AI’s first steps in transforming healthcare

Robert Wachter, a physician and author, offered a cautiously optimistic perspective on artificial intelligence in healthcare at a recent conference. He described the technology as a shift from novelty to necessity rather than a passing trend.

AI’s role in healthcare

Wachter based his views on over 100 interviews conducted for his book A Giant Leap: How AI is Transforming Healthcare and What That Means for Our Future. He explained that AI could help address persistent inefficiencies in hospital medicine, where clinicians still face heavy administrative burdens and patients struggle to receive timely care.

The electronic health record, once expected to solve these issues, did not deliver. AI might succeed where earlier systems failed—by automating chart summaries, drafting discharge notes, and aiding clinical decisions. Some of his colleagues already use AI-powered documentation tools, reporting that the technology saves time and allows more direct patient interaction.

Integrating AI-driven clinical decision support into electronic health records could be highly effective, though adoption remains incomplete.

The risks that won’t fade away

Not all concerns about AI carry equal weight. Wachter grouped them into two categories: those that are fading and those that are increasing. Hallucinations—where AI produces false or misleading information—have become less common as the technology improves. Issues like sycophancy and bias in outputs have also declined.

Other risks, however, are growing. Privacy and security remain major concerns, particularly the potential for AI tools to violate HIPAA regulations when processing sensitive patient data. Wachter emphasized the need for caution, noting that even well-intentioned use of clinical information could result in breaches.

His biggest concern was disinformation. To demonstrate the threat, he played a deepfake video of himself delivering statements he never made—a realistic but entirely fabricated conversation with clinicians in a hospital setting. The demonstration showed how easily AI could be used to spread false medical advice or manipulate public opinion.

If the technology can replicate a physician’s voice and mannerisms so convincingly, distinguishing truth from fabrication becomes difficult for patients and clinicians alike.

Speed over perfection

Wachter advised healthcare leaders to act quickly while implementing safeguards. He recommended establishing adaptable governance structures, including strong monitoring and oversight. Training staff to recognize skill gaps and use AI effectively should be a priority.

He reinforced this with a quote from Dr. Gianrico Farrugia: “The risk of not going fast enough is far greater than the risk of going too fast.” Hesitation could lead to falling behind, while recklessness could cause serious problems.

Wachter also discussed the fear of deskilling, where over-reliance on AI might weaken clinicians’ abilities. He shared a personal example: if he lost his phone, he might struggle to contact his wife because he no longer remembers her number. The same risk applies to medicine—if AI handles too much of the cognitive work, future doctors might lose their diagnostic skills.

Medical education must adapt. Curricula should teach students how to collaborate with AI, focusing on diagnostic reasoning and maintaining human oversight. The goal is not to resist AI but to ensure clinicians remain the final decision-makers in patient care.

This balance is especially important in fields where AI excels at pattern recognition and data processing. Wachter identified radiology, pathology, and dermatology as the most vulnerable to disruption, while surgical specialties faced the least risk. Hospital medicine, he noted, requires a mix of clinical knowledge, emotional intelligence, and care coordination that AI cannot fully replace.

The limits of accuracy

AI will never reach perfect accuracy, Wachter stated. Humans will always play a role, particularly in complex, high-stakes decisions. He illustrated this with a story from his book about a rounding team caring for a nonverbal patient. The interaction highlighted essential aspects of medicine—empathy, communication, and interpreting subtle cues—that AI, no matter how advanced, cannot provide.

“Patients will still need a doctor,” he said, reading an excerpt that positioned AI as a support tool rather than a substitute. The boundary between assistance and autonomy remains fluid, but Wachter’s optimism stems from the belief that AI can manage routine tasks, freeing clinicians to focus on what truly matters.

The technology is moving from curiosity to standard practice. The challenge for leaders is to adopt it without losing sight of its limitations or the people it is designed to help.