Mayo Clinic has implemented a novel technique called reverse RAG to combat AI hallucinations in healthcare data.
By employing backwards RAG, Mayo ensures that every data point links back to its original source content, reducing data-retrieval-based hallucinations.
This technique has enabled Mayo to utilize AI more extensively in its clinical practice.
Mayo initially applied AI in discharge summaries, where models excelled in extraction and summarization tasks.
By avoiding hallucinations through careful building and verification, Mayo fine-tuned the AI models.
Backwards RAG addresses limitations of RAG, such as irrelevant data retrieval, by using the CURE algorithm to double-check data extraction.
The CURE technique's effectiveness also extends to synthesizing new patient records, saving practitioners significant time.
Mayo sees potential in expanding AI capabilities to reduce administrative burden and assist physicians.
Collaborations with companies like Cerebras Systems and Microsoft aim to enhance genomic modeling and imaging analysis for improved patient care.
Mayo envisions using AI to personalize medicine by leveraging genomics and other 'omic' areas like proteomics to tailor treatments based on individual patient profiles.
While AI holds promise in transforming patient care, rigorous validation and testing are essential in clinical settings for successful implementation.