The technology has many applications in medicine, but hurdles to widespread adoption remain
In November 2022 the world was introduced to ChatGPT—an artificial intelligence (AI)-powered chatbot capable of mirroring intuitive human conversation. Since then, this technology has captured the attention of millions that have employed it to write everything from songs and poetry to essays and python code.
In medicine, ChatGPT is a major topic of conversation among providers and patients alike. This innovation has the potential to automate daily tasks like generating patient records or writing reports. While still in the early stages of development and use, it is projected to feature diagnostic and treatment use in the future.
But like other AI systems, ChatGPT has challenges, and further testing to undergo before achieving widespread use in health care.
What is ChatGPT?
ChatGPT is the latest evolution of a language model developed by OpenAI known as GPT-3. ChatGPT is trained on extensive amounts of data to learn language patterns. The process is designed to ensure accuracy when predicting what comes next in a sequence of words.
ChatGPT is taught and refined using a combination of supervised and reinforcement learning. IBM explains the former as the "use of labeled datasets to train algorithms to classify data or predict outcomes accurately." This approach instructs computers how to learn or identify topics.
ChatGPT is optimized through reinforcement learning from human feedback. Here, human experts teach the machine likely and ethical responses, or decisions to support users. Open AI describes ChatGPT’s dialogue features as being able to "answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests."
ChatGPT’s potential uses for physicians
According to an article from Yahoo! Finance, ChatGPT amassed 1 million users within its first week. Among these users are health care professionals leaning into AI as the near future of medical management. ChatGPT is ushering in a new phase of health care assistance in the following ways:
“In my view, patient interactions and care should be offered inherently through chat first, virtual care second, and in-person care, third,” says Ali Parsa, M.D., founder and CEO of Babylon, a global AI and digital health platform. “Moving to conversational systems powered through AI models such as GPT-3 (and others), which are contextualized with patient information, will provide more personalized and clinically accurate answers for patients.”
Automating administrative functions. Studies have shown thatdoctors and their staff spend around 16.4 hours per week navigating insurance approvals for patient medication, procedures, and other medical services. Focusing on administrative duties allows less time for providing medical care.
ChatGPT may be used to perform administrative tasks such as scheduling appointments, simplifying notes, and other repetitive daily tasks. “Conversational AI can support automation and delivery of improved clinical care,” Parsa explains. “In terms of automation, examples include summarizing clinical consultations writing referral letters.” Using suitable prompts, ChatGPT can be trained to draft letters seeking prior authorizations, appeals of insurance denials, and other claims, Parsa says.
Challenges to ChatGPT adoption in health care
While promising, ChatGPT technology faces challenges to widespread adoption in health care.
For one, the information used to train the algorithm only goes through 2021, limiting its usefulness about anything post-2021.
Users also need help with accuracy in answers fielded by the algorithm. ChatGPT has been known to make up references or factually provide incorrect information. And unlike human interactions that may admit uncertainty, this model may disperse wrong, biased, or unsuitable responses without caveats for potential error.
Speaking on the challenges and potential for ChatGPT in health care, Parsa says, “The challenge of mass adoption of ChatGPT is the same challenge of revolutionizing health care in the U.S. in general; it's a complicated system that takes time to change. The building blocks are in place, and it will be up to the innovators to unlock the value on a day-to-day basis, by leveraging technologies such as ChatGPT that can be personalized to health care.”
Plumptre is a freelance health tech writer, with a special interest in AI in health care.