Google worked with a panel of clinicians to get its medical large language model, Med-PaLM 2, to produce answers to patient questions that most aligned with theirs.
The tech giant took PaLM 2, its most advanced language model, and adapted it to the medical domain, according to a May 10 YouTube video from Google.
Google then trained Med-PaLM 2 and worked with clinicians in the U.S., India and the United Kingdom on a panel of questions.
The panel of clinicians answered medical questions, and Google trained Med-PaLM 2 to produce answers that were most closely aligned to that of the clinicians.
The clinicians then evaluated Med-PaLM 2 to see if its answers aligned with scientific consensus, had a lack of bias, had a low likelihood of medical harm and if the technology had precision.
The generative AI model is not yet publicly available, but was found to have 85 percent accuracy when it comes to answering medical licensing exam questions.