Over the past decade, I’ve kept a close eye on the emergence of artificial intelligence in health care. One truth remained constant: Despite all the hype, AI-focused startups and established tech companies alike have failed to move the needle on the nation’s overall health and medical costs.
Finally, after a decade of underperformance in AI-driven medicine, success is approaching faster than physicians and patients currently recognize.
The reason is ChatGPT, the generative AI chatbot from OpenAI that’s taking the digital world by storm. Since its launch in late November, ChatGPT has accomplished impressive feats—passing graduate-level exams for business, law, and medical school (the answers to which can’t simply be Googled).
The next version, ChatGPT4, is scheduled for release later this year, as is Google’s rival AI product. And last week, Microsoft unveiled an AI-powered search engine and web browser in partnership with OpenAI, with other tech-industry competitors slated to join the fray.
It remains to be seen which company will ultimately win the generative-AI arms race. But regardless of who comes out on top, we’ve reached a tipping point.
Generative AI will transform medicine as we know it
In the same way the iPhone became an essential part of our lives in what seemed like no time, ChatGPT (or whatever generative AI tool leads the way) will alter medical practice in previously unimaginable ways.
1. By becoming exponentially faster and more powerful. The human brain can easily predict the rate of arithmetic growth (whereby numbers increase at a constant rate: 1, 2, 3, 4). And it does reasonably well at comprehending geometric growth (a pattern that increases at a constant ratio: 1, 3, 9, 27), as well.
But the implications of continuous, exponential growth prove harder for the human mind to grasp. When it comes to generative AI, that’s the rate of growth to focus on.
Let’s assume that the power and speed of this new technology were to follow Moore’s Law, a posit that computational progress doubles roughly every two years. In that case, ChatGPT will be 32 times more powerful in a decade and over 1,000 times more powerful in two decades.
That’s like trading in your bicycle for a car and then, shortly after, a rocket ship.
So, instead of dwelling on what today’s ChatGPT can (or can’t) do, look ahead a decade. With vastly more computing power and more data and information to draw from, future generations of ChatGPT will possess analytical and problem-solving powers that far exceed current expectations.
This revolution will enable tomorrow’s technology to match the diagnostic skills of clinicians today.
2. By emulating how doctors make clinical decisions. Generative AI isn’t a crystal ball. Like Vegas oddsmakers and Wall Street investors, it cannot definitively predict the winner of the World Series or the next stock-market crash.
Instead, ChatGPT and other generative AI apps can access terabytes of data in less than a second (using hundreds of billions of parameters) to “predict” the next best word or idea in a series of words and concepts. But forming sentences is only the beginning.
Generative AI solves problems, unlike other AI tools. In fact, it closely resembles how doctors solve problems:
- Begin with a large database. For physicians, data comes from classroom lectures, published research and professional experience. For AI, it’s the totality of digitally published material.
- Extract useful information. A physician will recall (or look up) the relevant information that applies to a patient’s symptoms. Generative AI will use billions of parameters to pinpoint appropriate text.
- Use a predictive process to identify the right pieces. Physicians compare possible diagnoses, whereas today’s ChatGPT tests sentences. Both weigh the options and predict among (all of available possibilities) the best match.
Right now, the biggest difference is that doctors can perform an additional step: asking patients a series of clarifying questions and ordering tests to achieve greater accuracy when drawing conclusions. Next generations of generative AI will be able to complete this step (or at least recommend the appropriate laboratory and radiology tests). Already, Microsoft’s new
AI-powered interactive chat feature can ask iterative questions and learn from the conversations.
Just like residents in a hospital, generative AI will initially make mistakes that require a skilled physician to correct. But with greater experience and computing power will come increased acuity and accuracy, as happens with physicians. With time, ChatGPT will make fewer errors until it can match or surpass medical professionals’ predictive powers (and clinical quality).
3. By providing around-the-clock medical assistance. In the United States, 40 percent of Americans suffer two or more chronic illnesses, which, as the name implies, affects their health every day.
What these patients need is continuous daily monitoring and care. Unfortunately for them, the traditional office-based, in-person medical system is not set up to provide it. This is where AI can make a tremendous difference.
Unlike a solo doctor, the next generations of generative AI will be able to monitor patients 24/7 and provide ongoing medical expertise. Doing so would help patients prevent chronic illnesses like heart disease, hypertension, and diabetes and minimize their deadly complications, including heart attacks, strokes, and cancer. This service would cost just pennies a day (ideal at a time when chronic diseases contribute to 90 percent of all health care expenditures).
Generative AI could help patients with chronic disease by:
- Syncing with wearable devices and supportive consumer technologies like Alexa to provide round-the-clock monitoring while giving patients individualized, daily health updates.
- Comparing wearable-device readings against the expected ranges preset by each patient’s doctor—creating patient and physician alerts when something’s wrong.
- Reminding patients at home when they’re due for preventive screenings, Rx refills, or daily exercise (along with other lifestyle improvements).
4. By preventing medical errors. Given OpenAI’s success with Dall-E, an image-based AI platform, along with promising developments in video-based AI from companies like Meta, we expect machine-learning capabilities to evolve far beyond predicting text.
As an example, video-enabled AI in hospitals could help prevent medical errors, a leading cause of death in the United States.
Lapses in patient safety, especially in hospitals, kill tens of thousands of people annually (with some estimates reaching as high as 200,000 deaths). Scientists have defined the steps needed to prevent these unnecessary fatalities. Yet, doctors and nurses often fail to follow evidence-based protocols, leading to avoidable complications.
A recent paper published in the New England Journal of Medicine calculated that nearly 1 in 4 individuals admitted to a hospital will experience harm during their stay. Health care pundits have gone so far as to recommend that hospitalized people bring a family member with them to protect against deadly mistakes made by humans. That won’t be necessary in the future.
Next generations of ChatGPT with video capability will be able to observe doctors and nurses, compare their actions to evidence-based guidelines and warn clinicians when they’re about to commit an error.
This advancement would prevent nearly all medication errors and the majority of hospital-acquired infections, pneumonia, and pressure ulcers.
5. By helping all doctors perform like the best. There is an art and a science to medicine. Medical students and residents learn both skills through a combination of textbooks, journal articles, classroom instruction, and observation of skilled clinicians. Future generations of AI will follow the same approach.
Once ChatGPT is connected to bedside patient monitors and can access laboratory data and listen to physician-patient interactions, the application will begin to predict the optimal set of clinical steps. Each time it compares those decisions against the clinical notes and orders of attending physicians in the electronic health record, ChatGPT will learn and improve.
A matriculating first-year medical student needs ten years of education and training to become fully skilled. Future generations of ChatGPT will complete the process in months or less, learning from the actions of the best clinicians in hundreds of hospitals. And once generative AI becomes sufficiently adept at predicting what experts will do, it can make that expertise available to doctors and nurses anywhere in the country.
What ChatGPT can’t do
No matter how powerful and skilled ChatGPT becomes, it will have limitations. The application will always be dependent on the accuracy of human-inputted data. It will be influenced by the biases of doctors on which the application is trained.
But it will continually improve and address ever-more complex medical problems over time. Whether that requires ten years (and 32 times the computing power) or 20 years (and 1,000 times the power), future generations of generative AI will rival and ultimately exceed today’s physicians’ cognitive and problem-solving abilities.
To prepare the next generation of doctors, today’s educators must break health care’s unwritten rules and build this technology into medical school and residency training. Rather than viewing ChatGPT as a threat, trainees will benefit by learning to harness the clinical powers of generative AI.
Robert Pearl is a plastic surgeon and author of Uncaring: How the Culture of Medicine Kills Doctors and Patients. He can be reached on Twitter @RobertPearlMD.