J.P. Singh is a Distinguished University Professor at George Mason University, where he directs the Center for AI Innovation. He writes on the political economy of technology, and is currently in remission from esophageal cancer.
I have just been through nearly a year of grueling treatment, diagnosed with advanced esophageal cancer on a trip to India last July. Along the way, I used LLMs to inform myself. Do AI-informed patients improve their healthcare prospects and peace of mind? I did.
Most doctors were receptive to my AI-enabled queries, even helping me counter ‘hallucinatory’ results‘ from the LLMs. A doctor friend at one point did warn me that there is no definitive evidence confirming that informed cancer patients improve the success rate of their treatment.
That statement left me brooding and sent me back to my research. As a political economy scholar and one who studies AI policies, I draw from a large body of evidence showing that consumers and producers make better decisions with more information.
Patients are not economic actors, nor do they make all the decisions regarding their healthcare. Nevertheless, economic reasoning about the value of better information may be useful. To the extent that doctor-patient relationships are important for effective healthcare, a patient’s use of AI could either assist or damage that relationship – and, possibly, the course of treatment.
After a year of using AI to inform myself, my answer as a patient-scholar is consistent with most forms of scientific reasoning: it depends.
First, it depends on the type of query. AI was useful for interpreting lab, radiology, and pathology reports in everyday English, and for asking specific questions of my doctors. Preliminary searches of scientific publications confirm these benefits.
Second, it depends on patients and the way they value degrees of trust and thresholds for uncertainty. My preference for staying well-informed may not be shared by everyone. Some patients trust their doctors completely; I don’t distrust mine, but trust itself depends on the underlying information. Patients, like firms in the economy, are also trying to reduce uncertainty and anxiety. Once upon a time, we consulted books for health information. Now, having answers from LLMs helped reduce my anxiety and improve my well-being.
Third, it depends on the quality of the LLMs themselves – specifically their underlying data sets and inferential algorithms. When I was first diagnosed with esophageal cancer, the LLM provided me with an overall survival rate of 25-30 percent. Clicking through to the underlying studies showed the data were twenty years old. My treatment was based on two recent trials – the Matterhorn trial and the ESOPEC study. Once I mentioned these trials in my query, the LLM accurately provided the published two-year event-free survival rate of 67.4 percent.
But the overall survival rate depends on multiple factors: success of surgery (an esophagectomy), age, overall health, cancer staging, demographics, and ethnicity. Esophageal cancer typically strikes heavy drinkers and smokers. I do not smoke or drink – though it is relatively high among the North and Northeastern states of India (I grew up in Himachal). LLMs might not be able to calculate the complex interplay of these factors. I had lymphoma in 1998-99; radiation from 28 years ago could be another cause. The doctors told me this. The LLMs didn’t.
This brings me back to the overarching question of information itself. There is an intrinsic value to being well-informed, beyond whether it improves my chances of survival. I also understand that given competing demands on my time, I do not stay informed on all issues. Political scientists write about rationally ignorant voters who aren’t knowledgeable on issues that don’t concern them. Some patients may trust their doctors as their only source of information. I need multiple sources to trust my treatment and my physicians – and multiple queries across LLMs to trust their answers.
Did being well-informed improve the effectiveness of my treatment? As my doctor friend noted, we do not have definitive evidence. But scientific enquiries begin with well-informed gut feelings. My supposition remains that being informed improved my well-being and perhaps the effectiveness of my treatment.
The question for our age is no longer whether staying away from LLMs will improve the effectiveness of treatment. It is: given LLMs, under what conditions can we improve treatment? The answers will depend on how we formulate the questions and suppositions.
My cancer is now in remission, but I must wait 3-5 years to know whether I am cured. My AI chatbot, depending on the way I ask the question, gives me rates of 65-85 percent for making it to five years. This is suggestive and probabilistic, but not definitive. I prefer scientific answers over misleading certitudes.
The Tribune is committed to publishing a diversity of opinions. We’d like to hear from you. Send your articles to our email:opinion@merotribune.com.
Follow the Mero Tribune on Facebook and Twitter.
Advertisement
