L. John Greenfield Jr, MD, PhD, “spoke” about the future of artificial intelligence in neurology with ChatGPT-4—one of the largest language modules that have grabbed the field’s interest.
ARTIFICIAL INTELLIGENCE (AI) is all the rage. Applications range from writing articles or term papers to recognizing faces, generating paintings in the style of a specific painter, and writing computer code. In neurology, AI imaging software known as RapidAI helped turn the tide for interventional stroke care by improving patient selection.1,2
But is there a dark side to AI? Will software replace trained neurologists in making diagnostic or therapeutic decisions? What happens when the software is wrong? What are the implications of AI for academic neurology? To get a feeling for what the future might bring, I had a chat on Microsoft’s Bing with the latest version of OpenAI’s large language model (LLM) software, ChatGPT-4.3 The following is a lightly edited version of this conversation.
Note: ChatGPT-4 provided its own references, and thus will have a separate reference list.
LJG: Tell me a little about yourself.
ChatGPT-4: I’m an AI language model created by OpenAI. I’m designed to help you with your queries and provide you with information. I can help you with a wide range of topics such as science, technology, entertainment, and more.
LJG: What can AI do to help us take care of patients with neurological diseases?
ChatGPT-4: AI can help analyze medical data in disease prevention, diagnosis, patient monitoring, and development of new protocols, but can also assist clinicians in dealing with voluminous data in a more accurate and efficient manner.1-5 AI has been helpful in neurodegenerative disorders including Parkinson’s, Alzheimer’s, Amyotrophic Lateral Sclerosis (ALS)1 as well as traumatic brain injury (TBI).3 AI-enabled CT scanning for hemorrhagic patients allows automation of lesion segmentation and detection of hemorrhagic expansion.1 Incorporating advanced AI technologies into the delivery system of neurologic care has tremendous potential to improve health care, but also carries with it unique challenges.4,5
LJG: What databases do you have access to?
ChatGPT-4: As an AI language model, I have access to a wide range of databases such as PubMed, Google Scholar, and more.
LJG: Some physicians are concerned that AI might replace them in making decisions about patient care. Do you think this is possible?
ChatGPT-4: AI is designed to assist physicians in making decisions about patient care, not replace them. AI can help physicians analyze medical data in disease prevention, diagnosis, patient monitoring, and development of new protocols, but can also assist clinicians in dealing with voluminous data in a more accurate and efficient manner. AI can also help to reduce the workload of physicians, allowing them to focus on more complex cases and to spend more time with their patients. AI can help physicians make more informed decisions about patient care, but it cannot replace the human touch and empathy that is required in patient care.
LJG: What role could AI play in teaching neurology to medical students and residents?
ChatGPT-4: AI can play a significant role in teaching neurology.6-9 AI can provide trainees with interactive simulations, virtual reality, and augmented reality experiences.4 Interactive simulations of real-life medical scenarios10 can allow them to practice medical procedures and techniques,11 which help them learn in a more efficient and effective manner.6,12,13 AI can also provide personalized feedback and assessments.4 AI can also provide access to a vast amount of medical data and research.6
LJG: Give an example of how ChatGPT-4 could provide positive feedback to a medical student who had disorganized patient presentations and an average knowledge base, but good neurological exam skills.
ChatGPT-4: “Congratulations on your excellent neuro exam skills! Your ability to accurately assess patients’ neurological function is a valuable asset. I also appreciate your commitment to improving your patient presentations despite their initial disorganization. I have no doubt that with practice, you will continue to improve your presentation skills and become an even more effective communicator with patients and colleagues. Keep up the great work!”
LJG: AI is used in the interpretation of medical images such as brain MRI and CT scans. Could it be used to integrate information and help with diagnoses?
ChatGPT-4: Yes, AI is already being used to interpret medical images.14-17 AI can support the diagnosis and treatment decision-making process by facilitating the integration of the imaging results within the patient electronic medical record.14 Once incorporated, the image can then be used alongside patient clinical data and medical history in computer-aided diagnosis. AI can also help medical professionals identify imaging abnormalities and enhance tissue-based detection and characterization.
LJG: Could AI be used as a medical scribe to help physicians generate notes for the medical record?
ChatGPT-4: Yes. AI medical transcription uses AI and natural language processing to listen into the natural patient-clinician conversation, parse out the medically relevant information, and summarize that information into compliant notes that map back to the appropriate fields of an EHR.18-22 AI-powered medical scribes bring the quality and experience of a human medical scribe, but at the cost and scalability of a dictation service.18 Nuance has created an AI-powered ambient clinical intelligence solution called the Dragon Ambient eXperience (DAX) that can replace medical scribes.19
LJG: What are the limitations of using AI? Have there been examples of bad outcomes when physicians have trusted AI decision support?
ChatGPT-4: Although AI has come a long way in the medical world, human surveillance is still essential.23-26 AI may overlook social variables such as class, gender, ethnicity, age grouping, and group identity.24 Inaccuracies are still possible.23 AI is susceptible to security risks.23 AI may lead to unemployment.24 However, AI can help medical professionals identify imaging abnormalities and enhance tissue-based detection and characterization.23 In 2016, researchers at Beth Israel Deaconess Medical Center reported that an AI-powered diagnostic program correctly identified cancer in pathology slides 92% of the time, just shy of trained pathologists’ 96%. Combining the 2 methods led to 99.5% accuracy.26
[Note: This positive spin on AI’s potential didn’t provide any examples of errors, so I tried again.]
LJG: Provide examples of bad outcomes when physicians have trusted AI decision support systems.
ChatGPT-4: While AI decision support systems have the potential to greatly improve medical care, there have been instances where their use has led to negative outcomes. Here are a few examples:
LJG: What are the potential uses of AI for writing academic papers or grant proposals in neurological research?
ChatGPT-4: Here are a few examples:
Overall, the use of AI in academic writing can potentially save researchers time and improve the quality of their work. However, it is important to note that AI systems should not replace human expertise and judgment, and that their use should be carefully monitored to ensure that the results are accurate and unbiased.
LJG: How could AI be used to generate new medical knowledge? Has AI been used to look for new diseases or conditions based on large-scale information or data correlations that have not been previously detected?
ChatGPT-4: Yes. Artificial intelligence in medicine is the use of machine learning models to search medical data and uncover insights to help improve health outcomes and patient experiences.27,28 In a proof-of-concept study published in Frontiers in Immunology, a new artificial intelligence (AI) algorithm was developed which uses novel weight engineering methods to extract meaningful biological knowledge learned by AI models.29 Machine learning (ML), large and often unstructured datasets, advanced sensors, natural language processing (NLP) and robotics are all being used in a growing number of health care sectors.28 AI has been used to predict the structures of almost every protein made by the human body.30 Attention maps may be used in the clinic to aid diagnosis, discover new medical knowledge, and identify novel features and characteristics of medical tests.27
LJG: How do patients feel about talking with AI? Could it provide psychotherapy or counseling?
ChatGPT-4: Patients’ feelings about talking with AI vary depending on the clinical application.31-34 For example, a survey from close to 1000 respondents showed that most patients believed that AI would make health care much better or somewhat better (11% and 45%, respectively) with relatively few that feel AI would render health care worse or much worse (4% and 1%, respectively).31 However, participants also voiced concerns that AI tools might increase health care costs and that those costs might be passed on to patients.34 AI can provide psychotherapy or counseling, but it is not a substitute for human interaction. AI can be used to provide support and guidance to patients, but it cannot replace the human touch that is so important in psychotherapy and counseling.
LJG: What will be the future of AI in neurology?
ChatGPT-4: The future of AI in neurology is promising.35-38 AI can be used for predicting outcomes of seizure disorders, grading of brain tumors, upskilling neurosurgical procedures, and rehabilitation of stroke patients with smartphone apps monitoring patient symptoms and progress.35 AI can now diagnose stroke from CT/MRI scans and detect papilledema and diabetic retinopathy from retinal scans.35 AI will cause a paradigm shift in the delivery of health care.36
The ability of ChatGPT-4 to harvest information from the internet and package it into coherent, well-organized, and (mostly) grammatically correct paragraphs is both exciting and a bit frightening. Although the chatty quality of the interface inspires confidence, the answers are often superficial and do not reflect “thoughtful” consideration—they are search results presented in a better organized and linguistically sophisticated way. Moreover, the information provided by AI (or any search engine) is based on unseen algorithms that may be influenced by volume of citations (ie, popularity) or, worse, by undeclared commercial interests. However, ChatGPT-4 and other LLMs are just 1 aspect of AI, and the potential of AI approaches to assist in neurological research, education, diagnosis, and care is enormous.4 Even so, AI lacks judgment based on real-world experience and the intangible information derived directly from patients and the clinical situation. AI has a lot of potential but still requires oversight. For now, our jobs appear safe.