Google’s Expansion of Med-PaLM 2
Google plans to extend its healthcare-focused AI model, Gemini, to MedLM in the future. The MedLM suite includes both a large and a medium-sized model, both built on Med-PaLM 2, a language model trained on medical data that was first announced in March. Starting Wednesday, eligible Google Cloud customers in the U.S. can access MedLM. Although the cost of the AI suite varies depending on usage, the medium-sized model is more cost-effective to run. Companies like HCA Healthcare, a leading U.S. health system, have been testing Google’s technology and recognizing its potential impact.
Greg Corrado, Head of Google’s Health AI, explained that Med-PaLM 2’s use cases have evolved since its announcement. Rather than focusing solely on answering accessible questions about diseases, customers primarily seek AI assistance with back-office and logistical issues, such as managing paperwork. For example, HCA Healthcare has been successfully using MedLM to automate documentation for emergency medicine physicians, aiming to generate more than half of a note without human assistance. Dr. Michael Schlosser, Senior Vice President of Care Transformation and Innovation at HCA, expressed enthusiasm about automating laborious processes, such as nurse handoffs.
Challenges and Caution for Wider Adoption
However, the adoption of AI models in healthcare still faces several challenges. Schlosser mentioned the need to minimize the generation of incorrect information and the necessity of developing best practices in collaboration with Google. Managing token limits and ensuring the AI model’s accuracy over time have also proven to be additional hurdles for organizations like HCA. While initial responses to MedLM have been positive, healthcare providers remain cautious, recognizing that further refinement is required before these AI models can be scaled up across healthcare systems.