AI research by Google researchers shows an amazing increase in the accuracy of Med-PaLM 2 answers
Artificial intelligence has penetrated every field of technology today. Among them, one area that is particularly adaptable to emerging technologies is the medical sector. Because it has to deal with extremely sensitive areas that can lead to life-or-death situations, the medical field has been apprehensive about implementing new medical technology tools into general practice. However, AI has been knocking on doors for a while, and if a new study conducted by Google researchers is to be believed, Google’s in-house Med-PaLM 2 is scoring really well in answering. Medical Questionnaire (MedQA) and is in a prime position to enable medical professionals to provide faster medical care to patients.
In fact, Med-PaLM 2, is a medical large language model (LLM) being trained to synthesize information from medical images. In fact, not only Google, other players are also working on general AI in healthcare and among them is OpenAI’s ChatGPT led by Sam Altman. And the competition is fierce. A study published in JAMA Internal Medicine states that ChatGPT gives higher quality answers to questions than written answers from actual practitioners.
Now, on Wednesday, Google Health UK Study leader Alan Karthikesalingam posted on Twitter, highlighting the achievement. He speak, “Happy to share #MedPaLM2 – our team’s Med-PaLM evolution. A new state of the art to answer medical questions! Med-PaLM 2 scores 86.5% on MedQA-USMLE, surpassing Med-PaLM scores >19% and 81.8% on PubMedQA.”
It should be noted that the MedQA-USMLE dataset is a multiple choice questionnaire based on the US Medical Licensing Exam. So getting a good score basically means that AI could theoretically be certified to practice medicine in the US. PubMedQA is also a similar dataset. In the data set test, Med-PaLM 2 scored as high as 86.5% according to the study conducted by the team. The research is currently in preprint on arXiv. It should also be noted that this study has not been peer-reviewed or published in a journal to date.
Google’s Artificial Intelligence Scores High on Medical License Exam
Karthikesalingam stated in a series of tweets about the level of scrutiny being carried out to ensure that the results of the test are not by chance or misrepresent the capabilities of the AI platform. “We believe in rigorous, careful evaluation,” he said. Doctors even prefer #MedPaLM2 long answers to answers from other real doctors along 8/9 quality axes including medical accuracy (medical consensus). economics) and arguments, with less potential for harm.
He added: “To highlight the real-world importance of nuance assessment, we introduce a new dataset of ‘opposite’ questions specifically designed to probe the weaknesses of LLM, including #HealthEquity.
It’s currently unclear how much of an impact this new AI technology will have on the medical field, but Google seems optimistic about the results. However, research is just the beginning. For this technology to be adopted and used in real-life situations, it will have to go through much closer scrutiny to understand whether AI can help patients with healthcare consistently and reliably. trust or not.
Thus, even the director of Google Sundar Pichaiwhile speaking at a conference held recently Google I/O highlighted how the company is working on this technology carefully and responsibly to make sure it doesn’t go wrong.
