SYNTHESIS NOTE
Psychology, Society, and Alignment Language, Text, and Discourse

Can humans detect AI by passively reading its text?

When people read AI-generated transcripts without the ability to ask follow-up questions, can they tell it apart from human writing? This matters because most real-world AI encounters are passive.

Synthesis note · 2026-02-23 · sourced from Social Theory Society
What kind of thing is an LLM really? Where exactly do LLMs break down with language structure?

The displaced Turing test introduces a human judge who reads a transcript of a Turing test conducted by a different interrogator. The inverted test places an AI in the interrogator role. Both variations reveal that removing interactivity collapses detection capability.

Key finding: both AI and displaced human judges were less accurate than interactive interrogators, with below-chance accuracy overall. The interactive interrogator's advantage is the ability to adapt questions adversarially in real time — pursuing fruitful lines of questioning and probing inconsistencies as they emerge. Passive consumption of the same conversation loses this adaptive capacity entirely.

This has direct ecological validity implications. Most real-world AI content consumption is passive — reading chatbot outputs, encountering AI text in search results, seeing AI-generated social media posts. If interactive interrogation only marginally succeeds at detection (GPT-4 fools interrogators 54% of the time), passive consumption is far worse.

The inverted test (AI as interrogator) tests what Watt (1996) called "naïve psychology" — whether an AI system has the innate tendency to recognize intelligence similar to its own. The results suggest AI judges lack this capacity: they cannot reliably discriminate between humans and machines that fooled human interrogators.

Since Can humans detect AI text if machines can measure it?, there is a paradox: statistical analysis reveals clear AI signatures, but neither interactive nor passive human judgment reliably detects them. The detection problem is not that AI text has no distinguishing features — it's that human perception cannot access them.

Since How can AI text disrupt structure yet feel normal to readers?, the displaced Turing test confirms the mechanism: when you remove provenance information (who generated this?), the text enters the same interpretive circuits regardless of origin.

Since Does AI text affect readers the same way human text does?, passive consumption is where this equivalence matters most. If observers cannot tell, then AI text is functionally indistinguishable from human text in the settings where most content consumption actually occurs.

Inquiring lines that use this note as a source 10

This note is a source for these synthesized inquiries. Follow a line forward into its question, or open it to trace back to all of its sources.

Related concepts in this collection 4

This note in its neighbourhood — explore the map, then jump to a related concept in the list below.

Concept map
14 direct connections · 118 in 2-hop network ·medium cluster Open in graph ↗

Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph

your link semantically near linked from elsewhere

Related papers in this collection 8

Papers most semantically related to this note, ranked by cosine similarity in the embedding space.

Original note title

passive observers cannot distinguish AI from humans even when interactive interrogators partially can