Definition of AI hallucination as false fact

Summary

In AI, a hallucination (or artificial hallucination) is an AI-generated response that includes false or misleading information while presenting it as fact. The term is loosely compared to human hallucinations, which involve false perceptions, but in AI it specifically refers to fabricated or incorrect statements produced by the system. Large language model (LLM) chatbots can produce plausible-sounding random falsehoods, including fabricated citations, making detection and mitigation difficult—especially in high-stakes uses like medical diagnostics, chip design, and logistics. The concept has evolved over time: earlier uses in computer vision referred to “hallucination” as adding detail to improve images, and later it became associated with factually incorrect outputs in tasks such as machine translation and object detection. Although the metaphor is widely used, some researchers criticize “AI hallucination” for anthropomorphizing systems and for being vague, arguing that the models do not “understand” meaning and instead generate outputs via statistical pattern completion. Alternative terms in the literature include confabulation, fabrication, and factual error.

Key Points

• 1

  AI hallucination is false or misleading information generated by an AI system and presented as if it were true.

• 2

  LLM chatbots may embed plausible-sounding fabrications (e.g., incorrect or fabricated citations), and reliably detecting/mitigating these errors remains challenging.

• 3

  The term is debated: some researchers criticize it as anthropomorphic and propose alternative terminology such as confabulation or fabrication.