Artificial intelligence (AI) is rapidly entering health care including safety-net, rural, and underserved settings, where access, capacity, and workforce constraints are most acute.
Across the literature, AI is described as a tool that can support clinical decision-making, improve outcomes, and expand access. But those same systems also raise concerns about bias, data limitations, and uneven impacts on underserved populations.
This Digging Deeper series draws from a systematic review of peer-reviewed research to explore both sides of that tension: what AI could make possible, and what risks must be addressed to ensure it does not reinforce existing inequities.
Article Authored by Briana Martin | EHF
Getting AI right takes more than the technology
Our earlier posts focused on what AI can do and where it can go wrong. This final series post identifies what needs to be in place to use it well.
The review makes a broader point beyond just identifying risks: equitable AI depends on conditions that many health systems are still in the process of building.
And those conditions go beyond the technical. The literature points to structural challenges like gaps in regulation, limited governance, and workforce readiness that ultimately shape how AI shows up in practice.
Rather than treating these as a checklist, it’s helpful to think of them as layers of protection. Each layer plays a role in reducing risks that are otherwise predictable.
Layer 1: At the system level: rules and accountability
One of the biggest gaps the literature points to is regulation.
AI is moving quickly, but the frameworks that govern health care weren’t designed with these tools in mind. That makes it harder to answer basic questions like who is responsible when something goes wrong, or how safety should be evaluated.
A central issue here is accountability. In some cases, responsibility can fall on patients or clinicians using the technology, even though they had little to do with how it was designed or developed. At the same time, there are often many actors behind a single AI system, which makes assigning responsibility more complicated.
The literature also notes that regulatory approaches can vary even within a single country. Oversight may differ depending on how an AI tool is categorized, how it is used, or how responsibility is interpreted across different agencies and levels of government.
For example, a tool that supports clinical decision-making may be treated as a regulated medical device in one context but face different requirements in another if it is categorized as administrative support or implemented in a different care setting. That inconsistency means the same type of technology can be held to different standards depending on where and how it is used, adding uncertainty for health systems that are already navigating limited resources.
Layer 2: At the organizational level: how AI is overseen in practice
Even where regulation is still evolving, governance shapes what happens day to day.
The review connects governance challenges to the same issue raised in earlier posts: opacity. When AI tools operate as “black boxes,” it can be difficult for clinicians and patients to fully understand how decisions are made, whether they’re reliable, or how data is being used.
That uncertainty can erode trust and make adoption harder.
The literature suggests that more intentional oversight at the organizational level can help. This might look like dedicated committees or task forces that evaluate AI tools, set expectations for their use, and monitor how they perform over time.
These kinds of structures don’t eliminate risk, but they can make it easier to identify problems early and respond before they scale.
At the same time, the review points to the need for broader coordination across health systems, governments, and sectors to bring more consistency to how AI is governed, especially as these tools become more common.
Layer 3: At the human level: supporting the people using AI
The final layer is about people.
As AI becomes more integrated into care, the literature emphasizes the need for a workforce that understands how to use these tools and when to question them.
That includes building AI-related training into medical education, offering ongoing learning opportunities for current clinicians, and being clearer about how roles may shift in more technology-supported care environments.
There’s also a more practical takeaway here, especially for safety-net, rural, and underserved settings: tools need to be usable.
The review highlights the value of plain-language summaries and more accessible design, so clinicians aren’t left trying to interpret highly-technical outputs in already time-constrained environments.
In other words, trustworthy AI isn’t only about improving the technology itself, it’s about making sure the people using it are supported, informed, and confident in how it works.
Closing the loop: why transparency matters
In its conclusion, the review comes back to a theme that cuts across all of these layers: transparency.
AI has the potential to improve access to information and help reduce gaps in care, especially through tools like telemedicine and mobile technologies that reach remote populations.
But when systems are unclear or biased, they can have the opposite effect by deterring clinicians from using them and reinforcing existing disparities for underserved communities.
Transparency helps bridge that gap. It allows clinicians and policymakers to better understand how decisions are made, evaluate whether tools are working as intended, and make more informed choices about when and how to use them.
ATTRIBUTION & SOURCES
This post draws from Literature Review: AI in Safety-Net Health Care, prepared by the IC² Institute at The University of Texas at Austin.
More from Digging Deeper
How AI Deepens Inequity: Where the Risks Show Up in Practice
Artificial intelligence (AI) is rapidly entering health care including safety-net, rural, and underserved settings, where access, capacity, and workforce constraints are most acute. Across the literature, AI is described as a tool that can support clinical decision-making, improve outcomes, and expand access. But those same systems also raise concerns about
Where AI Could Help: Filling the Gaps that Safety-Net, Rural, and Underserved Systems Face Every Day
Artificial intelligence (AI) is rapidly entering health care including safety-net, rural, and underserved settings, where access, capacity, and workforce constraints are most acute. Across the literature, AI is described as a tool that can support clinical decision-making, improve outcomes, and expand access. But those same systems also raise concerns about
The opportunity to advance diabetes prevention through Texas Medicaid and the Rural Health Transformation Fund
EHF’s Shao-Chee Sim examines how new state and federal policy opportunities could help Texas expand and sustain diabetes prevention.