Start here
Thanks for stopping by the board. This page is everything the sixteen tiles could not hold, in the order most people want it: the poster as one scrolling page, what to try at work this quarter, a plain-English glossary for any term that sounded like another field, and the methods and readings behind it. Everything works offline once downloaded. If you have five minutes, read the poster; if you have one, read Try this quarter.
What is inside
| Page | What you get | Time |
|---|---|---|
| Digital poster | The whole story on one page with the result figures: how well humans agree, what the models did, the competence result, and a decade of nurses’ forum talk. The print deck is linked at the bottom. | 8–10 min |
| Plain-English glossary | Every term from the board and the paper in one plain sentence each, plus why it mattered here and one thing to read if you want more. Dip in; no need to read it straight through. | dip in |
| Methods and lessons | What we ran, what held up, what did not, and what we would do differently next time. | 15–20 min |
| Reading list | A short list, not a dump. Each entry says in a sentence why a practitioner might open it. | keep |
| Try this quarter | Two tracks: before you score employee text with a model, and before you automate part of a helping job. A twelve-week plan, vendor questions for both, and a one-page worksheet for a staff meeting. Grounded in the work-design and algorithmic-management literature. | use at work |
| Design notes | Why the charts look the way they do, keyed to the data-visualization books they borrow from. | optional |
The results in one screen
What the studies found
- Human coders agree strongly on agency (.93) and others-directed meaning (.84); communion (.48) and self-directed meaning (.52) need better human agreement before any model can be judged on them. The ceiling is dimension-specific, and it is now written down.
- Finding 2. None of six locked setups reached the agency bar. Agency .750–.798 against .910. Highest .798 is .11 below the bar.
- Meaning narrated as competence or mastery at Time 1 went with higher turnover intentions and exhaustion a year later (17 of 140 matched cases; held after adjusting for the full Time 1 survey set). Exploratory, and the lead for the next designed study.
- The codes track concurrent self-reported meaningfulness (r = .20–.24), and felt meaning rises with the number of pathways narrated up to two, then levels.
- Nurses name the apparatus, not AI: seven AI events in 7,709 posts, none after 2021, against a steady one post in twenty about ratios, metrics, and charting across ten years. What rose was a sense of lacking control (+4 points, with a 2019–2021 site redesign under that rise), and what is at issue is control over how the work is done, not skill.
How far to carry each result
- The planned next-year paths from the broad meaning dimensions came in near zero (three of five inside the band we treat as no effect), which is what singles out competence as the signal that reached next year. Read it as a lead, not a rule.
- Forum rates apply the codebook two humans built on the stories. A 200-post transfer sheet for a second human coding of forum prose is drawn and not yet coded. The year is when the archive copy was taken. Posts that name ratios, metrics, or charting also carry quit and lack-of-control flags more often, in the same post from the same model.
What is deliberately not here
No respondent narratives, no StudyIDs, no model instructions that would let someone rerun the rejected setups on new employee data, no forum post text, and no numbers from the 128 untouched cases (the one 65-case look is reported as descriptive). The live allnurses site was not crawled. Forum material is Internet Archive snapshots.
Monnot, M. J., & Thompson, I. (2026, September 30–October 1). The AI paradox: How optimization may undermine meaningful work [Poster presentation]. Society for Industrial and Organizational Psychology Leading Edge Consortium, Baltimore, MD, United States. https://mjmonnot.github.io/lec-2026/