OpenAI, the artificial intelligence startup based in San Francisco, generated massive media interest last month when it announced that its signature product, ChatGPT, is gaining memory.
According to the company, “We’re testing the ability for ChatGPT to remember things you discuss,” which means the generative AI system will, “carry what it learns between chats, allowing it to provide more relevant responses.”
While the tech community buzzes with both excitement and apprehension, initial media coverage has focused on (a) charming examples, like AI remembering your child’s fondness for jellyfish, or (b) familiar user concerns about bias and privacy.
So far, the national conversation is overlooking the biggest opportunity at hand. OpenAI’s company blog hints at it: “We’re taking steps to assess and mitigate biases, and steer ChatGPT away from proactively remembering sensitive information, like your health details – unless you explicitly ask it to.”
Although there are many technological hurdles to clear and fears to mitigate, memory-powered AI will be a pivotal step toward transforming U.S. medicine.
It holds the potential for making healthcare more personalized, patient-centric and affordable. These improvements—alongside the potential pitfalls of AI-empowered healthcare—are explored in depth in my upcoming book “ChatGPT, MD: How AI-Empowered Patients & Doctors Can Take Back Control of American Medicine.”
Even with the memory upgrade, ChatGPT’s “context window” (how many words it can recall before losing its memory), falls short of the nearly 17,000 words found in the average patient’s medical record. However, predictions are that ChatGPT will become 30 times more powerful within the next five years, dramatically expanding its data retention capabilities and enhancing its reliability.
Looking ahead, generative AI’s expanded memory capabilities will improve clinical outcomes and revolutionize U.S. medicine. Here are three potential breakthroughs:
1. More Accurate Diagnoses
For over a decade, clinicians have aimed to deliver precise, personalized care that’s tailored to each patient’s unique health profile, including their genetic makeup and personal health preferences. Their efforts, however, have faced significant obstacles.
One major challenge is the sheer volume of knowledge required to customize medical care. For example, the human genome consists of approximately 3 billion base pairs of DNA, which if typed out as letters (A, C, G, T) would fill about 200 New York City phone books.
Another is that medical knowledge doubles every 73 days, making it almost impossible for physicians to keep up with innovative medical findings and newer guidelines for helping patients.
A third hurdle is accessing a comprehensive medical record. With the average patient consulting 19 different doctors throughout their lifetime, their medical records are often dispersed across numerous medical offices and health systems. The lack of interoperability among electronic health record (EHR) systems compounds this issue, preventing clinicians—and by extension, generative AI—from accessing a patient’s complete medical history.
The introduction of specialized plug-ins (known as GPTs) combined with AI memory offers a promising workaround. Initially, generative AI might access a limited set of patient data through platforms like My Chart, which can be downloaded through home computers or smartphones. Eventually, however, generative AI apps (or plug-ins or individual GPTs) will enable patients to consolidate their digital medical records from various healthcare providers. This kind of comprehensive, personalized EHR will serve as a reliable resource for both patients and their healthcare teams.
With this information stored in an AI’s memory, patients will be able to input their symptoms and receive specific diagnostic and treatment suggestions. For people who are uncertain about the significance or urgency of new symptoms, the AI will provide reliable advice. And for patients with rare or complex conditions, it will offer invaluable second opinions. Advanced diagnostic ability combined with comprehensive healthcare information would be instrumental in reducing the 400,000 annual deaths attributed to misdiagnoses.
2. Fewer Complications From Chronic Disease
Chronic diseases like diabetes, hypertension, obesity, and asthma affect 6 in 10 U.S. adults and their complications account for 1.7 million deaths each year.
Unlike acute illnesses that appear suddenly and can usually be rapidly resolved with treatment, chronic conditions persist over time, impacting tens of millions of American lives every single day.
Caring for these conditions in a doctor’s office is episodic and far from optimal. Patients with chronic disease typically see their physician, at most, every three to four months. Periodic care means that doctors get only a snapshot of the patient’s health status, missing crucial day-to-day changes. As a result, diseases aren’t controlled as well as they should be, which leads to serious, preventable complications like heart attacks, strokes, kidney failure and cancer.
At a national level, for example, hypertension is adequately controlled just 60% of the time, and effective blood sugar management in type 2 diabetes is achieved less than half the time. The Centers for Disease Control and Prevention (CDC) highlights the damage caused by poor disease management in the U.S.:
- Proper management can reduce the risk of blindness, kidney failure, and heart disease by 40%.
- Managing blood pressure effectively can lower the risk of heart attacks or strokes by 30-50%.
- Regular eye exams could prevent 90% of diabetes-related blindness.
- Foot exams and patient education could prevent 85% of diabetes-related leg amputations.
- Early treatment of diabetes-related kidney disease could reduce the need for dialysis by a third.
Applying these percentages to the death toll from chronic disease complications in the United States, these CDC estimates suggest that more than half a million lives could be saved annually from prevented heart attacks, strokes, cancers, and infections.
Generative AI has the potential to transform the management of chronic diseases by connecting with wearable devices, informing patients about their healthcare status and suggesting medication adjustments or life-style changes. It can remind patients about necessary screenings and even facilitate testing appointments and transportation. These proactive approaches would greatly enhance disease management, reduce complications, and improve health outcomes.
3. Safer Hospitals
Patients admitted to hospitals, especially those who are critically ill, face significant risks—not just from their injury or illness but from the hospital, itself. Currently, only a fraction of patients benefits from continuous monitoring (such as ICU patients or those under telemonitoring) due to cost constraints. It’s expensive to continuously look for changes in blood pressure, alterations in pulse or unexpected cardiac arrhythmias. That’s why the majority of patients in hospital beds are checked sporadically, leaving gaps in their care.
ChatGPT, once integrated with wearable devices and bedside monitors, could radically improve inpatient care. Continuous monitoring, powered by AI’s ability to remember and compare a patient’s clinical status over time, would allow the technology to immediately alert healthcare professionals when a problem arose and facilitate rapid intervention.
Additionally, generative AI integrated with video monitoring could oversee the delivery of medical care, pinpointing any departures from established best practices. This real-time oversight could provide immediate alerts to caregivers, preventing medication mishaps and reducing the risk of infection.
In this way, this technology would help reduce the staggering 250,000 annual deaths attributed to preventable medical errors each year.
While ChatGPT and similar technologies hold immense potential, it’s essential to understand that today’s generative AI tools still require clinician supervision. Ensuring the precision of recommendations, safeguarding individual privacy and security, and addressing technological challenges are key considerations as healthcare and generative AI converge.
Still, the exponential growth of generative AI’s capabilities (doubling every year) points to a transformative future for the practice of medicine. To that end, it’s imperative for the medical community to begin to prepare for this evolution now. Medical school faculty should equip the next generation of healthcare professionals with the knowledge and tools needed to harness AI’s full potential. And medical societies must create the educational tools both clinicians and patients will use to achieve the best clinical results. With memory and GPTs, the doctor’s AI toolkit is quickly filling up.