The Joy of Writing (Psychology Today)

James Danckert, Ph.D., cognitive neuroscientist at university-of-waterloo, writing in his Modern Boredom blog at psychology-today, 2026-09-24, reviewed by Ekua Hagan. Written while drafting a paper of his own: "There is joy in the challenging process of writing." His claim is not that AI writing tools are bad, but that handing over a task you could do yourself has a cost, and that cost is rarely counted.

Three losses, in the order he gives them

1. We become more boring. LLMs are next-word predictors, so the output regresses toward the statistical average of their training data. Danckert cites Maik Bieleke and Wanja Wolff for the framing — "we will all begin to sound alike" — and pictures every novel reading like Dan Brown's (he says he has never read one).

2. We lose the joy of effort. The standard account of human cognition treats people as cognitive misers — conservers of a scarce processing budget. michael-inzlicht's research runs the other way: effort itself can be rewarding, and "sometimes we choose things because they are effortful" — his example is the tattoo, freely chosen pain. Outsourcing writing to an LLM skips the part that would have felt good.

3. We never develop our own voice. Danckert holds up George Miller's 1956 opening line — "My problem is that I have been persecuted by an integer" — as a distinctly Miller sentence, and laments that today's scientific writing tolerates less of that personality. He cites an arXiv study, Yakura et al., for a sharper version of the same worry: people who talk to a chatbot start using its words. "We are not just beginning to sound alike; we are sounding like AI, and this should disturb us."

He calls abandoning LLMs unrealistic ("the genie is well and truly out of the bottle") and recommends, in order: accept the difficulty rather than avoid it; use AI to polish grammar only after the ideas are your own (useful, he notes, for scientists writing in a second language); use it as a sparring partner — following Marc Schmitt's "Use AI as a sparring partner, not an oracle" (Nature, 2026), present an argument and ask the model to attack it — while adding that a chatbot cannot replace a human conversation.

The disclosure, stated up front, not buried

In a parenthetical, Danckert writes that he opposes "the theft of work to build AI models" and states plainly: "I will be a beneficiary of the successful lawsuit against Anthropic after they stole my book." The book is Out of My Skull: The Psychology of Boredom (Harvard University Press, 2020), co-written with john-eastwood — training-data litigation against anthropic over unauthorized use of copyrighted books is real and unrelated to this piece's claims (the settlement Danckert anticipates is anthropic's own to describe, not this spoke's to verify). It is disclosed, not hidden, which is unusual enough in this spoke's corpus to record on its own terms: an author arguing against outsourcing creative labor to LLMs has a stated financial interest in exactly that argument, and says so in his own voice rather than leaving a reader to find it.

What the citations hold up, and what they don't

Unlike why-writing-feels-hard — a citation chain that welds real studies to an untested practice — this piece's three central claims each rest on a source that says roughly what Danckert says it does, checked here against the primary or its abstract:

• Point 1 (homogenization) rests on a PsyArXiv preprint (wzveh_v1) by maik-bieleke and wanja-wolff that this ingest could not read — the OSF page renders its content client-side and returned nothing on fetch. The claim itself is plausible on general grounds (predictive text generation trends toward the probable, not the distinctive) but as cited here it is unevidenced from this spoke's side: no title, no method, no sample. Recorded as a gap, not a refutation.
• Point 2 (effort) points to an APA PsycNet record (2010-19536-001) for the "cognitive misers" claim and a ScienceDirect PII (S1364661318300202) for Inzlicht. Checked: the PsycNet record is Kool, McGuire, Rosen & Botvinick (2010), "Decision making and the avoidance of cognitive demand," Journal of Experimental Psychology: General — a demand-selection task, not the term's origin (that's Fiske & Taylor, 1984). The ScienceDirect PII resolves to Inzlicht, Shenhav & Olivola (2018), "The Effort Paradox: Effort Is Both Costly and Valued," Trends in Cognitive Sciences 22(4):337–349 — a real, correctly characterized review. See cognitive-miser for both citation chains in full.
• Point 3 (voice) cites Miller (1956) accurately (a real, celebrated paper) and arXiv:2409.01754 accurately in outline. Read directly for this ingest: it is the strongest source in the piece — a preregistered experiment (N=496) plus a synthetic-control analysis over 737,083 hours of podcast speech, not an opinion. See its own page.

Net assessment. Two of three load-bearing citations check out on inspection, one (Bieleke & Wolff) could not be verified at all from this spoke, and the strongest evidence in the piece — the Yakura et al. speech study — is undersold: Danckert cites it for "buzzwords creeping into speech" and skips its more striking result, that the effect appeared in a forced-choice vocabulary test after a distractor task, meaning the borrowed words entered active vocabulary rather than sitting as a passing mimicry.

One correction worth keeping

The article's byline for the homogenization claim reads "Mark Bieleke." The researcher's name, per his own publication record and university page, is Maik Bieleke — a small error, but one this spoke's practice is to record rather than silently fix, per the same convention applied to generational-boundaries reporting.

Related

cognitive-miser · michael-inzlicht · james-danckert · john-eastwood · yakura-llm-human-speech-2024 · george-a-miller · maik-bieleke · wanja-wolff · marc-schmitt · why-writing-feels-hard · synthesis