The Reverse Turing Test
12 May 2026
What Gets Lost When AI Makes Words Cleaner
AI-generated summary
Foster-Fletcher argues that when an LLM smooths writing it removes information. He opens with a fifteenth-century Novgorod bell whose cracked tone warned of danger; recast into perfect notes, its alarm was mistaken for a call to prayer. Cleaned text strips the metadata readers use to sense certainty, hesitation or pressure. He names the reverse Turing test: the point where humans become indistinguishable from the machine. Editing helps routine coordination and non-native writers, but for feedback, crisis and board messaging it makes text read better while telling the reader less.
When an LLM 'improves' writing, it removes information.
In fifteenth-century Novgorod, the city's signal bell was scarred by fires and invasions. So much so that its cracked tone carried meaning for those that had learned to read the bell and its implications.
The story goes that a new prince, embarrassed by the condition of the bell, had it recast into acoustic perfection. When next invaders came, the bellringer pulled the rope in terror, and the bell responded with serene, flawless notes. The city’s inhabitants mistook the alarm for a call to prayer. By the time they understood, the city had fallen.
Flaws are signal.
The Turing Test asked whether a machine could pass for a person. The reverse turing test would therefore be when the humans are unrecognisable from the machine. When an LLM reduces the friction in writing with its typical smoothing of the text, the reader can no longer tell whether the writer was certain or hesitant, comfortable or under pressure, still working it out. Bertrand Russell had the shape of this: the stupid are cocksure while the intelligent are full of doubt. AI editing takes the doubt out.
Reed Hastings built part of his Netflix leadership on rough, sometimes incomplete messages. The roughness forced teams to interpret intent rather than receive instruction. Had AI smoothed those messages, the friction that made people pay attention would have gone. And Hastings is just the visible case. Most organisational communication is read for the same metadata, less consciously. A long meandering paragraph signals that someone is still working through their thinking. A short abrupt note could carries urgency or irritation or an unusual word choice might flag discomfort. As writers, we rarely realise how much of the work the grooves were doing until they have been planed off.
AI editing has uses. It lowers the unfair barrier facing non-native English writers in English-dominant organisations. It handles routine coordination where signal is not the point. It supports compliance work where consistency of phrasing matters. But editing stops helping at a particular point. Performance feedback, crisis messaging, board updates, discussions about strategic uncertainty; these depend on the signals that editing removes. Without them the message reads better and tells the reader less.
Organisations now face a different problem from the one Turing posed. The people in them are starting to sound like the machine.
And when the bell rings, nothing tells us to act.
The analysis
- What information is lost when AI smooths or cleans human writing?
- What is the reverse Turing test?
- Why do flaws and roughness in writing carry meaning for readers?
- Which kinds of organisational communication are most damaged by AI editing?
- Where is AI editing genuinely useful, and where does it stop helping?
- How did Reed Hastings use rough messaging as a leadership tool at Netflix?
- What happens when everyone in an organisation starts to sound like the machine?