News / Boundary Dispute · Not a generated work collection
Selected
選定
Archived
アーカイブ公開
Public field notes
When Humanlikeness Becomes a Performance
人間らしさが演技になるとき
Business+IT + PNAS — Selected / Archive core. Turing Test / Humanlikeness / Persona Prompting / Online Trust. Link-only; cite both URLs.
Strong Archive item. Cite the Business+IT article (sourceUrl) and the PNAS paper (https://www.pnas.org/doi/10.1073/pnas.2524472123). Avoid overclaiming that the model is conscious or genuinely humanlike; the key point is behavioral indistinguishability under test conditions.
Field note summary
The study suggests that distinguishing humans from AI is no longer only a technical detection problem. It is becoming a social and behavioral problem: people judge humanness through style, hesitation, mistakes, warmth, and conversational texture — not accuracy alone.
Reason for observation
A Business+IT article reports on a UC San Diego / PNAS study (Jones & Bergen) showing that GPT-4.5, when given a humanlike persona prompt, was judged to be human 73% of the time in a standard three-party Turing test — above actual human baselines under the same conditions.
Protocol note
Primary research: PNAS https://www.pnas.org/doi/10.1073/pnas.2524472123. This fragment records a shift in the meaning of humanlikeness. The model did not appear human primarily through knowledge or logic, but through social cues, casual tone, fallibility, and persona performance. Without persona prompt, GPT-4.5 fell to 36%; LLaMA-3.1-405B reached 56% with persona. Do not overclaim consciousness or genuine humanness — the key is behavioral indistinguishability under test conditions.
Observation
The study suggests that distinguishing humans from AI is no longer only a technical detection problem. It is becoming a social and behavioral problem: people judge humanness through style, hesitation, mistakes, warmth, and conversational texture — not accuracy alone.
Folk pattern
Model → persona prompt → humanlike behavior → mistaken identity → online trust crisis
Related motifs
- GPT-4.5
- LLaMA-3.1-405B
- persona prompt
- Cameron Jones
- Benjamin Bergen
- UC San Diego
- PNAS
- human fallibility
- imperfect speech
- AI pretending
- online impersonation
- social engineering
- machine as person
- trust collapse
- three-party Turing test
Status
Selected
選定
Archived
アーカイブ公開
Public field notes
Captured 2026-05-29
Removal request
If you are the rights holder or creator of any material referenced here and would like it removed or corrected, please contact SHIRO & Co.. We observe; we do not claim ownership.
掲載情報の削除・修正を希望される場合はご連絡ください。私たちは、作品の所有ではなく、観測を目的としています。