MICROSOFT CHATBOT WILL BE YOUR MEDICAL DOCTOR......SOON!!!!

 

 There is a lot of talk about the future possibilities and dangers of AI, but health is one area where technology has already made tangible progress, and algorithms can be used for certain medical scans just like humans. can be read. Google first unveiled an AI tool for answering medical questions called Med-PaLM in a preprint survey in December. Unlike ChatGPT, it is not open to the public.

The US tech giant says Med-PaLM is the first major language model to pass the US Licensing Examination (USMLE). This is an AI technique trained on large human-made text sets.

The passing score for this exam, taken by medical students and prospective doctors in the United States, is about 60 percent. In February, a study reported that ChatGPT achieved passing or near-passing results.

In a peer-reviewed study published Wednesday in the journal Nature, Google researchers said Med-PaLM scored 67.6 percent on USMLE-style multiple-choice questions.

"Med-PaLM shows promising results, but still lags behind physicians," the study reads.

Google has developed a new metric to recognize and curtail "hallucinations," which refer to false information provided by AI models.

Karan Singhal, a Google researcher and lead author of the new study, told AFP the team used the benchmark to test a new version of the model, with "very exciting" results. . Med-PaLM 2 achieved a score of 86.5 percent on the USMLE exam, almost 20 percent better than the previous version. This was revealed in a non-peer-reviewed preprint study published in May. These AI-powered medical chatbots "have an elephant in the room," said James Davenport, a computer scientist at the University of Bath, UK, who was not involved in the study.

There is a big difference between answering "medical questions" and actual medicine, including diagnosing and treating real health problems, he said.

AI expert Anthony Cohn of the University of Leeds, UK, said hallucinations, due to their statistical nature, are always likely to pose problems for such large language models.

As such, these models "should always be viewed as assistants rather than final decision makers," Cohn said. Singhal said that in the future, Med-PaLM could be used to help doctors offer alternatives that were otherwise not considered. The Wall Street Journal reported earlier this week that Med-PaLM 2 has been undergoing clinical trials at the Mayo Clinic, a prestigious US research hospital, since April.

Singhal said he could not discuss specific partnerships.

But he stressed that not all tests are "clinical or patient-oriented and can harm patients."

Instead, it will focus on "more administrative tasks that can be automated relatively easily and with little effort," he added.