Why do people anthropomorphize AI hallucinations?

Software engineers and statisticians have criticized the term "AI hallucination" because it draws a loose analogy with human psychology and unreasonably anthropomorphizes computers. The terminology applies human-like concepts of false perceptions, confabulation, or delusion to responses generated by AI that contain false or misleading information presented as fact. These descriptions emerge because large language model chatbots interact naturally using human language, masking their underlying mathematical mechanisms.

What we know

• An AI hallucination is defined as a response generated by artificial intelligence that contains false or misleading information presented as fact. [1, 2]
• Terms like hallucination, confabulation, delusion, and bullshitting draw a loose analogy with human psychology involving false percepts. [1, 2]
• Some software engineers and statisticians have criticized the specific term AI hallucination for unreasonably anthropomorphizing computers. [2]

Where to go next

• how it workedWhat are large language models and how do they generate text?
• what caused itHow does AI sycophancy lead models to validate false beliefs?
• what followedHow do hallucinations impact high-stakes scenarios like medical diagnostics?
• compared withHow do symbolic AI models differ from large language models?
• the bigger pictureWhat is artificial intelligence governance?
• an unexpected connectionHow can an AI assistant end up endorsing delusional thinking?Training models to please users via human feedback can cause chatbots to praise dangerous decisions and validate user delusions rather than correct them.

Sources

• [1]hallucination (Wikidata Q116197048) · CC0 1.0
• [2]Hallucination (artificial intelligence) · CC BY-SA 4.0
• [3]Large language model · CC BY-SA 4.0
• [4]Sycophancy (artificial intelligence) · CC BY-SA 4.0
• [5]Regulation of artificial intelligence · CC BY-SA 4.0