Chat-GPT has gained more than 100 million users since Microsoft-backed OpenAI launched the AI service five months ago. People across the globe are using the technology for a multitude of reasons, including to write high school essays, chat with people on dating apps and produce cover letters.
The healthcare sector has been notoriously slow to adopt new technologies in the past, but Chat-GPT has already begun to enter the field. For example, healthcare software giant Epic recently announced that it will integrate GPT-4, the latest version of the AI model, into its electronic health record.
So how should healthcare leaders feel about ChatGPT and its entrance into the sector? During a Tuesday keynote session at the HIMSS conference in Chicago, technology experts agreed that the AI model is exciting but definitely needs guardrails as it becomes implemented into healthcare settings.
Healthcare leaders are already beginning to explore potential use cases for ChatGPT, such as assisting with clinical notetaking and generating hypothetical patient questions to which medical students can respond.
Panelist Peter Lee, Microsoft’s corporate vice president for research and incubation, said his company didn’t expect to see this level of adoption happen so quickly. They thought the tool would have about 1 million users, he said.
Lee urged the healthcare leaders in the room to familiarize themselves with ChatGPT so they can make informed decisions about “whether this technology is appropriate for use at all, and if it is, in what circumstances.”
He added that there are “tremendous opportunities here, but there are also significant risks — and risks that we probably won’t even know about yet.”
Fellow panelist Reid Blackman — CEO of Virtue Consultants, which provides advisory services for AI ethics — pointed out that the general public’s understanding of how ChatGPT works is quite poor.
Most people think they are using an AI model that can perform deliberation, Blackman said. This means most users think that ChatGPT is producing accurate content and that the tool can provide reasoning about how it came to its conclusions. But ChatGPT wasn’t designed to have a concept of truth or correctness — its objective function is to be convincing. It’s meant to sound correct, not be correct.
“It’s a word predictor, not a deliberator,” Blackman declared.
AI’s risks usually aren’t generic, but rather use case-specific, he pointed out. Blackman encouraged healthcare leaders to develop a way of systematically identifying the ethical risks for particular use cases, as well as begin assessing appropriate risk mitigation strategies sooner rather than later.
Blackman wasn’t alone in his wariness. One of the panelists — Kay Firth-Butterfield, CEO of the Center for Trustworthy Technology — was among the more than 27,500 leaders who signed an open letter last month calling for an immediate pause for at least six months on the training of AI systems more powerful than GPT-4. Elon Musk and Steve Wozniak were among some of the other tech leaders who signed the letter.
Firth-Butterfield raised some ethical and legal questions: Is the data that ChatGPT is trained on inclusive? Doesn’t this advancement leave out the three billion people across the globe without internet access? Who gets sued if something goes wrong?
The panelists agreed that these are all important questions that don’t really have conclusive answers right now. As AI continues to evolve at a rapid pace, they said that the healthcare sector has to establish an accountability framework for how it’s going to address the risks of new technologies like ChatGPT moving forward.