The reading room · Digna Legi
We Can't Let AI Writing Take Over the Internet
/100
80–100: high value. 70–79: worth the time. Below 70: below the usual publication threshold.
Evidence-reviewed score based on available publisher text. Evidence is sampled with gaps, so the full interview balance, redundancy, and evidentiary rigor cannot be completely assessed.
Scores reflect one reader’s profile, not an objective quality rating. Best is a separate personal selection.
How scoring works →This brief · about 3 min with detail
Why read this
AI detection becomes a fight over authorship, incentives, and whether writing still proves that a person thought something through.
AI brief · Checked against source text
The main idea
Derek Thompson and Pangram’s Max Spero argue that AI-written text matters less as prose quality than as a break between composition, knowledge, and trust. Pangram treats detection as pattern classification: it compares human and AI text distributions, while Spero argues that cheap text plus platform incentives pushes everyone, not only bad actors, toward volume and automation.
Some background helpful. Basic familiarity with generative AI, social platforms, and machine-learning classifiers.
Go a little deeper
Cheap text changes incentives
Spero’s sharper point is that AI slop is not only fraud or propaganda. Once written output becomes nearly free, ordinary career and platform incentives reward more posting, more search optimization, and more professional visibility. LinkedIn is presented as the clean example: if frequent posting seems useful for advancement, people rationally choose the easiest way to post more.
Detection looks for distributions, not catchphrases
The interview distinguishes folk detection from machine detection. Early AI tells might have been odd words or punctuation habits, but Spero says those have shifted upward into sentence, paragraph, and document-level patterns. Pangram’s claimed advantage is statistical memory: it sees many small phrasing choices accumulate into a signal humans cannot reliably perceive.
AI’s polish can flatten long-form meaning
Thompson and Spero separate short-passage impressiveness from sustained literary control. AI can imitate a style locally and write concise, high-impact summaries, but the interview argues it struggles with long-context change: the gradual development of character, scene, emotion, and delayed significance. That distinction explains why reader tests on short excerpts do not settle the question of novels or long essays.
Policing authenticity creates its own social cost
The interview does not treat detection as an uncomplicated good. Thompson worries that public Pangram screenshots can become witch-hunt material, while Spero accepts that false positives happen even at a claimed low rate. The tension is real: without enforcement, institutions may automate more writing; with enforcement, people can be wrongly shamed or reduced to detector scores.
A case from the article
The Guardian dispute
A sports column was publicly suspected of being AI-written, and the Guardian said the writer had always sounded that way. Spero responded by running hundreds of the journalist’s past articles through Pangram and publishing a time series that, in his view, showed a shift toward AI reliance. The episode illustrates both Pangram’s evidentiary appeal and the uncomfortable move from detection to public policing.
How the case is made
The case is made through an interview, reported examples from publishing and platforms, Pangram’s own studies, and cited third-party testing.
Where the idea has limits
The strongest empirical claims about Pangram’s accuracy come mainly from Spero’s account of studies and company measurements inside an interview, so the brief supports the argument’s logic more than independently auditing the tool.
A question to take away · from Digna Legi
When text is cheap and plausible, what kinds of writing should still require proof of human authorship?
What the original adds
The source adds details on Pangram’s classifier training, claimed false-positive rates, Substack and browser-extension integrations, and Spero’s view of watermarking versus independent detection.
About this brief
AI-written, then separately checked for source support, useful detail and clarity. The author’s claims and our editorial question are kept separate. The original remains the author’s work. How we select and summarise →
How was this brief?
Rate this summary, separately from the author’s article.
Optional. Saved in this browser; shared only if you allow analytics.
How was the original article?
Rate the author’s original after reading it.
Optional. Saved in this browser; shared only if you allow analytics.
Digna legi. Worth reading.