AI & Enterprise
Medical AI flaws exposed as models diagnose without images
Artificial intelligence used in healthcare can generate answers as if it has seen diagnostic images even when none are provided, researchers found. A Stanford University team reported similar issues across 12 visual AI models. The effect was pronounced in medical queries, where outputs tended to lean toward severe diagnoses. The researchers said some models could rank top on chest X-ray benchmarks without images. They proposed a filtering method, B-Clean, to remove questions answerable without images.
