Google’s medical chatbot is already being tested in hospitals
Google’s Med-PaLM 2, an artificial intelligence tool designed to answer health questions, has been undergoing trials at the Mayo Clinic Research Hospital since April, according to The Wall Street Journal. Med-PaLM 2 is a variant of the PaLM 2 language model that underlies Google Bard, but adapted to medical topics. It was announced at the Google I/O conference in May this year.
Med-PaLM 2 has been trained on a selected set of medical expert demonstrations, which Google believes will allow it to have better health conversations than general chatbots such as Bard, Bing, and ChatGPT. According to the WSJ, an internal email says that Google believes its updated model is especially useful in countries with “more limited access to doctors.”
A recent Google study showed that Med-PaLM 2 is still sometimes inaccurate, which is common for large language models. Thus, during the testing, doctors found more inaccuracies and irrelevant information in the answers provided by Med-PaLM and Med-PalM 2 than in the answers of real doctors.
However, in all other respects, Med-PaLM 2 performed about as well as real doctors. The chatbot was able to give reasons for its reasoning, provide answers confirmed by consensus, and understand the questions correctly.
Customers testing Med-PaLM 2 will be able to control data privacy, as all information will be encrypted and Google will not have access to it.
According to Greg Corrado, senior director of research at Google, Med-PaLM 2 is at an early stage of development. Corrado said that although he would not want the chatbot to become part of his own family’s “medical journey,” he believes that Med-PaLM 2 expands the possibilities of using artificial intelligence in healthcare, which is objectively valuable for both patients and clinics, as it significantly saves doctors’ time.
