At I/O this year, Google unveiled the PaLM 2 AI language model, a competitor to OpenAI’s GPT-4 and the brains behind Google’s AI chatbot, Bard. Today we learned that Google’s Med-PaLM 2, a version of the AI tool designed to specifically answer questions about medical information, is already in testing at the Mayo Clinic research hospital.
According to The Wall Street Journal, the technology has been used in the hospital, along with other locations, since April. Google had fed questions and answers from medical licensing exams to the bot, with the hope that it will ultimately be better able to have conversations about medical issues than rival Microsoft’s ChatGPT.
Med-PaLM 2 is capable of answering medical questions as well as summarizing documents and large volumes of health data and creating reminders. According to an internal email obtained by WSJ, Google thinks the language model can potentially “be of tremendous value in countries that have more limited access to doctors.”
Early results from the test highlight some of the same accuracy issues found in responses from other chatbots. However, in many other metrics, PaLM 2 performed similarly to human doctors.
The use of AI in medical settings has raised a few red flags with privacy experts who are concerned about the companies behind them using sensitive information for other purposes. According to WSJ, customers testing the language model retain control of their data. That data is also encrypted in such a way that Google itself will not have access to it.
Both physicians testing Med-PaLM 2 and Google say that while the product is promising, it isn’t quite ready to replace a trip to the doctor to diagnose or treat an issue. We’ll also likely see a bit more regulation come into play before AI can put on a virtual white coat and take the place of a doctor; however, we’re well on our way.