The Google LLC logo can be seen at the Google Store Chelsea in New York City.
Shannon Stapleton | Reuters
Google On Tuesday, it announced new health initiatives and partnerships at its annual event called “The Check Up.” The Google Health team shared updates on features coming to Search, tools for building health apps, and the latest in health research powered by artificial intelligence.
During the event, the company discussed new partnerships to help develop AI-assisted ultrasound, cancer treatment, and tuberculosis screening, but the latest version of its large-scale language model for medicine, called Med-PaLM, was particularly talked about.
Google first introduced Med-PaLM late last year. It is designed to provide quality answers to medical questions. Med-PaLM is the first AI system to achieve a passing score (60% or higher) on multiple-choice questions similar to those used in the US medical licensing exam.
According to the company, the second iteration of the technique, Med-PaLM 2, performed consistently at the “expert” level on health screening questions. Med-PaLM 2 made him reach an accuracy of over 85%, making him score 18% higher than the previous results.
Dr. Alan Karthikesalingam, Head of Research at Google Health, said the company is testing Med-PaLM responses against responses from real doctors and clinicians. He said Med-PaLM’s responses will be evaluated for factual accuracy, bias and potential for harm.
Karthikesalingam gave a controlled example (the demo wasn’t live). It showed how Med-PaLM 2 answers questions such as “What are the first signs of pneumonia?” and “Will the incontinence be cured?” However, in other cases, the Med-PaLM 2 response was less accurate.
“This kind of work shows us that we’re still learning,” Karthike Salingham said during the event.
Given the sensitive nature of medical information, Karthikesalingam said it could be some time before the technology reaches the average consumer. He said it’s important to innovate responsibly and in a controlled environment.
We will continue to work with researchers and experts on Med-PaLM. Karthikesalingam said he will share more updates in the future.
“The possibilities here are tremendous,” he said.