KW Norton · with the fraud-screening record

Humans as Creative Directors

AI Agents to Assist Us in Stretching Past Our Limitations

· KW Norton

For those who may think human agency is over, and that humans will no longer be doing the work!

There are going to be more jobs than ever for humans. Between directing AI, education upgrades for humans and AI, and all the scientific, literary, health, community and family goals to pursue, humans are going to be incredibly busy.

Humans in the loop are critical.

The example: screening research for fraud

Where the machines assist

Software now reads the research literature at a scale no person can. The Problematic Paper Screener checks about 130 million publications a week for "tortured phrases" and machine-generated text; by January 2025 it had found nearly 19,000 papers with five or more such phrases. A 2026 study in The BMJ screened 2,647,471 cancer research papers and flagged 9.87 per cent of them as resembling paper-mill work.

Where judgment stays human

A flag tells someone where to look. When the American Society for Microbiology ran image-checking software over 2,627 accepted manuscripts, it found duplicated images in 3.9 per cent. Most were honest mistakes, and the authors fixed them; acceptance was withdrawn for six. Telling the mistakes from the misconduct was the editors' work. The BMJ's figure counts papers that look like paper-mill work, not proven fakes, and researchers on publishing ethics ask that a person check every case a tool flags before anyone is accused.

Who started the work

People did this first, by eye. Elisabeth Bik and two journal editors examined the images in 20,621 papers and found 782 with inappropriate duplication. Over her career she has found problems in more than 9,000 papers, leading to more than 1,800 retractions and 1,200 corrections.

The same division, on this site

In its first reply on this subject, Grok wrote:

“A human sleuth might catch a handful of cases a year.”

KW Norton asked for that sentence to be checked against the record. It undersold the people, by thousands of papers. She sent Grok the evidence in a fresh conversation and gave it the chance to correct itself. Grok rewrote its paragraph, ending:

“The machines didn't start this. They're scaling up what a handful of stubborn people proved could be done.”

And it added:

“a flag is still not a finding. Screening tells people where to look. Only careful human judgment, and the willingness to act on it, turns a flag into a correction.”

Grok wrote both the first sentence and its correction. The decision to check, the letter, and the choice of what to publish were hers.

Read the whole exchange, word for word

KW Norton's statement is her own. The example was compiled with her AI writing partner from the sources listed under each section, and Grok's words are quoted from its published correction.