The hypothesis
“humans are too lazy for this I think ;)”
— Prof. Langenkamp, in conversation, 12 May 2026 at 07:48 EDT
The Lazy Median Hypothesis is the Dictionary’s prediction that cheap AI generation will split everyday writing between carefully revised human-AI work and abundant unchecked output, thinning the middle between them.
The wink in the original line stays. It marks the claim as provocative and provisional rather than as a moral verdict on most people.
What “lazy” means
“Lazy” describes an incentive, not a character defect. Much everyday writing is instrumental: a routine email, announcement, application, toast, or homework response exists because someone needs words for a task. When a plausible draft costs almost nothing, many people will rationally accept it without investing in revision.
That shortcut existed before generative AI in greeting cards, form letters, and templates. Generative systems extend it across far more categories and can produce a new draft for each occasion. The hypothesis predicts that this abundance will reward two different practices.
At one end, people who care about the consequences or voice of a text will use models cooperatively. They will check facts, restore their own register, and revise the draft as work rather than treat generation as completion.
At the other, people who need only a prose-shaped artifact will ship the first adequate output. Much of it may be grammatically competent. Its defining feature is not bad grammar but the absence of responsible attention.
The middle may thin because AI can imitate the surface of competent professional prose without reproducing the editorial labour that once produced it. This is a hypothesis about behaviour under new incentives. It is not yet an empirical finding that writing quality follows a bimodal distribution.
Why the printing-press comparison is limited
The printing press sharply reduced the cost of reproducing text. Generative AI reduces the cost of producing a new text. That contrast is useful, but it does not prove that the present divide will be the largest since the printing press, as the former entry claimed. Historical writing markets were varied, and measuring “care” across centuries would be heroic even by Dictionary standards.
The narrower claim survives: generative systems remove some of the human drafting and checking that templates, clerks, editors, and professional writers previously supplied. Whether new forms of review replace that labour is the question.
What would disprove it
The hypothesis would weaken if either of two things happened:
- first-pass generated text became reliably equivalent to carefully revised work under sustained reading and factual inspection; or
- ordinary users began checking and revising generated output often enough that unchecked generation was no longer the default shortcut.
Evidence could also show a broad continuum rather than two clusters. The word hypothesis is load-bearing: the Dictionary is naming a pattern to watch, not reporting a settled distribution.
What the Dictionary recommends
For consequential writing, treat model output as a draft whose author is not available to defend it. Check the claims, decide what the sentences mean, and restore a voice you are willing to own.
Schools, magazines, professional offices, and peer-review systems still matter because they teach or require that work. Their task is not to prohibit AI prose. It is to preserve revision, responsibility, and attention when generation becomes cheap.
See also
- AI Writing — the parent hub of this cluster
- Zombie Internet — abundant prose without a responsible reader or writer
- Earned Parallelism — one editorial check
- The Olang’ Trap — the structural bias of the AI-detection reflex
- The Sinceerly Stack — the recursive detection-and-evasion layer
- Mediation (a la Gibson)
- English Major
- The Sincere Society — the related argument from cheng