How will AI impact healthcare? You might live a bit longer, but it will not save us money.
On a recent trip to meet with fund investors one topic repeatedly arose- ‘How will AI affect healthcare?’ The question left me pondering the potential impact and benefits. Personally, I have found that AI helps improve my search results, nicely summarizes my Zoom calls and provides a concise synopsis of a corporate earnings call.
Looking back on historically similar situations, I think of a small personal example. With the advent of personal computers (PCs) in the workplace, my practice was an early adopter of computer technology. Computers made my front office more efficient and provided me with dramatically better information for managing the practice. Digital records improved our ability to track patient needs and we completed more treatment with fewer lost to follow up. However, the advent of computers did not change the actual patient treatment.
Looking at a larger example, radiologists have been using AI tools for decades. Back in 1998 R2 Technologies received FDA approval for its ImageChecker CAD system that helped radiologists detect breast cancer in mammograms. Initially sales were slow, but adoption took off after the software received reimbursement. (Physicians are interested in providing the best care possible. But they are much quicker to adopt innovative technologies if the economics are favorable.) With this new smart tool, breast cancer screening improved and screening saved more lives. However, there was no discernable reduction in the cost of screening or the treatment of discovered breast cancer. One might make the hypothetical argument that treating earlier stage cancers detected with CAD costs less than treating a larger later stage cancer. But overall, I am unaware of any significant cost savings.
Over the ensuing decades, medical imaging companies added thousands of image analysis tools to the radiologist’s armamentarium. Manufacturers used software to differentiate their machines and show that their equipment made the radiologists’ job easier. At the enormous RSNA radiology show in Chicago following Thanksgiving every year, Siemens, GE, and Philips put on awe-inspiring displays of image manipulation in booths the size of mini-Disneylands. The software tools can measure tumor size, segment out tissues such as fat, track the trees of blood vessels, and a myriad of other amazing tasks. On the business side software streamlined radiology office workflows by generating clinical notes and letters to referring physicians.
Yet, despite the software’s power, these tools have not reduced the need for radiologists. In 2025, American diagnostic radiology residency programs grew approximately 4% to a record 1,208 positions, and the field’s vacancy rates are high. In 2025, radiology was one of the better paying medical specialties. In the end, medicine’s ideal target for AI replacement remains unaffected, and radiologists are doing fine. The key point is that looking at the image is only a portion of the radiologist’s role. These clinicians also perform other critical services of putting the image data into context and managing the implications. The actual reading of the X-ray is just a small portion of their added value to the care continuum.
AI is driving tremendous advances in the treatment of rare diseases in children as it can efficiently perform the previously impossible analysis of the 40,000 genes and 3.3 billion base pairs in the human genome. This technology can comb through this enormous pile of data to identify seldom visible aberrant patterns and obscure correlations. With AI assistance families are receiving better diagnoses at earlier stages. And most importantly, the children are receiving better care as early problem identification leads to improved treatment. But is this tool reducing the need for medical geneticists? On the contrary. The profession is in more demand than ever. The health system needs a genetic expert to put the results into context and coordinate with the clinicians that provide care. The number of patients needing their services continues to grow rapidly.
These examples lead me to believe that AI will not reduce costs significantly on the labor side. But let us look at the hospital, which is an important large cog in the healthcare machine and specifically at an emergency visit for an appendectomy. At admission, AI may make the desk work a tad faster. But then, the patient is on a care pathway involving various steps and staff members. AI will not change that process. It may focus the team on delivering the correct care in the most efficient manner and help prevent missteps, but it cannot change the necessary cost of hands-on care.
As discussed earlier, AI will make it possible to improve the quality of care by accelerating the diagnostic process. Already medical residents are using multiple AI tools for this purpose. Having the right diagnosis early eliminates expensive hunting expeditions in which patients cycle through specialists for endless tests and second opinions, generating cost-efficiency savings. Once a patient is in the hospital AI tools will help the nursing staff better supervise patients and provide better care. On the one hand these alerts can prevent expensive interventions when things can go wrong. On the other hand, better care can add to costs. I once invested in a company that remotely provided close oversight of intensive care patients and kept them from failing. It surprised me to learn that hospitals saw the service increase average length of stay, which was bad for the intensive care unit’s profitability. The problem was that patients who were close to mortality were taking longer to die. In this case, better care was not better for the bottom line, and this ironic outcome delayed the company’s sales.
Looking further upstream AI is dramatically expanding our knowledge of human metabolism through extensive analysis of blood and tissue samples. AI can find patterns in the thousands of proteins, enzymes, RNA signals, and genetic signatures to match them with clinical outcomes. Over time, this encyclopedic map to good targets will lead to the development of better drugs that will help patients, but those new medications will not be inexpensive.
The good news is that AI will make better care and health possible. The unwelcome news is that AI will not halt the inexorable