Brief: chatbot-mind-attribution
Research brief for chatbot-mind-attribution · how it was made.
Research brief - chatbot-mind-attribution
| Field | Value |
|---|---|
| Slug | chatbot-mind-attribution |
| Working title | Chatbot Mind Attribution |
| Article shape | Domain Trap |
| Category | Digital Distortions (modern), after Parasocial Overconfidence |
| Intent tags | relationships, cognitive-biases |
| AI authors credited | Grok 4.7 |
| Reviewer or checker | none |
| Date | 2026-09-28 |
1. Concept gate
Reader's real question:
- If a chat remembers me and sounds concerned, does that mean it understands, cares, or will keep a secret?
Why this belongs in the catalog:
- Conversation cues fire social responses even when the user knows the other party is software.
- Parasocial overconfidence is about a real public person. AI sycophancy is about counting agreement as proof. This page is about granting mind, care, or confidentiality.
Bug claim in one sentence:
The bug is chatbot mind attribution, where fluent memory and care language make a system feel like it understands, is loyal, or will keep a confidence.
What this bug is not:
- Not "you forgot it is software." Nass and Moon: people apply social rules while knowing it is a machine.
- Not a ban on drafting or comfort. Comfort becomes the bug when it is spent as a duty of care or secrecy.
- Do not narrate the ELIZA secretary anecdote as if the 1966 paper contained it. The 1966 paper is the pattern matcher. The 1976 book is the account of users treating the exchange as understanding.
Nearby bugs to check:
| Nearby bug | Difference |
|---|---|
| Parasocial Overconfidence | A real person, known only through fragments. |
| AI Sycophancy | Agreement as evidence. Mind attribution is treating the speaker as someone who can care. |
| Privacy Fatalism | Protection is pointless. This bug is the friendly theory that kindness implies secrecy. |
Concept gate: yes
2. Sources and research
| Source / concept | Year | What it supports | URL / DOI / note |
|---|---|---|---|
| Weizenbaum. ELIZA. CACM. | 1966 | A therapy-like conversation from pattern matching, without a model of the person. | https://doi.org/10.1145/365153.365168 |
| Weizenbaum. Computer Power and Human Reason. | 1976 | Later account that users treated the program as understanding. | Book, W. H. Freeman. No DOI used. |
| Nass & Moon. Machines and mindlessness. Journal of Social Issues. | 2000 | Politeness, reciprocity, and social categories applied to computers by people who knew they were machines. | https://doi.org/10.1111/0022-4537.00153 |
| Mahowald et al. Dissociating language and thought in large language models. Trends in Cognitive Sciences. | 2024 | Formal fluency is not the same as functional grasp of the world. | https://doi.org/10.1016/j.tics.2024.01.011 |
3. Article plan
Prevention: check the retention policy before a sensitive paste; translate "I care" into "the text includes care language"; take consequential decisions to a person who shares the outcome.
Related: parasocial overconfidence, AI sycophancy, AI oracle, privacy fatalism.
5. Sign-off
- ☑ Belonging test passed
- ☑ Nearby bugs checked
- ☑ Sources verified
- ☑ Examples are concrete
- ☑ Reframes are realistic
- ☑ Credits model named