There are generative AI models. are increasingly being introduced into health care settings. – Premature in some cases, perhaps. Early adopters believe they will unlock increased efficiency by uncovering insights that would otherwise be missed. Critics, meanwhile, point out that these models have flaws and biases that can contribute to poor health outcomes.

But is there a quantitative way to know how helpful, or harmful, a model might be when tasked with things like summarizing patient records or answering health questions?

Hugging Face, an AI startup, proposes an all-in-one solution. A newly released benchmark test called Open Medical-LLM.. Created in partnership with the non-profit Open Life Science AI and researchers from the University of Edinburgh's Natural Language Processing Group, Open Medical-LLM aims to standardize the performance of generative AI models on a range of medically relevant tasks.

There is no Open Medical-LLM. From the beginning Benchmarks, per se, are rather tailoring existing test sets – MedQA, PubMedQA, MedMCQA and the like – designed to test models of general medical knowledge and related fields, such as anatomy, pharmacology, genetics and clinical practice. Is. The benchmark includes multiple-choice and open-ended questions that require clinical reasoning and understanding, drawing from material including question banks from American and Indian medical licensing exams and college biology tests.

“(Open Medicine-LLM) enables researchers and practitioners to identify the strengths and weaknesses of different approaches, furthering the field and ultimately contributing to better patient care and outcomes,” Hugging said. Fee wrote in a blog post.

Image credit: A huggable face

Hugging Face is positioning the benchmark as a “robust evaluation” of generative AI models connected to healthcare. But some medical experts on social media cautioned against putting too much stock in Open Medical-LLM, lest it lead to unreported placements.

At X, Liam McCoy, a resident physician in neurology at the University of Alberta, pointed out the difference between the “conflicting environment” of answering medical questions. real A medical practice can be quite large.

Face-hugging research scientist Clementine Fourier, who co-authored the blog post, agreed.

“These leaderboards should only be used as a first guess in which to explore a given use case (productivity AI model), but then always tested to test the model's limitations and applicability in real-world situations.” A deeper phase is required.” Fourier replied. On X. “Patients shouldn't be using the medical (model) themselves at all, but instead should be trained to be assistive devices for MDs.”

This brings to mind Google's experience when it tried to bring an AI screening tool for diabetic retinopathy to the healthcare system in Thailand.

Created by Google A deep learning system that scans images of eyes.Look for evidence of retinopathy, a leading cause of vision loss. But despite the high theoretical accuracy, The device proved impractical in real-world testing.Frustrating both patients and nurses with inconsistent results and a general lack of coherence with ground-based practices.

It is telling that the US Food and Drug Administration has approved 139 AI-related medical devices to date. No one uses generative AI.. It is extraordinarily difficult to assess how the performance of a creative AI tool in the laboratory will translate to hospitals and outpatient clinics, and perhaps more importantly, how the results may trend over time.

This does not mean that Open Medical-LLM is not useful or informative. The results leaderboard, if nothing else, serves as a reminder of how damaged Models answer basic health questions. But the Open Medical-LLM, and any other benchmark for that matter, is a substitute for carefully thought-out real-world testing.