Project Title

Sociotechnical Health-specific AI Impact Assessment

Organization

Confidential

About

Applications of AI in public health and health care are diverse and range from public health surveillance, to scheduling and decision support. As with all technological change, however, AI systems can also result in unintended consequences, from the exacerbation of health inequities through widening digital divides, to changes to professional work practices.

As applications of AI in health care continue to grow, ensuring responsible and trustworthy design, development, and use that balances risks and benefits will be essential. That is what this algorithmic impact assessment (AIA) set out to do. This AIA did not set out to solely assess model performance. It is intended to assess broader health system impacts of an AI system based on its potential uses.

Approach

This impact assessment was informed by a structured review of the research literature on health-related AI risks, benefits, and harms, and existing impact assessments. More than 200 sources were included in the final review.

AIA Domains

The AIA is comprised of a number of domains and sub-domains, each with their own questions, including:

  • Sponsorship

  • Intended Uses

  • Development Team

  • Value Proposition

  • Data Provenance

  • Model Development

  • Implementation Setting

  • Implementation Team

  • Reliability, Transparency, Safety, Explainability

  • Security

  • Maintenance, Monitoring, and Oversight

Governance Models

Different governance models of the AIA were also developed, each implicating different intended uses and users, including:

  • Evaluation of Individual AI Technologies

  • Informing Responsible Development & Implementation

  • Assessing Conformity with Legislation / Regulation

  • Condition of Access to Datasets for AI Model Training

Project Status

This project is complete.