The development of large language models like ChatGPT and, more broadly, generative artificial intelligence (gen AI), have been undeniably remarkable. These AI models, which can create new, synthetic text, images, and video by learning patterns from existing data, have pushed us to fundamentally rethink how we work — or what work even means. Healthcare is no exception. As resident physicians, we are excited by reports of ever-growing gen AI use cases in healthcare systems, from helping physicians respond to patient messages and order diagnostic tests to tracking patient care gaps and designing “smart hospital rooms” that integrate sensors for patient safety monitoring.
However, a predominance of these initiatives is unfolding at a select number of institutions. As we continue to develop and experiment with AI technologies, equitable access to these technologies is crucial to prevent a widening divide in healthcare quality. We must work to ensure that smaller clinics, community hospitals, and underfunded institutions are not left behind.
Drivers of AI Disparity
Innovative uses of AI at top institutions are commendable and exhilarating. But the adoption, or even awareness, of gen AI tools is lacking or absent at most other hospitals that do not have the resources of elite research universities.
At our institution, for example, there was perhaps one didactic session (out of more than 120) during this past academic year that specifically discussed AI, and only then at a broad, superficial level. Opportunities for trainees to learn more about gen AI tools are limited by lack of awareness and accessibility. Our experiences likely represent the overwhelming norm of residents and clinicians nationwide.
This disparity is influenced by several factors. First, there are financial constraints. Most institutions do not have the endowments and funding for the research and development of AI technologies. In a self-perpetuating cycle, underfunded institutions also face challenges in securing funding due to less grant-writing expertise and connections to attract investment.
Then there are the educational gaps. The expertise required to implement gen AI effectively and monitor its use is significant, often requiring the creation of entire teams or offices (Stanford, for example, has a dedicated “Artificial Intelligence Clinical Integration” team). Hospitals that have been able to deploy AI often harness the skills of engineers and researchers drawn from affiliated schools of engineering — affiliations absent at most hospitals. This educational gap extends to medical schools, where curricula often do not adequately cover the latest AI advancements.
Third, at elite research institutions, there is often strong support of the hospital administration in pushing for innovation, manifested through investments in AI programs and strategic partnerships. Smaller hospitals can struggle with administrative inertia, where leadership is either unaware of the potential of AI or understandably reluctant to invest in new technologies over uncertainty about the return on investment. Finally, cultural resistance to AI still exists. Healthcare professionals may be concerned about the possibility of gen AI fundamentally altering their roles. This hesitation is often rooted in a lack of understanding of the ways that AI can complement rather than replace human expertise.
Implications of Generative AI Disparities
Scientific and medical advances have always been pioneered by a few select institutions. So why do disparities in gen AI implementation matter? We believe the exponential pace of AI development and sheer extensiveness of its impact warrants special consideration.
Patient care quality
AI promises to enhance patient care by improving diagnostic accuracy and patient monitoring, creating more personalized treatment plans, and providing real-time medical decision support — all at a faster rate. By automating away documentation and other non-patient-facing tasks, AI also indirectly allows clinicians to spend greater amounts of time helping patients make sense of their care.
If these technologies fulfill their promises, disparities in AI accessibility and capability will lead to variations in patient care. In the long term, populations served at AI-equipped hospitals will likely experience lower rates of chronic diseases, higher life expectancy, and overall better health outcomes.
Economic efficiency
One of the most common applications of AI is improving operational efficiency in healthcare delivery. This includes streamlining administrative tasks, clinical workflows, and resource management. Hospitals can pass on the resulting economic savings to further improve patient care. However, ironically, well-resourced hospitals are best positioned to reap these benefits. Smaller and underfunded facilities are unable to leverage AI for administrative tasks and must continue to rely on labor-intensive and costly processes.
Technological lockouts
The momentum of AI advancement raises the risk of “technological lockouts,” in which institutions that fall behind in AI adoption find it increasingly difficult to catch up.
As well-funded institutions continue to invest in the latest technologies and modify elements in their health system (such as the electronic medical record) to incorporate these tools, they will develop proprietary AI solutions tailored to their specific needs and adapted to new challenges, thus maintaining their competitive edge. Moreover, gen AI thrives on vast amounts of high-quality data. Leading institutions often have access to comprehensive datasets and the infrastructure to manage and analyze this data effectively, while smaller hospitals can’t keep up, hampering their ability to refine their AI models and resulting in significant technological gaps over time.
As certain institutions demonstrate an ability to innovate in gen AI, they will also attract top talent skilled in these tools, creating a concentration of expertise in well-funded hospitals. Healthcare workers in smaller or underfunded hospitals also have limited opportunities for training in AI technologies, perpetuating a skills gap. Finally, well-resourced hospitals are often more adept at influencing and navigating policy and regulatory frameworks, ensuring compliance while advancing their technological capabilities.
Hospitals less equipped to do this will suffer delays in implementing AI solutions. Over time, a fractured healthcare landscape may result, in which a small number of institutions constantly advance technologically while others remain stagnant.
Bridging the Disparities
Preventing widening disparities in AI will require extensive strategies involving government intervention, educational initiatives, collaborative models, and contributions from both the public and private sectors.
Government and policy interventions
Government policies can play a critical role in promoting equitable AI implementation. Policies should focus on providing funding, training grants, and partnership mandates that encourage the adoption of AI in smaller, underfunded, and community hospitals. There are precedents for this; past initiatives have supported electronic health record and health IT adoption. Regulations should ensure that AI technologies address local health challenges and are equitably distributed across different regions.
Education and training programs
To reduce the educational gap, initiatives to enhance gen AI knowledge at all levels of medical education are essential. Professional associations such as the Association of American Medical Colleges have developed resources for this purpose and should continue to offer guidance on the design of medical school curricula and professional development programs. Healthcare systems can collaborate with academic and corporate organizations to create institution-specific AI training modules, as demonstrated by the abundance of existing online courses on gen AI use.
Collaborative models
Creating collaborating models for resource sharing between AI-equipped hospitals and other hospitals has vast potential to reduce disparities. Well-resourced and innovation-focused hospitals should mentor and provide technical support to underfunded or smaller hospitals. Establishing regional AI “hubs” that serve as centers of excellence can facilitate knowledge and resource distribution. Meanwhile, less AI-equipped hospitals ought to proactively consider how gen AI can benefit their workflows.
Similarly, the private sector should be encouraged to invest in affordable AI solutions tailored to the needs of underfunded and community hospitals. By uplifting smaller, underfunded hospitals, the entire system becomes more resilient and capable of handling both public health crises and everyday medical issues alike.
Given how quickly AI tools are evolving, it is not premature to continue developing gen AI in an equitable manner. Neglecting to do so risks creating gaps that will be ever more difficult to bridge. By implementing the strategies outlined above, we can fulfill our ethical imperative to realize more inclusive healthcare systems in which AI technologies benefit all patients, regardless of where they are or the resources available to their healthcare providers.
Henry Bair, MD, MBA, and Mak Djulbegovic, MD, MSc, are resident physicians at Wills Eye Hospital/Jefferson Health in Philadelphia. Bair previously directed several courses on digital health at Stanford University School of Medicine. Djulbegovic is an AI researcher whose work focuses on biomedical applications of large language models.
Please enable JavaScript to view the