🩺 There are few industries that AI has yet to drastically overhaul, but none are as consequential and delicate as healthcare. If you’ve ever found yourself pleading with an AI agent to get to the bottom of that “strange rash on your left shoulder,” you’re not alone. People are now increasingly turning to consumer-facing AI tools to self-diagnose before ever scheduling a doctor’s appointment.

How AI is being used in healthcare

Today, AI is fundamentally altering how med is practiced, how patients interact with their own health, and how diagnostic decisions are formed by medical professionals. There are exciting new developments that could genuinely improve health outcomes for many, but questions still remain about accuracy, privacy, and how patients and doctors alike will be shaped by the increasing use of AI.

Ever felt like your doctor wasn’t listening to you? Medical institutions and universities worldwide are now deploying what has been described as ambient AI scribes, which are essentially tools designed to listen to patient-doctor conversations in real-time and automatically draft clinical notes. Doctors say that handing over this kind of clerical work to AI allows them to look their patients in the eye rather than typing away at a keyboard, ensuring more accurate note-taking is happening.

But the extent of AI use in healthcare goes beyond minor administrative tasks. The far more groundbreaking way AI is being deployed in the medical field is through diagnostic reasoning. A recent study published in Science has shown a substantial hierarchy in performance when it comes to comparing diagnostic accuracy. Doctors working alone, without the assistance of AI software, score lowest in accuracy, the study claims. Meanwhile, doctors using AI as a “co-pilot” perform significantly better, and — most shockingly — AI operating entirely on its own scored the highest.

While clearly outperforming doctors in diagnostics, AI still lags behind when it comes to clinical reasoning, particularly in the face of limited information, according to a study from Mass General Brigham, Harvard Medical School’s teaching hospital. But what might count as a turning point is patients’ use of AI tools to translate complex medical charts, lab results, and symptom profiles into plain, comprehensible language.

For some doctors, this new dynamic can be a nightmare scenario, especially if it means patients act based purely on the recommendations of a large language model. What this could mean for individuals is that they enter consultations with a slightly more informed baseline understanding of their health status. For patients who live in countries with healthcare systems that are overburdened with months-long waits for medical appointments, or for patients who can’t afford medical care, is getting medical advice from a chatbot better than getting no medical advice at all? It’s certainly a debate amongst medical professionals.

What’s next?

This is just the beginning. The next phase of AI integration into healthcare is the development of technologies that are designed to analyze vocal biomarkers in patients during telehealth consultations. By evaluating subtle, sub-audible variations in pitch, rhythm, and acoustic patterns, AI models are now starting to be used to detect early micro-changes in speech that could serve as passive screening tools, flagging potential neurodegenerative disorders like Parkinson’s disease, or chronic conditions like clinical depression, long before explicit physical symptoms become apparent.

But the USD mn question remains: is AI good for our healthcare systems? Some might argue that AI integration could help reduce physician burnout (think the overworked ER doctors in the drama The Pitt), contribute to early disease detection, and offer more democratic access to medical translation. Yet, navigating the trust factor remains very sensitive. AI is not infallible; it can hallucinate or misinterpret nuances, leading to false reassurances that delay necessary care, or conversely, cause unnecessary panic over benign symptoms.

Because the stakes are life and death, medical AI will likely never be able to operate in a vacuum. The more plausible outcome going forward is that AI will become an increasingly embedded tool within a healthcare practice that combines human oversight with the fast parsing abilities of large language models.