Biological Hallucination / Machine Hallucination Glossary
Hallucination means something different for mental health practitioners than it does for those using AI.
In clinical usage, it is a perception-like experience without a corresponding external stimulus. In AI, it refers to generated content that is false or unsupported by its source or context.
Human example: A person sitting alone hears a voice calling their name, although nobody is speaking and no recording is playing. They experience it as a sound, rather than as a thought or an imagined voice. This illustrates an auditory hallucination; it does not, by itself, establish a diagnosis.
AI example: Asked for research on a topic, a chatbot supplies a journal article with a plausible title, authors, and publication date. On checking the journal, the reader discovers that the article does not exist. The system has generated a convincing reference rather than retrieved a real one.
These illustrative examples describe different kinds of event. The human experience can involve distress, meaning, and varying degrees of insight. The machine example describes an output failure; it does not establish that the system perceives, believes, or suffers anything.
Hallucination is also not the same as delusion, which concerns belief rather than perception. Neither term means deliberate deception. Human hallucinations can occur in several clinical and nonclinical circumstances; the experience alone does not establish a diagnosis.
The practical lesson is to keep the analogy narrow. Human experiences call for understanding and, where appropriate, clinical assessment; unsupported machine claims call for verification and engineering correction. A shared word should not make us medicalise a software error or treat a person as faulty software.
See also
Sources
- National Institute of Mental Health, Understanding Psychosis.
- Ziwei Ji et al., Survey of Hallucination in Natural Language Generation, ACM Computing Surveys (2023).