paper advocates for a dual approach:
  • Negative Aim: Design and implement AI to prevent the creation, sustainment, or exacerbation of health inequities.
  • Positive Aim: Design and implement AI to actively redress or eliminate existing health inequities and promote health equity.
  • The Digital Divide is More Than Access: It's not just about who can use AI tools, but also about who benefits from the insights and interventions they generate, and who bears the burdens. This divide can exist between different socioeconomic groups, geographic locations, and even between researchers and the public.
  • Algorithmic Bias is Deeply Embedded: AI systems learn from data, and if that data reflects societal biases (e.g., regarding race or gender), the AI will perpetuate and potentially amplify them. This raises questions about what constitutes "fairness" in AI outputs.
  • Plurality of Values Creates Inequities: Different health systems and cultures may embed different values into AI technologies. This can lead to varied outcomes across systems, potentially creating new forms of inequity based on where one receives care.
  • Fair Processes Over Perfect Outcomes: Given the difficulty in achieving consensus on what "fair" AI looks like, the paper suggests prioritizing fair and inclusive processes for AI development and implementation. This includes transparency, accountability, and active engagement with underrepresented groups.

Key Results

Not applicable, as this paper is a conceptual framework and not an empirical study with quantitative results.

Limitations and Future Work

The authors acknowledge that their four considerations are not exhaustive and that the specific nuances of individual AI technologies and their deployment contexts will further shape equity concerns. They suggest that these considerations serve as a starting point for reflection. Future work should focus on:

  • Detailed Mapping: Further research is needed to map and understand the specific equity considerations that arise with different AI technologies in various public health settings.
  • Capturing Multiple Perspectives: Efforts to address equity must actively incorporate the diverse perspectives of all populations who will be impacted by AI in public health practice, policy, and research.
  • Developing Practical Frameworks: Translating these conceptual considerations into actionable guidelines and tools for developers, implementers, and policymakers is a crucial next step.

Summary updated

BibTeX
@article{smith2020four,
  author = {Maxwell J. Smith and Renata Axler and Sally Bean and Frank Rudzicz and James Shaw},
  title = {{Four equity considerations for the use of artificial intelligence in public health}},
  journal = {Bulletin},
  year = {2020},
  doi = {10.2471/blt.19.237503},
}
Figure