Google’s AI tool, Med-PaLM 2, has been undergoing testing since April at prominent medical institutions, including the Mayo Clinic research hospital, according to The Wall Street Journal. Derived from PaLM 2, Med-PaLM 2 is a specialized version introduced at Google I/O in May (pdf).
Med-PaLM 2 has been trained on a carefully curated collection of expert medical demonstrations, enabling more effective healthcare conversations than generalized chatbots. While accuracy challenges persist, Med-PaLM 2 has demonstrated comparable performance to human doctors in various metrics, including reasoning, consensus-supported answers, and comprehension.
During the testing phase, customers will have control over their encrypted data to ensure privacy. Greg Corrado, the Senior Research Director at Google, recognizes that Med-PaLM 2 is in its early stages of development but believes in its potential to enhance healthcare significantly.
At the recent “The Check Up” event, Google announced new health initiatives and partnerships, highlighting Med-PaLM as a key component. Med-PaLM 2 has consistently performed at an “expert” level, surpassing 85% accuracy and outperforming its predecessor by 18%.
Controlled examples showed Med-PaLM 2‘s detailed answers to medical questions, often comparable to or better than those provided by clinicians. However, accuracy discrepancies still exist, indicating ongoing learning.
Dr. Alan Karthikesalingam of Google Health emphasized the importance of responsible innovation and the need for real-world applications to be explored ethically. Med-PaLM’s widespread accessibility may take time as Google continues to refine the technology in collaboration with researchers and experts.
Google’s Med-PaLM 2 represents a promising advancement in AI-driven healthcare. With ongoing testing, user-controlled data, and a commitment to responsible innovation, Google aims to enhance healthcare outcomes and improve access to medical expertise.