Stop Calling AI Errors "Hallucinations." They Are Product Failures

The term "hallucination" used to describe AI errors misrepresents the issue as a charming psychological quirk. When a paid professional confidently invents false information, they would be quickly fired. The dictionary definition of hallucination involves a false sensory experience, such as seeing or hearing something that has no external physical source.

However, AI does not perceive anything in the human sense. When AI presents invented, unsupported, or outdated information as fact, it is a product failure. Describing it as a hallucination anthropomorphizes the machine, softens accountability, and leads users to accept false answers as an unavoidable trait. This is unacceptable, especially from services users pay for.

Google is a prime example, as it attaches uncertainty to a search engine built on verifying reality. When an AI-powered search product is asked a simple question about a current event, it can return a confident, false answer without any qualifiers or admission of unreliability. It presents fabricated information as fact, which is a significant failure.

Unlike a human hallucination, which involves a sensory experience with no external stimulus, AI processes inputs and generates outputs without any sensory experience or confusion about its inputs. AI does not wake up distressed or make errors due to hallucinations; it simply fails to provide accurate information. Describing AI errors as hallucinations is a cop-out that smacks of misplaced pity and forgiveness.

If a human did this at work, it would be called incompetence or fabrication. Calling it a hallucination is a way to avoid accountability and protect the company from criticism.