Physicians and health systems must align on AI to avoid repeating EHR mistakes

Artificial intelligence will change health care. But how will physicians and health care organizations align to deliver on AI's potential in patient care? That potential includes reduced physician burnout, lower costs and higher quality, efficiencies that improve patient access, greater patient satisfaction and less fragmentation of services across the health care system. Alignment means physicians and health care organizations seeing similar value and benefits in AI, agreeing on when and how AI should be implemented in patient care, and working collaboratively to improve health care delivery outcomes with AI.

There is a long way to go. Here's why AI alignment may be difficult in the short run, and what physicians and health care organizations can do to achieve effective AI implementation.

The physician perspective on AI and why it matters

When electronic health records (EHRs) were implemented several decades ago, health care organizations made critical mistakes: They did not understand how physicians would want to use EHRs in their work, did not consider those views enough in EHR design and implementation, and did not fully appreciate how physicians perceived the value of EHRs for themselves, patients and their employers. This hurt both early and later EHR implementation, reducing the tool's overall effectiveness and delaying widespread adoption by physicians. To this day, many health care organizations are still working with physicians to optimize EHR usage.

The same mistakes could be made with AI. Physicians have strong opinions about the use of AI in their work, even as their daily use of AI grows. Most important, physicians are more willing to use AI when it makes their everyday lives easier and gives back time lost to administrative and documentation demands. In other words, physicians want AI first and foremost to help them on a personal level. Health care organizations need to acknowledge this.

Other physician opinions about AI are also important to consider. A recent national survey of 1,000 physicians revealed high levels of dissatisfaction with how health care organizations are thinking about AI and how quickly they are moving to implement it in patient care, with many physicians feeling organizations are not moving quickly enough. At the same time, physicians appear to trust health care organizations less to consult with them on which AI tools should be used in patient care and how. In the same survey, just under three-quarters of physicians said they did not have much influence over which AI tools are being implemented, even as two-thirds wanted more say in those decisions. Other surveys show physicians need more assurance that AI will be used appropriately in patient care.

Why are these opinions important to appreciate now, in the early phase of AI implementation in health care? Because if they are not addressed, they form the basis for later conflict between health care organizations and physicians. They also reduce the potential for alignment on the goals and implementation of AI in patient care, meaning AI use may fall short of expectations. Here are three ways to address physicians' early opinions and achieve greater alignment now.

1. Seek early wins that demonstrate why alignment on AI is beneficial

The most widespread early adoption of AI in everyday patient care involves ambient dictation to help physicians with documentation and medical record completion. Physicians are adopting ambient dictation rapidly and in large numbers. Research supports using generative AI in medical dictation to reduce physician burnout and documentation burden and to save physicians time.

Organizations already achieving early success include large health systems such as Cleveland Clinic and Kaiser Permanente. Both have touted successful ambient dictation implementation and mutually beneficial outcomes. Early wins that demonstrate the benefits of alignment get physicians and health care organizations to agree on what AI can achieve in patient care. They also build trust between the two that AI can help physicians. Getting physician support early is crucial, as the history of EHRs shows.

2. Collaborate now on the clinical decision support question

Ambient dictation is less controversial than what is coming next to patient care: the use of AI to assist physicians with diagnosis and treatment. Physicians report already using clinical decision support AI tools such as OpenEvidence in their work, often unofficially and under their own control. But many also want their employers to sanction these tools, invest in them and help them integrate them into patient care now.

Organizations are hesitant, citing the resource investment needed and unclear return on that investment; liability and privacy concerns; the lack of a clear company strategy for integrating AI; and concern about the current state of AI technology. At the same time, AI for clinical decision-making both excites and scares physicians, given its potential to reduce physician autonomy and discretion.

Effective collaboration between physicians and health care organizations on AI for clinical decision support requires a commitment that any AI used in clinical decision-making remains under physician control. Health care organizations must acknowledge the threat of AI to replace rather than complement what physicians do. Giving physicians real input into which AI tools should be adopted for which areas of patient care helps ease these fears. Also important is redesigning clinical workflows so that AI clinical decision support tools let physicians decide in real time whether to use them. This preserves physician autonomy and discretion over the tool.

3. Adopt an incremental, data-driven experimentation mindset

Showing proof of concept with AI in patient care is critical for physician acceptance, mutual trust and physician-organization alignment. Both physicians and organizations must adopt an experimentation mindset when considering if, when and how AI should be used. That means pilot testing AI tools in incremental doses that allow close study of quality and efficiency outcomes, examination of how AI functions in everyday clinical workflows, and collection of real-time data on clinician and patient satisfaction. This phased, process-improvement approach is exactly what Cleveland Clinic and Kaiser Permanente followed when rolling out ambient dictation to their physicians.

Implementing AI incrementally by identifying physicians who can serve as early adopters is also crucial to early alignment. Both Cleveland Clinic and Kaiser Permanente let physicians volunteer to use the ambient scribing tool during the first phase of implementation. That drew in the physicians who either felt more comfortable with the technology or had the greatest need for it because of their existing note-taking burden.

Diffusion of innovations theory explains why starting with a smaller group of early adopters enhances the long-term success of any innovation. First, peers are likely to view early adopters favorably as opinion leaders, so their embrace of an AI tool influences others to adopt it. Second, they serve as test cases, lowering the risk for later adopters and demonstrating feasibility. Finally, early adopters are willing to make mistakes, adapt how they use the tool and, in the process, fine-tune it so later adopters have fewer negative experiences.

AI implementation in health care will be more successful if alignment remains a key goal for both physicians and health care organizations. In these early phases of AI use, it is imperative to jointly pursue collaboration, experimentation and evaluation. The trust that results provides a deep pool of goodwill that can be drawn on to navigate what looms as a decades-long transformation — perhaps the most important one in the history of the U.S. health care system.

Timothy Hoff is a professor of management, health care systems and health policy at Northeastern University in Boston and author of the book "Searching for the Family Doctor: Primary Care on the Brink."