9-7-2023 (CALIFORNIA) Google’s Med-PaLM 2, an AI tool developed to provide answers to medical queries, has been undergoing testing since April at prestigious research institutions, including the Mayo Clinic, according to a report by The Wall Street Journal. Med-PaLM 2 is a variant of PaLM 2, which was unveiled at Google I/O in May. PaLM 2 serves as the underlying language model for Google’s Bard.
Internal emails viewed by WSJ revealed that Google believes its updated model can be particularly valuable in regions with limited access to medical professionals. Med-PaLM 2 was trained using a curated dataset of medical expert demonstrations, which Google believes will enhance its performance in healthcare-related conversations compared to general chatbots like Bard, Bing, and ChatGPT. However, the report also mentions a study published by Google in May, indicating that Med-PaLM 2 still exhibits some of the accuracy issues often observed in large language models. Physicians participating in the study identified more inaccuracies and irrelevant information in answers provided by Google’s Med-PaLM and Med-PaLM 2 compared to those provided by other doctors.
Nevertheless, in terms of metrics such as evidence of reasoning, consensus-supported answers, and correct comprehension, Med-PaLM 2 performed on par with actual doctors. WSJ reported that customers participating in the testing of Med-PaLM 2 will retain control over their encrypted data, with Google having no access to it.
Google’s Senior Research Director, Greg Corrado, stated that Med-PaLM 2 is still in its early stages. While he personally wouldn’t want it to be part of his family’s healthcare journey, Corrado believes that Med-PaLM 2 expands the areas where AI can be beneficial in healthcare by tenfold.
