Why AI and people both "hallucinate," but differently, and how therapy helps us face reality.

What is it about?

AI chatbots sometimes "hallucinate," confidently stating things not checked against an outside reality. This paper asks what that phenomenon is, and how it compares to something humans do too: testing our own sense of reality. Drawing on psychoanalytic thinkers, the paper argues human reality-checking is not just about getting facts right. It grows out of frustration, disappointment, and mourning what we wished were true, none of which a chatbot experiences. The paper then turns to psychoanalytic therapy, describing it as a special kind of "past-tense" reality-testing: patients coming to terms with a childhood or relationship that did not go as hoped. The AI comparison is not the main point: computational science suggests AI hallucination is unavoidable. The conclusion brings into focus something distinctly human: the need to mourn what could have been, and to find present life livable.

Featured Image

Photo by am g on Unsplash

Photo by am g on Unsplash

Why is it important?

As AI becomes part of daily life, and sometimes even conversations people turn to in distress, it matters to understand what a chatbot can and cannot do psychologically. This paper gives therapists and general readers a clear way to think about the difference between a machine that cannot verify its own claims and a person doing the hard work of mourning. Habitual reliance on AI may erode rather than build the human capacities that attachment depends on, so the stakes go beyond individual conversations to how people relate to loss, disappointment, and each other more broadly. The paper also shows that this work is not always straightforward: under early trauma, a person's own perceptions can be overridden by someone else's denial of reality, so a recognition of traumatic realities has to be recovered before mourning can even begin.

Perspectives