Becker’s Healthcare spoke with Peter Bonis, MD, chief medical officer at Wolters Kluwer Health, about the healthcare AI landscape in 2025 and what developments could spur headwinds for emerging technologies.
Editor’s note: Responses have been lightly edited for length and clarity.
Question: What are the most important AI trends that healthcare leaders should monitor in 2025?
Peter Bonis, MD: Competition in AI technology is leading to rapid innovation, so there’s a lot to watch. The capabilities of AI models are improving, and developers are jockeying to establish their position in the evolving marketplace.
Electronic medical record (EMR) vendors have promised to deliver many new capabilities that involve AI. Adjacent technology companies are also launching AI solutions. At the same time, healthcare systems are evaluating how to use their existing technologies and add new ones to improve healthcare services and financial performance. It’s a challenging set of objectives.
Not surprisingly, a lot of the action in the healthcare AI arena is taking place in areas most closely tied to cost savings and revenue capture. We are also seeing enhanced clinician workflow and patient communications starting to gain traction. Within those domains, ambient intelligence is a maturing technology that can aid workflow and improve financial performance. It’s definitely an area worth watching in 2025. It aligns with our strategy to meet the clinicians where they are in the workflow as seen by our collaboration with Abridge and other partners. The other area to focus on is how much progress the EMR vendors will make around AI capabilities.
Q: How are emerging technologies and evolving tech legislation shaping business strategies?
PB: The regulatory landscape is evolving. There’s a tangle of federal, state and international proposals for existing regulations, as well as regulatory guidance. We have the new administration plus the yet untested muscle from the Chevron deference. It’s exceptionally challenging for regulators to keep pace with changing technologies, but guidance continues to emerge.
I’d like to take a step back and discuss U.S. healthcare challenges and where technology can help. We have the most expensive healthcare system in the world, but the performance doesn’t reflect the spending. Consumers are caught in the middle.
Technology and data aren’t going to solve those problems on their own. They can be powerful enablers, but you need to look at the policies, procedures, financial incentives, and stakeholders to really make progress on these deeply entrenched problems.
A great example is the abrasion caused by our payment system. Both payers and providers are using technology to enhance their processes. It’s created a kind of technology arms race. That will change only when strategic and financial incentives become better aligned.
Q: What ongoing concerns exist regarding the responsible use of AI, particularly when it comes to patient data and the use of generative AI in patient interactions?
PB: The published literature on GenAI is increasing and there are a lot of headlines about the utility that GenAI can have. We’ve seen foundational models deliver differential diagnoses, pass board-like exams, do decision support, look at images, summarize healthcare data, help with documentation, assist with revenue optimization, help with patient communications, and more.
When you look more closely at the literature and focus on the clinical domain, the landscape is more concerning, and the dangers must be recognized. Vendors, regulators, healthcare systems, doctors, nurses, and other clinicians need to understand the limitations.
Hallucinations in GenAI use are a well-known problem, but the real concern is the false belief that hallucinations can be identified. Clinicians seem fairly confident that they would recognize when something presented by a large language model isn’t right such as by checking a reference. But that confidence isn’t easily validated. In some cases, even the references may be hallucinations. I’ve seen instances of this first-hand.
Another concern is automaticity. It’s so easy to get seemingly trustworthy answers from a GenAI solution that there is concern about physicians acting without adequately engaging in the decision-making.
Reliability is problematic when providing clinical guidance. Different users should get the same information from GenAI, if they enter similar prompts. However, that’s not the case.
In addition, while AI models give efficient answers, the efficiency may come at the cost of providing sufficient context. For example, I searched on how to treat a urinary tract infection in a patient with a penicillin allergy. One of the commercially available models gave me clear options, but it didn’t respond with important caveats such as whether the patient is pregnant. The drugs it recommended could be harmful in that setting.
Persistent biases still exist in GenAI. Racial biases are well documented, but there are many more subtle biases. Models are subject to all sorts of cognitive biases, just like people. For example, the clinical finding you name first in a search query about a patient with various symptoms may influence the probability of which diagnosis is suggested by the large language model (a bias called the “primacy effect”). There are also biases around the selection of which information resource to use.
At Wolters Kluwer, we are doing everything we can to deliver effective and highly-tested solutions in this high-stakes domain. Healthcare systems understand the risks, in part explaining why they are adopting GenAI solutions in lower-stakes domains like operations and revenue capture.
Q: How can organizations ensure AI is implemented responsibly, and why is this critical for the future of clinical GenAI development?
PB: It’s a real problem. Most healthcare systems don’t have the resources to establish a robust governance process. It’s equally important that vendors and workflow systems ensure that solutions are reliable and effective over time. There must be a governance process among vendors to ensure that solutions perform as expected.
Q: How will AI continue to reshape healthcare workflows and operations?
PB: We’ll see adoption cycles with AI, just as with other technologies. There will be early adopters and then innovations will go more mainstream. The winning technologies will be the ones that work well with the existing workflow technology footprint, particularly EMR systems, and deliver a hard ROI. The large investment into EMR systems and high switching costs suggest they will continue to be at the center of workflow for some time.
Q: As AI-driven revenue optimization evolves, how can healthcare organizations balance implementing innovative technologies while addressing economic challenges and other priorities?
PB: There’s the expression “No margin, no mission,” which means that healthcare systems need sufficient financial performance to deliver on their charter over time. It takes strong leadership to adhere to the mission while creating sustainable financing so the mission can take place. Healthcare leaders must balance investments in technologies that enhance financial performance with tools that provide the best possible work environment so healthcare professionals can deliver exceptional patient care and patients consistently receive affordable, consumer-friendly, effective care wherever they are seen.
At Becker's 4th Annual CEO + CFO Roundtable, taking place November 2–5 in Chicago, more than 1,500 hospital and health system executives tackle decisions that determine whether organizations thrive or merely survive: protecting margins under cost pressure, choosing where to grow, renegotiating payer relationships, stabilizing the workforce and proving real ROI on technology. This is where leaders work through them together, face-to-face. Apply for complimentary registration now.
