Google's AI chatbot for medical support has been tested in hospitals

Google's Med-PaLM 2, a new artificial intelligence tool designed by Google to answer medical information questions. Has been tested at the Mayo Clinic research hospital, among others, since April, according to The Wall Street Journal.

Med-PaLM 2 is a variant of PaLM 2, announced at Google I/O this past May. PaLM 2 is the language model that underpins Google's Bard assistant.

According to the WSJ, an internal email they viewed says that Google believes its updated model could be useful especially in countries with "restricted access to doctors". Med-PaLM 2 is trained on a set of presentations from medical professionals, which Google believes will make it a better performer in medical discussions than generic chatbots like Bard, Bing, and Chat GPT.

The WSJ article also mentions research that Google published in May showing that Med-PaLM 2 still suffers from some of the accuracy problems we often see in large language models. In this study, physicians discovered more errors and irrelevant information in Google's Med-PaLM and Med-PaLM 2 responses than similar responses from other physicians.

 

In most other cases, however, the answers were supported by consensus or showed no sign of misinterpretation. Med-PaLM 2 works relatively well like real doctors.

The Wall Street Journal reports that Med-PaLM 2 trial customers will control their data, it's encrypted, and Google won't have access to customer data.

According to Google senior research director Greg Corrado, WSJ said that Med-PaLM 2 is still in the early stages and is gradually being finalized. Corrado said that while he doesn't want it to be part of his "family healthcare journey," he believes Med-PaLM 2 "expands more in the medical field where artificial intelligence can be useful and expands them tenfold."