Does ChatGPT Hallucinate Like Humans? Scientists Say the…

Does ChatGPT Hallucinate Like Humans? Scientists Say the Answer Is More Complicated Published: June 2026 | Category: Artificial Intelligence, Mental Health, Digital Medicine | TALMedora Editorial Desk Artificial intelligence can write essays, answer medical questions, summarize research papers, and even hold surprisingly human-like conversations. But it can also be confidently wrong. Anyone who has used an AI chatbot has likely encountered an answer that sounds convincing but turns out to be completely inaccurate. These mistakes are often called "AI hallucinations." But according to researchers, that label may not tell the whole story. A new analysis suggests that some AI errors may actually resemble a different phenomenon seen in human psychology: confabulation. The comparison is helping scientists better understand both artificial intelligence and the human brain. Large language models such as ChatGPT generate responses by predicting which words are most likely to come next based on patterns learned during training. Most of the time, this works remarkably well. However, problems arise when the system lacks sufficient information. Instead of admitting uncertainty, AI can sometimes generate answers that sound believable but are factually incorrect. Examples include: • Invented references • Incorrect medical advice • Fictional historical events • Made-up statistics • Non-existent scientific studies The response often appears coherent and confident despite being wrong. In medicine, confabulation refers to the unintentional creation of false memories or explanations. People experiencing confabulation are not deliberately lying. Instead, the brain fills gaps in memory with information that seems plausible. This can occur in conditions such as: