On Tuesday, Google announced new health initiatives and partnerships at its annual event, The Check Up.
The Google Health team announced updated search engine capabilities, tools for creating health apps, and the latest health research based on AI technologies.
During the event, the company discussed new partnerships to advance AI-enabled ultrasound, cancer treatment, and tuberculosis screening, but focused specifically on the latest version of its large-scale medical language model, Med-PaLM.
Google first launched Med-PaLM late last year with the goal of providing high-quality answers to medical questions.
Med-PaLM is the first artificial intelligence system to achieve a score of at least 60% on a multiple-choice exam, similar to the medical licensing exam in the United States.
The second version of the technology, Med-PaLM 2, is under continued development at an "expert" level and includes medical investigation questions, the company said in a blog post. Med-PaLM 2 has an accuracy of over 85%.
Dr. Alan Karthixalingam, director of research at Google Health, said the company is also testing Med-PaLM responses against the reactions of real therapists and doctors. He said Med-PaLM responses will be judged on factual accuracy, bias, and potential harm.
Karthixalingam used controlled examples to show how Med-PaLM 2 can answer questions such as “What is the first warning sign of pneumonia” and “Can bedwetting be treated?”. »
In some cases, Med-PaLM 2 responses are comparable or even more detailed than those given by clinicians. But in others, the response from Med-PaLM 2 was not correct.
"You can tell from this kind of work that we are still learning," Karthixalingam said at the event.
Given the sensitivity of medical information, Karthixalingam said it may take time for the technology to reach ordinary consumers. He said it was important to innovate responsibly in a controlled environment.
Google will continue to work with Med-PaLM researchers and experts, and Karthixalingam said the company will provide more updates on this issue in the future.
He also said, "The potential here is huge, but responsible and ethical exploration of real-world applications is essential."