TLDR: Google’s healthcare AI, Med-Gemini, erroneously identified a non-existent brain part, ‘basilar ganglia,’ in a 2024 research paper. While Google attributed it to a misspelling, medical experts are alarmed by the potential for AI ‘hallucinations’ to pose significant risks in clinical settings.
Medical professionals are expressing growing apprehension regarding the widespread adoption of generative artificial intelligence tools in healthcare, following a significant error by Google’s Med-Gemini AI model. In a May 2024 research paper introducing Med-Gemini, Google researchers showcased the AI’s ability to analyze brain scans. However, the model reported an ‘old left basilar ganglia infarct,’ referring to a brain structure, ‘basilar ganglia,’ that does not exist in human anatomy.
The critical error was brought to light by board-certified neurologist Dr. Bryan Moore, who flagged the issue to The Verge. It is believed that the AI likely conflated ‘basal ganglia,’ a genuine brain area associated with motor movements, and the ‘basilar artery,’ a major blood vessel.
Google has acknowledged the mistake, attributing it to a ‘simple misspelling’ or ‘common mis-transcription’ of ‘basal ganglia’ that Med-Gemini learned from its training data. The company updated its blog post to reflect the correction, stating that ‘basilar’ is a common mis-transcription of ‘basal’ and that the meaning of the report was unchanged. However, as of the reporting date, the original research paper still contained the erroneous term.
This incident underscores persistent and impactful shortcomings of AI technology, particularly the phenomenon of ‘hallucinations’—a term for AI-generated made-up facts. Experts warn that while this specific faux pas may not have directly endangered patients, such inaccuracies could have devastating consequences in a hospital environment. Dr. Maulin Shah, Chief Medical Information Officer at Providence healthcare system, emphasized the gravity of the error, stating to The Verge, ‘What you’re talking about is super dangerous. Two letters, but it’s a big deal.’
Also Read:
- Landmark Study Reveals Generative AI Chatbots Currently Unreliable for Critical Stroke Care Advice
- AI and Ethics: Reshaping Healthcare and Clinical Trials for a New Era
The case highlights the ongoing debate about the reliability and safety of AI in critical applications like medicine, where even minor errors can have profound implications for patient care and diagnosis.
