As two people who have spent careers at the intersection of healthcare delivery and innovation, we have both watched the AI revolution unfold with both excitement and concern. While artificial intelligence promises to transform healthcare, however, we must separate reality from hype to harness its true potential.

Let’s be clear: AI in healthcare didn’t start with ChatGPT. For decades, healthcare organizations have been using machine learning and pattern recognition to assist medical professionals. In the mid-1990s, we saw AI applications in pathology for cancer detection. These weren’t replacements for physicians but rather tools to enhance their capabilities.

This fundamental philosophy remains crucial today: AI should augment, not replace, healthcare professionals.

The current wave of generative AI has captured public imagination, but we must temper our expectations. While these models demonstrate impressive capabilities in processing vast amounts of medical literature and generating coherent text, they’re far from perfect. 

Indeed, current large language models have error rates between 15% and 40% — which is far from acceptable when lives are at stake.

What patient would trust a physician who offered the right information 70% of the time?

When summarizing clinical trials, analyzing patient-physician conversations, or making treatment recommendations, we cannot afford hallucinations or inconsistencies.

But this doesn’t mean we should dismiss AI’s potential; rather, we must be strategic as an industry about where and how we deploy it. What Roth and others have pointed out is that generative AI models may not ever be great at medical reasoning, but they can indeed be good at identifying key factors that should go into decision-making.

As a result, the real promise of AI in healthcare lies in its ability to support human decision-making, not replace it. These tools can digest massive amounts of medical literature, analyze patient records, and identify patterns that might escape even the most experienced clinicians.

The key is building reliable decision-support systems on top of these foundation models; systems that combine AI’s pattern-recognition capabilities with rigorous clinical validation and human oversight.

Across many organizations in healthcare, including those with which we have firsthand experience, there have been explorations in targeted applications where AI can enhance specific healthcare processes. We’ve learned that while large language models offer impressive general capabilities, specialized models focused on specific tasks often provide better results, and they can do it faster than general purpose models.

Critically, such targeted applications can also yield such results at a fraction of the cost, when analyzing electronic health records for specific conditions or medications.

This practical approach to AI adoption reflects a broader truth: healthcare innovation isn’t about chasing the latest technology trend – it’s about finding sustainable solutions to real problems. As healthcare leaders, we must focus on applications that demonstrably improve patient outcomes, support our healthcare workers, protect patient privacy, and maintain cost-effectiveness.

The path forward requires a balanced approach that takes all of the above factors into careful consideration. We should leverage AI’s capabilities to handle routine tasks, analyze complex medical data, and support clinical decision-making, all while maintaining high standards for accuracy and reliability. Success in healthcare AI will depend on our ability to match the right technology to the right challenge.

But we need to resist the temptation to view AI as a magical solution to all healthcare problems, and to lavish our attention and resources on one-size-fits-all technologies, or those that support autonomous AI decision-making. 

The future of healthcare isn’t about AI taking over; it’s about creating a symbiotic relationship between human expertise and artificial intelligence. By maintaining this perspective, we can harness AI’s potential while avoiding its pitfalls, ultimately creating a healthcare system that better serves both patients and providers.

But to get there we must ignore the hype, and we must not let ourselves be fooled by fluency. As any physician could tell you, there’s a universe of difference between a brilliant med student and a competent clinician. We’ll do well by patients and clinicians to bear this difference in mind as we harness AI’s power to benefit all.

Photo: metamorworks, Getty Images

This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.