Omission

An AI error where the model fails to mention important information that is present in the input. In image descriptions for BLV users, omission can include failing to mention warning labels on medication, not describing important text in a document, skipping relevant objects in a scene, or overlooking details that are critical for the user's task. Omission is difficult to detect without visual access to the original image because users cannot know what they are not being told. Research shows that with single descriptions, lack of detail is the most common indicator of unreliability (54% of flagged claims), but users often only recognize omissions when comparing multiple descriptions that include different details.