noun
An AI hallucination occurs when a language model or AI system generates incorrect, fabricated, or nonsensical information while presenting it as factual. This happens because AI models predict the most likely next word or image based on patterns in their training data, not on real-world accuracy.
Hallucinations often arise when the AI is asked questions outside its training scope, lacks sufficient data, or is pushed to generate creative or speculative content. For example, a language model might invent a fictional research paper with fake citations or a fake author.
For users, this means AI outputs must always be fact-checked, especially in professional or safety-critical contexts. In 2023, a lawyer used AI-generated case citations that didn’t exist, leading to a court sanction. AI hallucinations also force developers to refine models with better data filtering and human oversight.
A chatbot claims a celebrity was born in 1995 when the correct year is 1990, despite no evidence in its training data.
In other languages
See also: AI bias · AI training data