Publishing-brain limits people’s understanding of AI usefulness

Source: Mike Caulfield Substack Author: Mike Caulfield Original source: https://mikecaulfield.substack.com/p/publishing-brain-limits-peoples-understanding Published: 2026-05-14 Source type: essay

Private backup: the full article text is archived in the private repository at archives/articles/mikecaulfield-substack-com-publishing-brain-limits-peoples-understanding.source.md. It is not published on the public Quartz site.

Summary

Mike Caulfield argues that critics misunderstand AI usefulness when they judge every AI-assisted output against publication-level accuracy. Using an example where Claude first offered a plausible but untested claim about 1990s “high-concept” comedies, then helped investigate the claim through Wikidata, Caulfield shows AI’s value as part of a provisional, iterative inquiry process.

His key point is that most human argument-making is not publication: people constantly use partial evidence to test assumptions, form judgments, make decisions, and refine questions. In that context, imperfect first passes from AI can be useful because they provoke classification questions, evidence checks, and deeper understanding. The educational implication is that teachers should not merely repeat “AI is a bad answer box,” because that accepts the corporate frame. Instead, they should model better AI use as evidence-seeking, iterative, source-aware inquiry.

Pull quotes

Publication is the exception

“Publication — and the accuracy people expect from it — is both the bedrock of a literate society and also an outlier to most human experience where we are looking for just a little bit of data to test our assumptions, make a point, inform a decision, or explain how we see the world.”

The better question

“The question is not whether the answers you get are perfect (there are no perfect answers to most things anyway). The question is whether a person who learns to use these tools will have more evidence-informed opinions on things than someone who does not.”

Bad first passes can be useful

“It’s not necessary to have a perfect pass to get that process started. You just have to have a pass. In fact, an initial bad categorization (and your reaction to it) is most likely to produce your most valuable theoretical insights.”

Don’t accept the answer-box frame

“To that I’d just say yelling over and over again ‘It’s a bad answer box!’ simply accepts the corporate frame that it is meant to be an answer box, and our argument is over how good at that function it is.”

Durability note

The specific tool example may date as LLM-tool integrations change, but the core distinction between publication-level accuracy and everyday evidence-informed inquiry is likely to remain durable. The article is especially useful as a conceptual bridge between SIFT-style verification and practical AI use.

Big ideas

Claims

Key evidence and examples

  • Caulfield starts from a practical media-research task for a podcast episode about The Frighteners.
  • Claude initially offers a plausible claim that the film fit a 1990s surge of “high-concept” comedies.
  • Caulfield then asks for evidence, specifically a Wikidata-based look at comedies from three major studios between 1986 and 2015.
  • The evidence complicates the initial claim: 1996 was not especially dominated by high-concept comedies, while a later decline in high-concept comedy appears more interesting.
  • The value of the AI-assisted process is not that the first answer is publishable, but that it gets the user to better questions about classification, evidence, scope, and uncertainty.
  • Caulfield argues that imperfect first passes can be productive because they expose definitional problems and create opportunities to refine the inquiry.

Education relevance

This is highly relevant to AI literacy, media literacy, research instruction, and classroom modeling of AI use. It reframes AI from an “answer box” to a tool for provisional inquiry: students and teachers can use AI to generate first passes, check evidence, revise definitions, and develop more evidence-informed judgments without pretending that every intermediate result meets publication standards.

My notes