In an article for the Journal of the American Medical Informatics Association (JAMIA), Dr. Joachim Roski, Booz Allen health analytics leader; Dr. Ezekiel Maier, Booz Allen analytics leader; Dr. Kevin Vigilante, Booz Allen chief medical officer; Elizabeth Kane, Booz Allen health operations expert; and Vanderbilt University Medical Center's Dr. Michael Matheny, present insights that organizations of industry stakeholders can use to adopt self-governance as they work to maintain trust in artificial intelligence (AI) and prevent an "AI winter."

Industry stakeholders see AI as critical for extracting insights and value from the ever-increasing amount of health and healthcare data. Organizations can use AI to synthesize information, support clinical decision making, develop interventions, and more—creating high expectations for AI technologies to effectively address any health challenge. However, throughout the history of AI development, streaks of enthusiasm have been followed by periods of disillusionment. During these AI winters, both investment in and adoption of AI best practices wane.

Today, publicity around highly touted but underperforming AI solutions has placed the health sector at risk for another AI winter. To respond to this challenge, we propose that industry organizations consider implementing self-governance standards to better mitigate risks and encourage greater trust in AI capabilities.

Building on the National Academy of Medicine’s AI implementation lifecycle, we created a detailed organizational framework that identifies 10 groups of AI risks and 14 groups of mitigation practices across the four lifecycle phases. AI developers, implementers, and other stakeholders can use this analysis to guide collective, voluntary actions to select, establish, and track adherence to trust-enhancing AI standards.

Without industry self-governance, government agencies may act to institute their own compliance requirements. However, industries that have proactively defined, adopted, and implemented standards complementary to government regulation have reduced the urgency of public-sector action while allowing for the appropriate use of available resources. Industry self-governance also enables exceptional agility to respond to evolving technologies and markets.

When considering self-governance, there are a number of key success factors to take into account. These include the creation of an industry-sanctioned certification and accreditation program. It’s also important to understand that self-governance success is based on stakeholder confidence that all standards and methods have been developed in coordination with consumers and patients, clinicians, AI developers, AI users, and other key parties. 

While AI advancement continues with government support, there are also signs of a technology backlash, underscoring the need to mitigate AI-related risks. The government-led management of such public risks occurs in various ways; however, targeted, AI-specific legislation does not yet exist. Diverse organizations of health industry stakeholders could step in to help manage AI risks through self-governance. Using evidence-based risk mitigation practices could be effective across the industry and simultaneously promote and sustain user trust in AI, fending off the next AI winter.