As Google continues to fine-tune its Gemini generative artificial intelligence model—which is positioned as a core competitor of OpenAI’s ChatGPT—the tech giant is also working on other, more specialized generative AI projects.
Among those is MedLM, a group of AI models specifically geared toward the healthcare industry. Google unveiled MedLM in a blog post Wednesday, noting that its first two models are now available to U.S. Google Cloud customers via the company’s Vertex AI platform, which uses machine learning workflows to guide users through the process of training, evaluating and deploying generative AI models.
Certain markets outside the U.S., meanwhile, can access the tools “in preview.”
The MedLM models are built on Med-PaLM 2, the second iteration of Google’s foray into “medically tuned” AI tools called large language models, or LLMs. After the first Med-PaLM model made headlines last year for earning a passing score on U.S. medical licensing exam questions, Med-PaLM 2 improved on its predecessor’s score by 18% this year to achieve 85% accuracy, per Google, which equated the score to “‘expert’ doctor level.”
Differences between the two MedLM models and their uses come down to their respective sizes: “The first MedLM model is larger, designed for complex tasks. The second is a medium model, able to be fine-tuned and best for scaling across tasks,” according to the blog post.
And future additions to the MedLM family are already in the works. The company is planning to integrate Gemini-based models into the healthcare-focused group in the coming months, to further expand MedLM’s AI-powered capabilities.
The newly launched AI tools are designed to be used across the healthcare industry—in hospitals, drug development, patient-facing chatbots and more. In the blog post, Google highlighted several of the organizations it tapped to test out the MedLM suite, adding that many of them “are now moving it into production in their solutions, or broadening their testing.”
In a pilot project at HCA Healthcare, for example, the MedLM technology was used to augment technology from Augmedix that’s designed to listen in on clinician-patient conversations and automatically translate them into medical notes. According to Google, adding MedLM’s generative AI to Augmedix’s natural language processing will help to continuously improve the quality of the resulting notes, while also making it easier and more affordable to scale up the transcription technology across hospitals.
BenchSci, meanwhile, added MedLM to its existing AI-powered R&D platform, with an aim of enhancing the ASCEND engine’s ability to quickly sift through massive amounts of clinical data and identify connections between certain diseases and biomarkers.
Accenture and Deloitte also signed on as early adopters of the MedLM tools—the former to speed up AI-powered processing of clinical documents, enrollment forms, medical claims and more, and the latter to improve patient-provider matching for contact center agents.