If this Google test goes well, your next hospital visit could benefit from the help of artificial intelligence (AI). According to recent information, Google has been using its Med-PaLM-2 chatbot in a clinic in the United States, the Mayo Clinic, since April.
In a document describing in detail its work with the new AI, Google explains that its Med-PaLM-2 chatbot is capable of answering health-related questions and proposing a diagnosis based on symptoms. But the program can also perform laborious tasks, such as summarizing a medical document or organizing a patient’s data. In the future, Google hopes that AI can even help analyze an X-ray.
The chatbot is based on the Med-PaLM-2 language model, which was announced a few months ago. Med-PaLM 2 is formed from a corpus of data based on medical demonstrations performed by experts and questions / answers from medical examinations. According to Google, this training based on specific information makes the chatbot much more efficient in health-related conversations than general chatbots.
The chatbot was as efficient as a real person
According to Google, Med-PaLM-2 has shown a correct understanding of the situations it has faced in almost all areas. In short, the chatbot was as efficient as a real person. The company even said that AI has sometimes generated answers that have been preferred to those of real doctors.
Of course, it’s not perfect. Speaking on Med-PaLM-2 at the beginning of the year, Google noted that the chatbot still suffers from the accuracy problems similar to other AI products. However, most of the time, these inaccuracies resulted in irrelevant answers.
A Google research director recently said that Med-PaLM-2 is still in the early stages of development. But he also said that language models make it possible to increase the applications of AI in the medical world tenfold.
Naturally, the use of AI in the field of health raises privacy issues. However, Google has specified that users’ health data is encrypted and used exclusively by Med-PaLM-2 at the local level.
Source: “ZDNet.com “