This week's theme is in praise of friction. Nearly everything in this issue circles the same question from a different angle: which parts of learning are valuable precisely because they are slow and effortful, and which can be safely handed to a machine? A historian of "teaching machines" reminds us this question is a century old; a New York Times columnist stakes out the maximalist answer for writing; a calligrapher in China offers a gentler version of the same case; Brookings publishes a premortem; Brussels moves a deadline; and a viral interview thread hints at how the world of work is re-answering the question for us.
History
Before drafting an AI policy, it helps to remember we have been here before. In an interview with the MIT Press Reader, Audrey Watters, author of Teaching Machines, retells the century-long history of automated instruction: psychologist Sidney Pressey's 1924 "Automatic Teacher," B. F. Skinner's behaviorist teaching machines of the 1950s, and the recurring sales pitch — automation plus personalization — that reappears with every generation of technology. Watters notes a "strange amnesia" in how ed-tech reformers talk about supposedly unchanging schools, and argues the rhetoric of inevitability is itself the problem: "Educators want students to understand what it means to have agency over their own learning. We want them to develop curiosity. To say that the future is already written is antithetical to that."
Worth a look: the photo of Skinner's original teaching machine, mid-article — see it in the original.
Small liberal-arts colleges are the institutional counterargument to automated instruction: what we sell is precisely the non-automatable part — seminar discussion, mentorship, disagreement in real time. Watters's history is a useful framing device for faculty meetings: the question isn't whether AI tutors "work," but which of our practices we'd defend even if they did.
Source: "Can the Classroom Ever Be Automated?" — The MIT Press Reader (interview with Audrey Watters)
Opinion
In an August 4 New York Times column titled "I'm Begging You: Never Write With A.I.," Bret Stephens makes the maximalist case: "Don't use artificial intelligence to help you write. Never let A.I. do your writing for you" — school papers, work briefs, even routine emails included. His argument is cognitive rather than ethical: AI-assisted writing is "mentally enfeebling," "an escalator toward a result when you really need to make a daily habit of taking the stairs," because writing is the activity that compels thought. He also points to Dan Sarofian-Butin's analysis of 100 EdD dissertations published in 2025 as a sign AI is seeping into the scholarly record — a single-sourced finding worth holding loosely. Critics note he draws no line between generative AI and ordinary grammar tools, and doesn't engage the accessibility case.
Our writing-intensive seminars are built on exactly Stephens's premise — writing as thinking. But a blanket "never" is hard to hold in classrooms where many students write in a second language and AI is a real leveler. The column is a good provocation for a department conversation: which assignments exist to produce text, and which exist to produce thought?
Sources: Bret Stephens, The New York Times (gift link) · summary and reaction at AI Weekly
Essay
Karolina Pawlik, an assistant professor at Xi'an Jiaotong-Liverpool University, makes the slow-writing case from a very different tradition. In an essay for The Conversation, she argues that as generative AI spreads, "humans are turning from writers into prompting masters and editors of machine-generated content" — and points to Chinese calligraphy, "one of the oldest handwriting practices still in use today," as a counter-discipline: writing with a brush demands slowing down, attentiveness to posture and movement, and "a clear mind, sincerity, humility and self-restraint." She profiles ink artist Pan Jianfeng, whose works resist machine text-recognition but reward human imagination. Pan's line lingers: "People have little trust in themselves and believe too much in technology, so they easily get lost."
This one is close to home: Pawlik writes from a Sino-foreign joint-venture university much like ours. Calligraphy is a resource our context offers that most AI-and-education debates ignore — an embodied, culturally grounded practice of attention that could anchor a signature DKU course or co-curricular on writing, thinking, and AI.
Sources: The Conversation (original) · republished at Asia Times · related: Rebecca Winthrop's LinkedIn essay on the same theme (login required; not independently verified)
Research & policy
Rebecca Winthrop, director of Brookings' Center for Universal Education, is making the rounds — most recently a Talks at Google conversation on rethinking the purpose of education in the age of AI — with findings from the Brookings Global Task Force on AI and Education. The underlying report, A New Direction for Students in an AI World, drew on interviews, focus groups, and consultations with over 500 students, teachers, parents, education leaders, and technologists across 50 countries, plus a review of over 400 studies. Its blunt conclusion: "at this point in its trajectory, the risks of utilizing generative AI in children's education overshadow its benefits." The report organizes twelve recommendations under three pillars — Prosper, Prepare, Protect — including "use AI tools that teach, not tell" and preserving deliberately AI-free human interaction.
The premortem framing travels well to higher ed: rather than waiting to autopsy what AI did to our students, name the risks now and design against them. The three pillars map neatly onto things a small college can actually do — rework pedagogy, teach holistic AI literacy, and set procurement and privacy standards — without waiting for sector-wide consensus.
Sources: Brookings report page · Talks at Google video · via Winthrop's LinkedIn post (login required)
Regulation
August 2, 2026 was long circled as the date the EU AI Act's "high-risk" rules would bite for education — covering AI used to evaluate learning outcomes, screen applicants, and monitor candidates during examinations. But under the Digital Omnibus on AI, agreed this spring, that Annex III deadline moves from 2 August 2026 to 2 December 2027 (with AI embedded in regulated products getting until 2 August 2028). The substance is intact: institutions deploying high-risk AI still owe human oversight, monitoring, logging, and a Fundamental Rights Impact Assessment before use — and the Commission can pull the deadline forward once standards are ready. One assessment-platform provider's advice to universities: build a genuine register of AI in assessment now, because "December 2027 is not a snooze button."
We aren't governed by Brussels, but the AI Act is becoming the global reference point — and partner institutions, accreditors, and exchange programs will increasingly ask AI Act-shaped questions. The homework is portable: could we list every tool on campus that scores, screens, or monitors students? For a small college, that inventory is a one-semester project, not a bureaucracy.
Sources: UNIwise on the AI Act and assessment · WINS Solutions on the deferred deadline · Jones Walker on what still applied August 2
Student experience
A widely shared X post from late July — an engineer's account of interviewing at OpenAI — struck a nerve for what it implied about hiring in the AI era: the expectation that companies are moving away from traditional data-structures-and-algorithms drills toward live, practical coding. That matches published accounts of OpenAI's process, which describe interview problems that are "actual things that you might do at work" — building a file-system traversal, a spreadsheet formula engine with caching, a small SQL engine — rather than LeetCode puzzles, with a probing deep-dive into what candidates really did on past projects. In other words: the assessment frontier in industry is shifting from rehearsable artifacts toward observed, authentic performance — exactly the direction many faculty are dragging their own assignments.
Students often hear "AI-era skills" as a reason to grind more certifications. This thread suggests the opposite: employers are probing for judgment, ambiguity tolerance, and the ability to explain one's own work — which is what oral exams, vivas, and signature-work projects at a liberal-arts college already train. Useful ammunition for advising conversations and for career services.
Sources: spotted via @coolcoder56 and @dotey on X · corroborating detail: interviewing.io's OpenAI interview guide · Exponent on interviewing at OpenAI in 2026
We are collecting examples for AI experimentation at DKU.
— Yisu