Google's Med-Gemini AI hallucinates nonexistent brain structure
Google's Med-Gemini healthcare AI model generated a reference to 'basilar ganglia,' a nonexistent brain structure, in a 2024 research paper and blog post. After a neurologist flagged the error, Google edited the blog post quietly but left the research paper unchanged, characterizing it as a typo rather than a hallucination.
What happened
Google's Med-Gemini healthcare AI model generated a reference to 'basilar ganglia,' a nonexistent part of the brain, in a 2024 research paper introducing the model. The term appears to conflate two distinct anatomical structures: the basal ganglia and the basilar artery, which serve different functions and require different treatments if affected by stroke or abnormality.
The error appeared in a section on head CT scans in the research paper and an associated blog post. Neurologist Bryan Moore flagged the mistake to Google after noticing it. Google quietly edited the blog post to fix the error with no public acknowledgement, but the research paper remains unchanged.
Reported impact
Medical professionals have raised concerns about the error demonstrating the risks and limitations of healthcare AI, particularly the danger that doctors using such models might not catch hallucinatory errors and could act on incorrect information without verification.
Index analysis
Google characterized the error as a simple misspelling of 'basal ganglia.' The company later added a caption to the blog stating that 'basilar' is a common mis-transcription of 'basal' that Med-Gemini learned from training data. Medical experts disputed this characterization, arguing the error represents a hallucination rather than a simple typo.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.