The release of ChatGPT in November 2022 ushered in a new era of artificial intelligence (AI), the latest developmental phase of a technology that continues to evolve in its generative capacities for ever-broadening applications by ever-larger audiences.
On a recent Ezra Klein podcast on AI, Casey Newton, the editor of the newsletter Platformer, noted that for the first time, ChatGPT gives the average person a way to interact with AI (Transcript: Ezra Klein Interviews Casey Newton and Kevin Roose. The Ezra Klein Show. The New York Times. December 1, 2023). “Even a year into folks using this now, I don’t think we’ve fully discovered everything that it can be used for,” Newton said during the podcast. “And I think more people are experiencing vertigo every day as they think about what this could mean for their own jobs and careers.”
Vertigo is an apt word for the dizzying effects of a technology that, like the cloning of Dolly the sheep, is seen with a bit of both awe and trepidation for its power to transform how we live and our understanding of what distinguishes us as human in a world with “intelligent” machines. Although AI has been around for decades, it’s only recently that the full weight of its potential use is being felt as generative AI technologies such as ChatGPT bring AI into the hands of “average people.” With such a powerful technology literally at the fingertips of anyone and everyone, both government and industry grapple with appropriate uses of AI and the potential dangers lurking within the “intelligence” of this technology.
Like many terms used repeatedly and over time, AI is now a part of everyday lexicon. But it isn’t a simple concept or an easy technology to understand. It involves multiple branches, definitions, evolutions, and applications, and is increasingly viewed as a technology for which guard rails are needed to ensure its appropriate and safe use. (See Sidebar A for a partial list of AI terminology.)
This article is the first in a series of five that looks at the broad dimensions of AI—what it is, its evolution into generative capabilities, as exemplified by tools such as ChatGPT, and how AI is applied to and supports extended (virtual and augmented) reality—within the context of healthcare in general and otolaryngology where apt. This first article lays the foundation by describing basic principles of AI and some of the ways AI technologies are currently used, or will hopefully be used, in healthcare.
As generative AI tools such as ChatGPT promise ever more sophistication and power, many wonder if AI is inching (or leaping) toward replacing humans in daily tasks, or even specialized ones, like clinical care. Is AI poised to take over essential activities of clinical care? Which ones? And should physicians be concerned over their future role in healthcare?
What Is Artificial Intelligence?
In an interview he gave prior to his death in 2011 (http://jmc.stanford.edu/artificial-intelligence/what-is-ai/index.html), John McCarthy, PhD, the Stanford University computer and cognitive scientist considered one of the founders of AI, defined AI as “the science and engineering of making intelligent machines, especially intelligent computer programs” and defined intelligence as “the computational part of the ability to achieve goals in the world.” He noted that intelligence involves mechanisms, and AI research is helping computers figure out how to carry out mechanisms. Computer programs that are able to carry out a task that requires well-understood mechanisms, he said, can be considered “somewhat intelligent.”
Dr. McCarthy cited another founder of AI, Alan Turing, as arguing in 1950 that a machine can be considered intelligent if it can “successfully pretend to be human to a knowledgeable observer” (the Turing test). Of the test, Dr. McCarthy noted that “a machine that passes the test should certainly be considered intelligent, but a machine could still be considered intelligent without knowing enough about humans to imitate a human.”
These distinctions may seem like splitting hairs or semantic gymnastics, but understanding what the intelligence part of AI means gets at the core of what AI is and is not capable of—in short, what sorts of activities AI can do as well as or better than humans, and what it falls short of, at least for now.
“Intelligence in general is a controversial topic,” said Alfred Marc Iloreta, Jr., MD, assistant professor of artificial intelligence and emerging technologies at the Graduate School of Biomedical Sciences at the Icahn School of Medicine at Mount Sinai Hospital in New York City, where he is also an assistant professor of head and neck surgery and neurosurgery and co-directs endoscopic skull base surgery.
AI has not shown the capacity for [adaptability, resilience, and variability] as of yet. But who knows? These things are changing so rapidly. —Alfred-Marc Iloreta, Jr., MD.
“Breakthroughs in AI have had an impact on our understanding of intelligence, as it is multifaceted and continually evolving,” he said. “To me, it is the ability to take input, or data, in any form, draw inferences from general data and create new principles or rules, and then apply that to solve problems and rationalize things, in addition to being able to adapt and change in a dynamic way. All of this leads to decision making and critical thinking.”
Given the infinite memory and data processing speed of computers, AI has already eclipsed humans in its ability to understand patterns and regurgitate data, Dr. Iloreta said. He pointed out that the ability of computers to find and understand patterns is part of the evolution of AI throughout many decades of research that fundamentally anchors AI in the concept of machine learning.
“Machine learning is the ability to predict outcomes based on a particular data set, and then expand on that data set by employing and deploying neural networks,” he said. “These neural networks have enabled AI systems to perform extraordinarily complex tasks and identify intricate patterns within diverse data sets, often uncovering relationships that are not immediately apparent.”
He emphasized, however, that these types of activities, as complex and sophisticated as they are, won’t necessarily be able to replace human intelligence, which includes emotional understanding and empathy. “Humans understand their own emotions and have empathy,” he said, suggesting that without these abilities, AI will not exceed or replace humans.
Other traits that humans are still better at than AI, he said, include adaptability, resilience, and variability. “AI has not shown the capacity for these as of yet,” he said. “But who knows? These things are changing so rapidly.”
The rapid pace at which AI is advancing and the still unknown answers about its full capabilities are expanding the groundswell of interest and concern about the broad application of AI across industries, governments, and public use.
In healthcare, where AI is already strongly rooted and poised for substantial growth, efforts are fully underway to educate physicians and all healthcare providers on the current and potential uses of AI in clinical practice and research. In December 2023, the publisher of the New England Journal of Medicine (NEJM) announced the launch of a new, peer-reviewed monthly journal called NEJM AI (Peters D. NEJM AI to educate clinicians about artificial intelligence applications in medicine). The first issue (January 2024) is currently available at https://ai.nejm.org/toc/ai/current.
“What many people don’t know is that AI is already being used in medicine,” said Eric Rubin, MD, PhD, editor-in-chief of the NEJM and NEJM Group publications, in the same press release. Below is a sampling of some of the many ways AI is currently used in healthcare in general, and the currently limited ways it’s being used in otolaryngology.
Artificial Intelligence in Healthcare
Current use of AI and machine learning is expanding throughout healthcare, as is research into its potential applications that many people predict will revolutionize the way healthcare is delivered. From drug discovery to personalized medicine to medical image analysis and predictive analytics, the power of AI to be trained in and excel at a range of clinical and administrative activities increasingly demonstrates its revolutionary capacities. (Table 1 lists a number of areas in which AI-based technology is being used or studied.
Among the most common current uses of AI-based technology in healthcare is in medical imaging and radiology. A 2022 study on the characteristics and intended use of FDA-approved AI/machine learning-enabled devices in 2021 found that most devices were in radiology (70.3%), followed by cardiology (12.0%). Of the 241 radiology-related devices, 117 (48%) were for diagnostic assistance (20.5% for assessing breast lesions and 14.5% for assessing cardiac function on echocardiogram) and 14.1% were for image reconstruction (Int J Med Inform. 2022. doi:10.1016/j.ijmedinf.2022.104828). The most common applications for the 41 cardiology-related devices were for detecting electrocardiography-based arrhythmia (46.3%) and monitoring hemodynamics and vital signs (26.8%).
One huge benefit of AI-assisted radiology for diagnostic assistance is its impact on early disease detection. Machine algorithms trained on large medical data sets to recognize abnormalities on imaging are able to rapidly and accurately identify potential issues that may take significantly longer for radiologists to identify through manual analysis. In a 2021 study (Nat Commun. 2021. doi:10.1038/s41467-021-26643-8), researcher at Tulane University, for example, found that AI did as well as or better than pathologists in accurately detecting and diagnosing colorectal cancer from tissue scans (Shah P. From Pixels to Precision: The Impact of AI and ML on Medical Imaging. Einfochips. Sept. 28, 2023).
This is just a sampling of what lies ahead for healthcare as it grapples with how best to apply the power of these AI tools—a power that’s only increasing as generative AI (covered in the second and third articles in this series) continues to show its marvels, excelling at activities once found only in the realm of human activity.
Dr. Iloreta underscored the reality that healthcare is still trying to figure out how to deal with AI. “We are in a mode right now of trying to figure out the rules of engagement,” he said. Basic questions still need to be answered, he added, such as, “What can be allowed?” “What data can we input?” and “Do you let AI read a CT scan or make a medical decision?”
For otolaryngologists, these questions may not be as pressing yet, given that the specialty is not as data driven as other specialties such as nephrology and cardiology. Dr. Iloreta noted that a lot of what otolaryngology manages are quality-of-life issues, for which the data used to make decisions or determine success are not easily measured. However, he sees AI playing a key role in otolaryngology when it comes to repetitive tasks such as documentation, which, he said, most otolaryngologists aren’t happy performing. He also sees otolaryngologists as a natural fit for technologies using extended reality, and highly doubts that AI will or can replace their jobs as clinicians.
Dr. Iloreta underscored, however, that as AI broadens its reach into otolaryngology specifically and healthcare in general, attention needs to be paid to safe and ethical practices. “I am hopeful that there will be a rationalization and set of guidelines of what is acceptable, not acceptable, and what is a meaningful and safe use of AI in an effort to use it in what is useful and helpful,” he said. “It is very important for us to approach AI with cautious optimism and not run haphazardly into using it, but at the same time we shouldn’t run away from it or shun it.”
Measures are underway by those companies and practices using AI to ensure its safe and appropriate use. Among these is a commitment by 28 healthcare providers and payers to align industry action on AI to ensure that AI leads to healthcare outcomes that are Fair, Appropriate, Valid, Effective, and Safe (called “FAVES”) following an executive order by the Biden Administration addressing the safe and responsible use of AI in healthcare. In December 2023, the Biden Administration finalized its rule on transparency requirements for AI and other predictive algorithms used in healthcare to promote the responsible use of AI.
Mary Beth Nierengarten is a freelance medical writer based in Minnesota.
AI Terminology
The following terms are part of the AI lexicon. Although this list isn’t exhaustive, it’s enough to get you started.
Algorithm: A set of rules or instructions telling a computer what to do, typically to perform a computation or solve a problem.
Deep Learning: A subset of machine learning and a complex form of AI that imitates the way human brains process data and create patterns for making decisions. It uses neural networks that allow for unsupervised learning from unstructured and unlabeled data.
Large Language Model (LLM): A computer program, such as ChatGPT, designed to understand and generate human-like text based on patterns and information it has learned from large amounts of inputted text data such as books, videos, pictures, etc. It uses natural language processing to recognize and learn patters and relationships between words, which helps LLMs to generate responses that sound human.
Machine Learning: A branch of AI using large sets of data to create algorithms to recognize patterns, make predictions, or solve specific tasks.
Neural Networks: A technique used in deep learning that relies on a neural network of tiny, interconnected nodes that work together to process information, similar to the interconnection of neurons in the human brain. Feedback and training improve neural networks.
Natural Language Processing (NLP): A branch of AI that uses algorithms to train machines to understand and work with human language and simulate human conversation.
Sources:
https://connect.comptia.org/content/articles/artificial-intelligence-terminology
https://www.heinz.cmu.edu/media/2023/July/artificial-intelligence-explained