insiderblm
According to a peer-reviewed study released on Wednesday, Google’s artificial intelligence-powered medical chatbot passed a challenging US medical licencing exam, but its responses still fell short of those provided by actual doctors.
The launch of ChatGPT, whose creator OpenAI is supported by Google’s competitor Microsoft, last year marked the beginning of a competition between computer giants in the developing field of AI.
While much has been said about the potential benefits and risks of artificial intelligence, one area where the technology has already made real strides is in the field of health. Algorithms are already capable of reading some medical scans just as well as people.
In a preprint paper published in December, Google initially introduced Med-PaLM, its AI tool for addressing medical queries. It hasn’t been made available to the general public like ChatGPT has.
The US tech company claims that Med-PaLM is the first large language model to pass the US Medical Licencing Examination (USMLE), an AI approach trained on enormous amounts of human-produced text.
The exam, which is taken in the US by medical students and future doctors, has a passing score of roughly 60%.
According to a research published in February, ChatGPT had received passing or nearly passing scores.
Google researchers said that Med-PaLM had a score of 67.6 percent on multiple-choice questions in the form of the USMLE in a peer-reviewed study that was published in the journal Nature on Wednesday.
The study concluded that while Med-PaLM still performs below doctors, its performance is encouraging.
Google claimed it has created a new evaluation standard in order to recognise and reduce “hallucinations,” which are instances in which AI models provide incorrect information.
The researchers utilised the benchmark to test a newer version of their model, with “super exciting” findings, according to the primary author of the new paper and Google researcher Karan Singhal.
