Critical Thinking with AI • Hist 300 • Last Day

What was this course, actually?

A semester of readings, dialogues, prompts, broken builds, and one stubborn question: how do we know what we know — and what changes when a machine starts “knowing” too?

Slide 01 · The Course in One Sentence

Every reading we did was the same question wearing a different costume.

The Question

Whose word counts — and why? Plato asked it about writing. The Royal Society asked it about strangers. Johns asked it about printers. Trouillot asked it about archives. We asked it about chatbots.

The Trick

The trick is to recognize the question when it shows up in a new costume. If you can do that — you can read AI like a historian, not like a tourist.

Slide 02 · Old Questions, New Costume

AI didn’t invent the trust problem. It just made it impossible to ignore again.

The Greatest Hits

Plato: writing will erode memory.
Bacon: don’t trust books, trust your eyes.
Shapin: science runs on the gentleman’s word.
Johns: fixity is a social achievement.
Postman / Nichols: the kids can’t tell experts from posters.

The Punchline

Every new information machine triggers the same panic — and the same scramble to figure out who gets to be a source. AI is the latest installment. Not the first. Not the last.

Slide 03 · What the History Actually Taught

Trust machines are built by people — and they break the same way every time.

Printing Press (Johns)

Two copies of a book aren’t reliably identical. Fixity is social labor, not a feature of the press.

Science Networks (Shapin)

Most of what scientists “saw” they got by mail. Credibility was anchored in the gentleman, not in the data.

The Pattern

Whenever “objectivity” looks effortless, somebody is doing a lot of invisible work upstream. That is also true of your favorite model.

Slide 04 · New Questions Only AI Raises

Some of these problems really are new. We need a new vocabulary.

Genuinely New

A sentence with no author. A “source” that is the average of a scraped continent. RLHF as an invisible authority. Bias not at the level of a single bigot but baked into a corpus. Fluency that increases as grounding decreases.

Why It Matters

You can’t critique what you can’t name. Half the work this semester was vocabulary acquisition: stochastic parrot, jagged frontier, hallucination, hybrid fact, archival silence. Use the words. They earn their keep.

Slide 05 · The Expertise Question

AI raises the bar for expertise — not by replacing it, but by industrializing the fakes.

Beer-Mat vs. Real

Postman and Nichols were already worried about beer-mat expertise — knowing just enough trivia to sound smart at the bar. AI didn’t cause that. It just industrialized it.

What Survives

The expertise that survives looks more, not less, like the kind we read about all semester: slow, social, situated, accountable. The kind a model can imitate at the surface and not at all underneath.

Slide 06 · Why We Made You Build a Website

A different part of your brain has to wake up when the build fails.

The Reason

Reading and writing essays uses one circuit. Editing YAML, fixing a broken link, pushing to GitHub uses a totally different one — closer to fixing a bike than writing a paragraph.

Why It Mattered

AI is happiest in the easy mode — where everything is text and everything is plausible. The build either works or it doesn’t. There is no “sounds about right” in a missing semicolon. The friction was the point.

Slide 07 · The Discomfort Was the Point

If nothing broke, nothing was learned.

What You Probably Noticed

The first time the site failed to deploy was probably the first time you really understood what files do. The first time AI got it wrong was probably the first time you noticed your own reasoning. Both of these are gifts disguised as bugs.

Translation

Comfort is the enemy of skill. Don’t bring a forklift to the weight room. You came here to lift.

Slide 08 · The Driving Question, Revisited

Did AI help us think more critically — or did the course just use AI as bait?

“How can AI help us think more critically, develop skills, and produce higher-quality work than we could without it?”

My Answer (Tentative)

Both. AI is most useful when it gives you better questions, not better answers — and you only know it’s a better question if you brought your own. The course is the place to practice bringing your own.

Slide 09 · Habits Worth Keeping

Four moves — one for each ghost from the syllabus.

Shapin

Triangulate

Don’t take any single source’s word for it — not the model’s, not the textbook’s, not mine.

Trouillot

Notice fluency

The smoother the answer, the more it has hidden. Smoothness is a clue, not a virtue.

Johns & Crawford

Ask “from where?”

Every claim has an address — a printer, an archive, a server farm, a labeler in Nairobi.

This Course

Show your work

The artifact isn’t the value. The process is. AI raises the bar for proving the process is yours.

Slide 10 · Questions Worth Keeping

More questions than answers — on purpose.

  • When you trust an AI answer, which Stationer are you really trusting?
  • What is literally unthinkable to your favorite model?
  • What goes silent when the prose gets fluent?
  • What does it mean to be the author of an AI-assisted thing?
  • What kind of expertise is worth your next four years?
  • Which of these questions will still be the right question in 2030?

The Hope

The course wasn’t about AI. It was about giving you tools to keep evaluating it after it changes.

AI in two years will be unrecognizable. The questions won’t. If you walk out asking better ones — out loud, in writing, of yourself, of the model — the course did its job.

Thanks for the semester. Now go break something on purpose.