Search engine giant is developing an updated version of its medical AI that can answer health-related questions

The Google Health team shared updates about new features coming to search, tools for building health apps, and the latest in artificial intelligence-powered health research Tuesday at its annual event called "The Check Up.".

‍

The company discussed new partnerships with the goal of developing AI-assisted ultrasounds, cancer treatment, and tuberculosis screenings. However, one of the most talked-about aspects of the event was the release of the latest version of its medical large language model, Med-PaLM.

‍

Med-PaLM was introduced by Google late last year. This system is designed to respond to medical questions in a high-quality manner with accurate responses. It is known that Med-PaLM is the first AI system to have achieved a passing score, or more than a 60%, on multiple choice style questions similar to those used in the U.S. medical licensing exams, which is the highest possible score.

‍

In the second iteration of Med-PaLM, the company claims that it consistently performed at an "expert" level on medical exam questions when using the technology. Compared to its earlier version, Med-PaLM 2 has achieved more than 85% accuracy, an improvement of 18% over its earlier version.

‍

Dr. Alan Karthikesalingam, one of Google Health's research leads, told Trade Algo that his company was also testing Med-PaLM's answers against those provided by real doctors and clinicians. There is a thorough scrutiny of Med-PaLM's responses in order to determine their factual accuracy, bias, and potential for harm.

‍

As part of the demo, Karthikesalingam showed controlled examples of how Med-PaLM 2 might be used to answer questions such as “what are the first warning signs of pneumonia? ” and “can incontinence be cured? Med-PaLM 2 was able to provide answers that were on par with, and even more detailed than the answers provided by clinicians. However, Med-PaLM 2's responses weren't as accurate in other cases.

‍

“From this type of work, it is evident that we are still learning,” Karthikesalingam said during the event.

‍

Karthikesalingam said it may take a while before consumers can access this technology because of the sensitive nature of medical information. Responsible innovation and controlled environments are important, he said.

‍

The company plans to continue to collaborate with researchers and experts on Med-PaLM, and Karthikesalingam said in the future Google will be sharing more information about the project.

‍

“There is tremendous potential here,” he said, “but it is crucial that real-world applications are explored in a responsible and ethical manner.”