Billions of dollars have been invested in healthcare AI. But are we spending in the right places?

Much has been made about the promise of AI in healthcare. Venture-capital investments in health AI in the US alone will reach $11 billion this year, with additional funding from institutional investors and other organizations across the hulking $4.5 trillion US healthcare market.

The rush to invest and implement generative AI for all kinds of applications is coupled with assurances that this technology is the silver bullet we’ve been waiting for to disrupt our broken system. Simultaneously, critics counter that we are ill-prepared and lack the capacity and capability to scale this technology in an industry where lives are at risk and health equity is persistently beyond our reach.

The reality is that it’s too early to take a position on whether generative AI in healthcare will help, harm or simply squander billions of dollars with no improvement in people’s lives. We can, however, be sure that our decisions about resource allocation in the immediate future will determine the outcome of this latest healthcare innovation.

Currently, the adoption of generative AI solutions is largely limited to leading academic medical centres, which is typical of the diffusion of innovation in healthcare. Let’s consider the technological advances that drove the advancement of laparoscopic (keyhole) surgeries.

These new minimally invasive techniques that rely on a camera through a small incision, instead of open surgeries, were pioneered by surgeons at academic institutions before becoming widely adopted across the country. In this way, diligent research eventually resulted in better patient experience, dramatically better outcomes and substantial cost savings.

This diffusion of innovation and outsized ROI did not happen organically; enormous investments were poured into resourcing the minimally invasive surgery revolution. From refitting surgical suites for these new technologies and reimagining clinical workflows to substantial retraining of the full clinical team, success at scale did not happen by accident.

This approach to putting real investment in studying the clinical workflows that generative AI demands, as well as training and supporting the medical workforce to safely and effectively implement these tools, must be table stakes as we plan for the AI-enabled future of healthcare.

The broader adoption of laparoscopic surgery required new knowledge, skills and expertise, but not a new care team. Whether laparoscopic or open, successful surgery relies on the surgeon’s knowledge of physiology, anatomy, disease mechanisms and surgical techniques.

However, identifying and managing the risks of using AI in healthcare – such as limited understanding of how AI makes decisions (“black box reasoning”), keeping the AI's performance consistent over time (“AI drift”), and avoiding over-reliance on AI recommendations (“automation bias”) – along with ensuring privacy and security, requires a fundamentally different skill mix in the care team, combining clinical and technical expertise that extends far beyond the capacity of the traditional clinical workforce.

It is not sufficient to simply add requirements for new skills to the existing clinical workforce. Instead, we must begin investing in a multidisciplinary healthcare workforce that combines traditional clinical expertise with technical competencies, and supplement both with new caregiving and community-based roles as we prepare for AI to fundamentally change how we deliver care in the digital era.

In the AI era of healthcare, we don’t simply want the equivalent of a new way of performing surgery; we must strive to prevent as many people as we can from needing it in the first place.

Moving beyond this analogy, we don’t want to simply use AI to better triage patients in the emergency room; we must travel upstream to prevent people from needing emergent care in the first place.

We don’t simply want to make it easier for people to navigate today’s complexities of a late-stage cancer diagnosis. We must ensure they receive the proper screening based on their personal risk, followed by rapid diagnosis and optimized treatment that is easily accessible and affordable, regardless of their current access to care.

This vision requires a fundamental shift in our current focus and resource allocation strategy related to health AI. We must go beyond “the model” and immediately begin being more intentional about the problems we are trying to solve with this new generation of tools.

The billions of dollars already and yet to be invested will not return commercially or in clinical outcomes unless we prioritize the needs of patients, building and using AI solutions that drive a fundamentally different approach to healthcare.

To achieve this, we must invest in turning scientific discoveries into practical, real-world healthcare solutions. This includes evaluating and improving workflows, preparing the workforce through training and planning, and building the necessary infrastructure. If we don’t match these efforts with our current investments in developing new models, there will not be a market for these new solutions.

These investments must be globally distributed across all care settings, or generative AI will not reach the health systems and patients who most need – and who will benefit most – from these innovations. Nor will it reduce the overall cost of care that we all bear through our insurance contributions and taxes.

This investment in preparedness – far downstream of where resources are currently being pumped – cannot wait. The billions of dollars poured into health AI will be wasted, and the opportunity to move from a broken, sick care system to an equitable, affordable healthcare system will be squandered if we don’t plan for successful implementation at scale.