The Doctor’s New Assistant: AI Analytics for Medicine
Generative artificial intelligence is at its most basic only a predictive algorithm, but with modern improvements in complexity, it can be used for non-trivial problems. One of these areas of use is data analytics. Chat bots such as chatGPT can be given input data such as text, numbers, or even a pasted excel file and are able to output accurate and insightful analysis. There is little doubt that this function of generative AI is of great use in the medical field, perhaps most obviously in the practice of diagnosis.
Data Analysis and Diagnosis
What is a diagnosis? Although a seemingly trivial question, it prompts a perhaps more interesting question, what produces a diagnosis? When a patient exhibits symptoms of illness, a physician is charged with determining the most likely illness at hand and this is all a matter of data. Collecting information from a physical examination, assessing patient and family health histories, and considering current common health conditions are all in an effort to build a more complete database of information about a patient. After acquiring this data, it is a physician’s job to determine the most likely ailment, essentially a process of analyzing the available data. As it turns out though, this process of analysis is something generative AI can do remarkably well. Thus, there is clearly an opportunity for this technology to have a place in the diagnostic procedure.
When considering the data necessary to diagnose a patient, there is not an incredible complexity of data. In fact, other than measurements from test results, most all of the data necessary to diagnosing a patient is textual, in the form of research. Indeed, what makes a physician so good at diagnosing patients is their consumption of knowledge throughout their career. Books in medical school, case studies during residency, and years of experience practicing are invaluable data to a doctor. This same data is easily consumed by a generative AI, but even better, much of the training data applied to modern large language machine learning models is scholarly works, including medical research and case studies. That is to say, generative AI is positioned perfectly to aid in the process.
Consider the optimal use case, a general practitioner. This is the type of doctor that deals with a wide array of health areas. Their knowledge is vast, however more shallow than a specialist in any particular field. If given access to a patient’s medical history and results from current testing, generative AI such as chatGPT, is able to correlate the given input with all the exhaustingly detailed research it had access to as training data. Given enough input, these models can suggest a number of possible ailments. Here, the reality of generative AI as a predictive algorithm seems to align perfectly with this task. Just as a doctor aims to predict the most likely diagnosis, generative AI can accomplish the same task, the only bounds are the information provided as input and as training data.
While there are considerations to be made here regarding information privacy, this could be an incredibly useful tool for medical professionals. Suggesting possible diagnoses could be a welcome aid to doctors confused by a patient's symptoms or looking to confirm their own assessment.
Aside from generative AI aiding in the greater assessment of a patient’s health, perhaps it can contribute to specific laboratory work. Given data from a test such as metabolic panels or viral antibody tests, generative AI can discuss the data textually or even generate code to plot data in a simple script. It is hard for humans to gain any insight into data by simply looking at a sheet of numbers, but this is a simple task for generative AI. Getting simple metrics is trivial, but chatbots can also talk about it. With a background of research training data, generative AI can recall implications and context around test results, offering valuable information to healthcare workers. Essentially, generative AI can have a place removing the burden of result analysis and simply giving a doctor or lab scientist the critical information.
Along the same line, generative AI is poised to offer insight past what a human may come up with. Given the vast amount of information generative AI is trained on, these models have developed an incredible ability to contextualize. In other words, with essentially immediate access to millions of articles of research (and likely access to live internet in the near future), generative AI can draw associations between any different areas. This offers the possibility of producing new insights, yet unconsidered by the modern medical field. Given large amounts of medical data, generative AI can run predictive analysis, identifying trends and patterns that may not be otherwise visible. In areas such as radiology, specialized computer vision AI has already been used to aid in data analysis, but the flexibility of generative AI would allow it to be used in just about any area of medicine limited only by the context available and data provided! This is possible because of an incredible trait of generative AI. Although in the past machine learning was a field divided into many specific data types, the trend has been towards translation of data into different forms and thus research has focused on building models that are applicable to any possible input data. To explain, consider that a number can be written as a word in English and an image can be converted into a list of numbers identifying color values for each pixel. Thus, it is technically possible to convert a picture into words which is enough for it to be used as input for any generative AI chatbot available today.
By this logic, it is clear that the assistance of generative AI is not limited to any field of medicine. With any form of data available, the contextual abilities and vast amount of training data built is accessible through this technology. The availability of chatbots such as chatGPT means that professionals have this technology at their fingertips and can benefit from its versatility in analysis. Additionally, the use of generative AI can extend beyond patient consultations and medical research. It can be utilized for medical education and training purposes, allowing doctors to simulate challenging cases, practice surgical procedures, and enhance their clinical skills in a safe and controlled environment.
Another vein of generative AI’s use in the medical field is around patient interaction. One of the remarkable utilities of modern models is their ability to perform sentiment analysis. Getting feedback from patients about care or hospital facilities or even symptomatic descriptions can be easily analyzed by artificial intelligence. This analysis can help healthcare providers gauge the overall patient experience and identify areas for improvement. Additionally, generative AI can assist in generating personalized responses to patient queries or concerns, enhancing the efficiency of communication between patients and healthcare professionals. By automating certain aspects of patient interaction, AI can free up valuable time for medical staff, allowing them to focus on providing quality care and attention to patients' specific needs all while collecting important information for care providers. Further, perhaps there is an opportunity to perform metaanalysis of doctor-patient interactions in an attempt to improve care quality. Implementation of this type of system could be through an applet integrated into an electronic health records (EHR) system to provide ease of access to doctors in their normal course of work. While preparing to see a patient, a doctor could consult a generative AI chat that has been continuously updated with patient responses to visits and would be able to provide advice for interaction and reminders of important details. Essentially, generative AI could act as an assistant for navigating a patient’s health record! The versatility of ChatGPT can also be harnessed for telemedicine, enabling doctors to provide remote consultations and guidance to patients, especially in underserved areas where access to healthcare may be limited. Given generative AI’s ability to analyze, AI could be the best option. For only the price of a computer, access to contextualized, abundant research information can be at the fingertips of every human on earth. Doctors have been the statistical algorithms of the past, but they will have help in the future. In truth, by democratizing access to such advanced tools, we can empower people worldwide, ensuring that quality healthcare reaches even the most remote corners of the globe.
One growing concern in the medical field is the shortage of doctors across the world. The ratio of doctors to humans is simply not steady and AI may be the solution. AI-powered medical chatbots like ChatGPT can help alleviate the burden on doctors by providing preliminary assessments and information to patients, allowing doctors to focus on more critical cases. This can improve access to healthcare, especially in underserved areas where the shortage of doctors is more pronounced. The abilities of generative AI to analyze data aligns perfectly with the aims of the medical field and without a doubt AI will have a place in the field. Though more complex models will be deployed for research, algorithms available right now such as chatGPT can be used to analyze patient data, give insight about test results, and even help improve doctor-patient dynamics.