Google’s AI-driven medical chatbot, named Med-PaLM, has successfully cleared the rigorous US medical licensing examination, although its responses still lag behind those of human physicians, as per a recently published peer-reviewed study.
Advancements in AI and Healthcare
Despite ongoing debates surrounding the benefits and risks of AI, the healthcare sector has witnessed significant progress in leveraging the technology. AI algorithms have demonstrated their ability to interpret medical scans as effectively as human professionals.
Med-PaLM’s Success and Improvements
Google unveiled Med-PaLM, its AI solution for medical queries, in a preprint study last year. The AI model, which is not publicly available like its sibling ChatGPT, became the first large language model to clear the US Medical Licensing Examination (USMLE), according to Google.
Reducing “Hallucinations” and Enhanced Performance
Google researchers introduced a new evaluation benchmark to identify and minimize “hallucinations” – instances when AI models generate incorrect information. The team utilized this benchmark to assess a more recent version of Med-PaLM, achieving promising results. A preprint study released in May, although yet to undergo peer review, reported that Med-PaLM 2 scored an impressive 86.5% on the USMLE, marking a nearly 20% improvement over its predecessor.
