The new frontier: Pros and cons of AI in medicine

Let’s talk about the elephant in the room: artificial intelligence (AI). With a market projected to leap from $150 billion to $1.3 trillion by 2032, AI is becoming ubiquitous, weaving itself seamlessly into the fabric of our daily lives. Currently, many of us underestimate just how much we rely on it; studies by Tableau show that up to 77% of our devices and platforms already use AI to optimize algorithms and user experiences.

Whether we’ve fully grasped the gravity of this shift or not, AI is here to stay. It is a milestone no less groundbreaking than the advent of the home computer, the internet or even broadcast television. As a digital native, I believe we must embrace this challenge. Regardless of where you stand on the issue, one thing is certain: You cannot effectively adopt—or regulate—something you haven’t tried to understand.

Clinical shifts vs. technological shifts

Medicine is a state of constant metamorphosis. If you practice long enough, you’ll see clinical habits fall in and out of favor as evidence evolves. As physician-scientists and patient advocates, we are usually at the forefront of these changes, pivoting our workflows to meet the latest guidelines.

Interestingly, we often accept these clinical shifts willingly because they feel familiar. We understand the science behind them. However, when it comes to technology, billing or insurance, many of us feel like we’re in uncharted territory. That apprehension is exactly why we should push ourselves to use these tools. If we don’t participate in their integration, we lose our seat at the table when it comes time for their regulation.

History repeats itself

The last time medicine faced a systemic disruption of this magnitude was the transition to the electronic medical record (EMR). There are many physicians still practicing today who trained entirely on paper charts; that was the norm until the internet and home computers became household fixtures in the 1990s.

Following the 1996 Health Insurance Portability and Accountability Act (HIPAA) and the 2009 American Recovery and Reinvestment Act (ARRA), the digital shift became very difficult to avoid. At the time, the change was met with significant resistance, with many highlighting workflow disturbances and security headaches.

Yet, those same years also brought the benefits of centralized communication, better continuity of care and consistent record-keeping. Today, for better or for worse, EMRs are integrated into daily practice. I believe AI in medicine is just as disruptive—and will become just as mainstay—as the EMR was 30 years ago.

Pros and cons of AI

AI allows us to offload clerical “scut work” like note-taking and billing, freeing us to stay current on medical literature. For those of us in training, it is a powerful tool for organizing differentials and streamlining research. Some useful tools include:

• Evidence-based retrieval: Platforms like Open Evidence, Doximity and UpToDate allow users to search for information methodically. Unlike a standard Google search, these tools pull from peer-reviewed sources, ensuring that the information improving our patient care is both quick and accurate.
• Ambient scribe: Tools like Doximity’s ambient scribe feature allow us to capture notes without the fear of forgetting critical details. While these must be reviewed for errors, removing the “screen barrier” between the doctor and the patient goes a long way in building rapport.
• Synthesizing the chart: EMR-based AI, such as features found in Epic, helps us synthesize a patient’s history efficiently, reducing the risk of human error during chart-checking.
• Learning and research: AI tools like NotebookLM can turn a 400-page clinical guideline into a succinct summary or even a riveting podcast for your commute. Furthermore, AI has lowered the barrier to entry for research. Even those not well-versed in complex research processes can use AI to teach themselves the workflow, allowing productive projects to take flight that might otherwise have stalled.

This is not to say that AI is without its pitfalls. Just as we faced security and pricing concerns with EMRs, AI carries its own baggage. There are environmental implications to the massive cloud computing and Large Language Model (LLM) power required.

Most uniquely, we have the issue of “hallucinations.” This is a widely recognized phenomenon where AI generates plausible, but entirely false information—it’s also possible for AI to inaccurately summarize the requested information. AI-provided or summarized information often needs to be thoroughly reviewed.

These pitfalls are less likely when we use healthcare-specific LLMs like OpenEvidence, Doximity or UpToDate, among others. Additionally, giving limits, sources and context to queries can also help physicians avoid these hazards.

Beyond the technical, we face nuanced ethical issues like algorithmic biases and the issue of liability. AI is famously training on homogenous data that is representative of only a subset of the population. AI also draws from numerous sources, much of which have not been vetted for accuracy or veracity.

Blindly using AI without being aware of this can perpetuate racial and socioeconomic biases. Recognizing that these biases exist (much like being aware of biases in research) is the way to offset these issues. Additionally, running the same query in different LLMs and asking for sources can be a way to verify authenticity.

Also, if mistakes occur as a result of AI in medicine, who is to blame? The physician? The hospital? Or will it be a large overseas tech company? It’s hard to say, so it’s imperative that we double-check things and remain cautious.

Using AI to benefit all

I’ve touched on some real problems, but they can only be addressed through constant use and advocacy by physicians. As more corporations move toward AI, we cannot afford to be passive observers. If we don’t jump on this now and lead the way in active participation and regulation, we will be stuck playing by a set of rules that someone else envisioned.

As organizations like the AOiA and AOA take steps to foster AI literacy in physicians and trainees, we should all be seeking to master the technology within a culture of “informed skepticism.” We should use AI to offload tasks that make us feel like machines, so we can focus on the work that makes physicians.

Related reading:

Exploring how physicians can make themselves ‘AI-proof’

Essential apps for healthcare professionals: Streamlining care, education and wellness