Building on their groundbreaking innovation in medically tuned language models with Med-PaLM, Google has unveiled MedLM. This is their new family of foundation models based on Gemini that is fine-tuned specifically for the healthcare industry. MedLM is now available to select Google Cloud customers through Vertex AI.
MedLM offers robust performance on complex medical tasks like summarizing patient conversations and answering doctors' questions. Currently there are two models tailored to different use cases,
medlm-large for complex tasks and
medlm-medium optimized for scale across numerous workflows. Google has not yet shared any performance benchmark metrics for MedLM, so direct comparisons to Med-PaLM 2 or other models are not yet possible.
Healthcare organizations are already putting MedLM to work in real-world pilots. HCA Healthcare is testing a solution from Augmedix that uses MedLM to instantly transcribe clinician-patient discussions into medical note drafts. This ambient documentation approach stands to increase physician efficiency, reduce burnout, and enable affordable scaling across specialties.
BenchSci is utilizing MedLM to accelerate pre-clinical drug research and development. By integrating MedLM into its ASCEND platform, BenchSci enhances the speed and quality of research, ultimately contributing to quicker scientific discoveries and drug development.
Accenture and Deloitte have also teamed up with Google Cloud to apply MedLM's capabilities in areas like improving healthcare access, experiences, and outcomes as well as streamlining provider search for plan members.
As Google Research continues advancing its medical AI - from passing medical board exams to achieving expert-level performance - MedLM represents the next step in responsibly translating these breakthroughs into real improvements in healthcare. The incremental rollout focuses on safe, monitored application of generative AI rather than premature, unchecked deployment.
While MedLM's capabilities are impressive, Google AI emphasizes the model's limitations and the responsibilities of its users. MedLM, not intended as a medical device, requires careful handling, especially in clinical settings. Healthcare professionals are advised to use MedLM as an assistive tool, subject to rigorous review and validation.
The deployment of MedLM also raises important considerations regarding equity and bias. Given the probabilistic nature of LLMs, Google AI urges users to implement equity-focused evaluations to ensure that MedLM's usage does not exacerbate health disparities or perpetuate harmful biases.
With this measured approach, Google hopes MedLM will open up transformative new possibilities in medicine while prioritizing patient wellbeing above all else. The future looks bright, but there is still much progress ahead as collaborations expand and research persists. MedLM is not an endpoint, but rather part of an ongoing journey to unite human ingenuity and AI for the greater good.