In a Nutshell:

• Health care providers are beginning to use artificial intelligence (AI) tools in their practice, largely without federal or state regulation.
• AI has the potential to improve efficiency, drive down costs, and improve patient outcomes, but the technology presents significant safety and privacy risks.
• The role of government in the oversight and regulation of AI in health care should be considered by the state, and the state should not wait to engage in those deliberations.

Artificial intelligence (AI) is “technology that enables computers and machines to simulate human intelligence and problem-solving capabilities.” While AI has been around for decades, a recent wave of developments in generative AI – AI that can generate text, images, video, and other kinds of data – has brought the utilization of AI into the mainstream, particularly following the release of user-friendly generative AI programs (e.g., ChatGPT). Governments, businesses, and individuals are utilizing these modern AI tools in countless ways, with a major focus on searching for information and preparing written responses (e.g., chatbots, essays, etc.).

Mainstream AI utilization is touching every aspect of life, but health care is an area where the stakes are particularly high. Many of AI’s current applications across all fields involve substituting technology for human labor in the name of efficiency. In health care, AI has the potential to create efficiencies that save time and drive down costs. AI utilization may also improve patient outcomes through enhanced research, diagnosis, and treatment. It is important to balance the risks that come with using AI with the implications of this emerging technology to ensure its utilization serves the public interest.

AI Utilization in Health Care

Many potential avenues for the use of AI in health care exist and are being evaluated. The ability of AI to scan large swaths of information and search for patterns has significant potential for medical diagnosis, treatment, and self-care, as well as aiding health professionals’ knowledge and improving clinical trials and studies. For instance, providers or patients could enter symptoms into an AI platform and receive information about the most likely diagnoses or the preferred tests to administer. Similarly, AI platforms could answer patient questions about procedures or conditions. AI could also be used as a force multiplier in areas with a lack of access to certain specialties.

The benefits of AI utilization in health care are straightforward. Health care is expensive and labor intensive, so replacing human time with automation has value in terms of cost and in making up for provider shortages. Beyond efficiency, AI has the potential to improve diagnosis and treatment by aiding (or even outperforming) the average clinician at a variety of tasks. Even in cases where highly trained providers perform better than existing AI in complex health care settings, AI can be used to supplement provider expertise, getting the best of both worlds at very little additional cost. 

As the technology develops, it is likely that the best AI tools will meet or exceed the diagnostic accuracy of many providers, which if used properly, would improve average diagnostic accuracy overall. Similarly, making medical advice and expertise more accessible via chatbots could improve health outcomes for people who cannot afford or access medical care. Many entities are working diligently to capitalize on this potential. For instance, Michigan Medicine is working to implement AI tools in administrative and clinical contexts while also setting policies around patient privacy and safety. 

While AI has the potential to be extremely valuable, two major categories of risk come with its utilization in health care. The first centers around data privacy, as utilizing AI in many of the proposed contexts would require giving personal health information to some form of AI software that inherently learns from every bit of information it receives. Existing laws and regulations – primarily those that contemplate medical records being housed in siloed records management systems – may not be sufficient to ensure that confidential medical information shared with an AI software is not inadvertently disclosed to those without authorization.

The second category of risk is accuracy. While human decision-making is also imperfect, many examples exist of AI tools presenting wrong, nonsensical, and/or dangerous results. Many different AI tools already exist with varying levels of performance on different tasks. Some AI tools may far exceed the average clinician’s diagnostic accuracy, while others may perform worse, even if they avoid blatantly flawed results. AI functionality depends both on the operation of the code and on the data it learns from, meaning any particular AI platform may do better or worse for a variety of different reasons. 

It is important that providers do not have incentives to utilize AI in a manner where doing so could lead to less accurate results than relying only on human expertise. Similarly, providers will likely have a choice of AI tools and should be incentivized to use the tools that maximize accuracy rather than speed, cost savings, or other variables not directly related to patient outcomes. Related to broader concerns about accuracy, AI has the potential to exacerbate discrimination if the technology was trained on data with underlying biases. 

Questions about health care AI accuracy need to be analyzed and debated as a matter of public policy. Given the potential benefits AI could bring to health care, governments should prepare for increased AI utilization with an eye on improving efficiency and health outcomes, while also protecting providers and patients from potential pitfalls.

Government Response to Technological Advances

Governments are often slow to respond to new technologies, both in terms of utilization and regulation. Laws and regulations do not always contain the flexibility needed to address novel technologies and it takes time to enact and implement necessary changes. Rather than waiting for the AI landscape to unfold, policymakers should engage early with the implications of AI on modern life, both in terms of promoting the value of AI and in establishing appropriate safeguards.

Governments, including the State of Michigan, are starting to think about how they might deploy AI to carry out its functions. As part of the State of Michigan’s FY2025 budget, $10 million was appropriated to explore “strategy, platforms, and tools for the integration of artificial intelligence and develop[ment of] pilot projects that capitalize on the potential of this new generative technology to transform the provision of government service.” While this is a worthwhile step, it only covers one part of the government’s role in the emerging AI landscape.

The government is unlikely to be a leader in the development and deployment of AI tools. Improvement in the technology’s capabilities and implementation will flow from the private sector and research community, and it is important that the government avoid getting in the way of advancements in AI.

The goal for the government should be to support an environment in which AI development and utilization can flourish, while also ensuring proper guardrails are in place to protect people from intentional misuse and unintended negative consequences. The government does have a role in regulating the utilization of AI in all walks of life. The extent of those regulations are very much up for debate, but in health care alone, AI has implications for the workforce, individual privacy, patient safety, and costs, not to mention research and development. Policymakers should grapple with the implications of AI before it is fully enmeshed within the industry and consider whether any laws or regulations need to be put in place.

Existing Policies and Proposals on AI in Health Care

The emergence of mainstream generative AI platforms over the last few years has moved the discussion of AI utilization and regulation into the public consciousness, but policy action on AI in health care has been limited thus far. Federal agencies have engaged with the implications of AI for health care through executive actions such as establishing task forces and issuing guidance, but there have not been major legislative or rulemaking actions. In fact, proposing regulations at all has been challenging because the government usually regulates specific drugs, devices, and procedures. AI cannot necessarily fit into the existing authorization framework because the software changes over time. 

While the European Union has responded to these challenges by enacting a new regulatory framework specifically for AI, the U.S. Congress has not embraced policy discussions related to health care and AI in earnest. Some legislation has  been introduced related to relatively minor aspects, like allowing AI to prescribe medication if it is authorized at the state level, setting guidance for Medicare payments for the use of specific AI monitoring devices, and authorizing the use of AI in certain government-funded health research. 

Broadly, many have taken note of Congress’ lack of action on AI generally and the role states are playing in filling the void. Some recent examples include an Illinois bill that would require AI programs used to diagnose patients to be certified and shown to be accurate. Legislation in Georgia would prohibit making insurance coverage or health care decisions solely with AI. A proposal in Massachusetts would regulate the use of AI in mental health treatment, including requiring the provider to seek approval from their licensing board, use an AI system that prioritizes patient safety, and inform patients of their use of AI and receive their consent. Similar legislation was proposed in Texas and Rhode Island. States have also pursued and enacted broader legislation related to AI, with some of that legislation including data collection and opt-out requirements that could have some impact on health care even if health data is not the specific focus of the laws. 

While a case can clearly be made for the federal government to take the lead on AI, both generally and as it relates to health care, the lack of action in Washington, D.C. has prompted states to engage in their own policymaking process.

Michigan’s Next Steps

The novelty of mainstream AI technology leaves policymakers with no easy answers about when and how to regulate its use in health care. However, there are a variety of key questions for policymakers to consider when deciding what kinds of laws and rules might be necessary. 

Who, What, When

The core regulatory questions for policymakers to consider are which providers can use AI tools, what AI tools they can use, and when they can use them for patient care. Policy around notifying and receiving consent from patients to utilize AI tools as part of their care is also a key aspect of patient choice. 

The state should consider whether it should set criteria around which AI tools meet accuracy and safety thresholds. Within the context of regulating which tools meet the necessary criteria, the state should consider whether it is necessary to set transparency standards about the operation of an AI tool and its training data. 

Additionally, while AI tools are supposed to be easy to use, using them well is a skill and the state should consider whether it needs to set additional licensing requirements on providers who want to utilize AI in their practice. The state should also consider whether it is appropriate to set guidelines on intended use. While it is challenging to envision every scenario in which AI could or will be used, the state could give providers direction on the manner in which AI should fit into the scope of practice.

Other Considerations

Beyond the core questions of using AI for patient care, the state should also engage with a variety of other policy-related questions, including:

• Data Privacy: Whether in the context of broader data privacy laws or in a specific health care context, the state should assure patients their data is safe from disclosure even if AI is involved in their treatment.
• Discrimination: The state must ensure it can police health care discrimination when AI is involved, but that may look different than the existing framework.
• Cost: Health care providers will certainly bill for services provided by AI, but patients should expect to share in any efficiencies, either through better or cheaper care.
• Liability: Existing liability law may need to be adjusted to clearly delineate who is at fault if health care that was delivered or augmented by AI leads to some sort of adverse outcome, or if a patient suffers when a provider ignores AI.
• Labor Protections: Laws may need to specify whether individual clinicians can be subjected to employment sanctions if they decline to use and/or overrule recommendations made by AI tools.

The specific provisions of any of these policies, and the extent to which these issues can be addressed under existing law, are very much open for debate. The state has a variety of interests to balance and lacks concrete data to inform best practices, but it has an interest in making sure that existing protections within health care can be applied as AI becomes more commonplace. 

AI is coming to health care and the state should be thinking about answering all of these questions now. That is not to say that the state should pursue an overly strict regulatory framework at the outset, but that it should be actively considering the merits of each of these questions. It is easy to fall into the trap of waiting to see how things shake out before taking action, but waiting too long could have significant consequences in terms of patient privacy and safety.

Conclusion

Mainstream AI utilization is likely to transform health care over the next decade and the stakes of properly managing that transition are high. More efficient, cheaper, and more accurate health care is a noble goal that AI can help achieve, but there are significant risks to allowing the unregulated use of AI in health care.

With the federal government largely absent from the arena, states are beginning to pursue AI regulations. While the implications for health care are huge, very few laws and rules on AI in health care have been enacted. Rather than following the usual pattern of waiting for the technology to mature and establish itself, policymakers in Michigan should actively engage with a series of key questions surrounding the use of AI in health care to ensure that the state’s residents get the most out of this technology without opening the door to major downside risks.