Google’s AI tool, Med-PaLM 2, designed to address medical inquiries, has undergone testing at the Mayo Clinic research hospital and other facilities since April. Med-PaLM 2 is an advanced variant of PaLM 2, which was introduced at Google I/O in May. It serves as the language model behind Google’s chatbot, Bard.
According to The Wall Street Journal, an internal email revealed that Google believes the updated model can be particularly valuable in regions with limited access to healthcare professionals. Med-PaLM 2 was trained on a curated collection of medical expert demonstrations, enhancing its ability to engage in healthcare conversations compared to generalized chatbots like Bard, Bing, and ChatGPT.
The article also mentions a May research publication by Google, demonstrating that Med-PaLM 2 still exhibits accuracy issues common to large language models. Physicians participating in the study found more inaccuracies and irrelevant information in responses provided by Med-PaLM and Med-PaLM 2 compared to their human counterparts.
However, Med-PaLM 2 performed comparably to actual doctors in various other metrics, such as displaying reasoning evidence, delivering consensus-supported answers, and demonstrating accurate comprehension.
The Wall Street Journal reports that customers testing Med-PaLM 2 will retain control over their encrypted data, which will not be accessible by Google.
Greg Corrado, Google’s senior research director, mentioned that Med-PaLM 2 is still in its early stages. While he would not personally rely on it for his family’s healthcare, Corrado believes that Med-PaLM 2 expands the possibilities of AI’s positive impact in healthcare by a significant margin.