Google Launches Generative Artificial Intelligence Models for Healthcare

Google announced the launch of MedLM, a suite of foundation models designed to help healthcare organizations meet their needs through generative artificial intelligence (AI), according to a blog post shared with HealthITAnalytics via email.

The launch of MedLM builds on existing work the company has done in the health AI space.

In March, the company provided updates on its healthcare AI research at Google’s annual health event, The Check Up. At the conference, Google announced a second iteration of its healthcare-tuned large language model, Med-PaLM.

At the time, the new Med-PaLM 2 achieved high performance on a United States Medical Licensing Exam (USMLE)-style test, but Google noted that more improvements needed to be made to the model before it could be utilized in real-world applications.

The model became available for limited customer testing in April to allow users to share feedback and help improve the tool.

The blog post indicates that now, Google Research team members are exploring multimodal capabilities of the large language model.

The two models under MedLM are built on Med-PaLM 2, and are designed to “offer flexibility to healthcare organizations and their different needs.” The blog post further notes that healthcare organizations are exploring a wide range of potential applications and use cases, and that the most effective model for a particular use case will vary.

To offer some flexibility for users, MedLM currently consists of two models. The first is larger to allow users to undertake more complex tasks, while the second is described as a “medium model” to help users fine-tune the tool and scale it across tasks.

In 2024, Google plans to introduce additional models into the MedLM family that are based on the recently released Gemini—the company’s newest multimodal generative AI model.

A handful of organizations with whom Google has been testing MedLM are now broadening their testing or incorporating it into workflows.

The blog post states that HCA Healthcare has been piloting a physician note-taking solution over the past several months across four emergency departments. The platform, known as Augmedix, is designed to provide Health Insurance Portability and Accountability Act (HIPAA)-compliant medical notes based on clinician-patient conversations. The Augmedix platform utilizes a combination of natural language processing (NLP) and MedLM to convert conversation data into medical note drafts, which clinicians can then review, finalize, and transfer into the patient's electronic health record (EHR).

The blog post asserts that the incorporation of MedLM is set to improve Augmedix’s performance, which could increase efficiency, reduce clinician burnout, and improve care.

BenchSci, Accenture, and Deloitte will also reportedly be utilizing MedLM for various use cases. The suite of tools is currently available to US-based Google Cloud customers through allowlisted availability in the Vertex AI platform.

Despite the promise of these technologies, some have raised questions about their use in healthcare. In an August letter to Google CEO Sundar Pichai, US Senator Mark R. Warner (D-VA), Chairman of the Senate Select Committee on Intelligence, highlighted concerns around the deployment of Med-PaLM 2 in terms of transparency, patient privacy, and ethical guardrails.

Warner underscored that while healthcare AI tools hold significant potential to improve patient care and outcomes, premature deployment of those tools could lead to a host of negative consequences, including erosions of trust in medical institutions and professionals, the exacerbation of racial disparities, and increased risk of care delivery- and diagnostic-related errors.