Lecture 12 — Closing arguments

What eleven weeks of casework add up to, and where the discipline goes from here

10 May 2027

No single method from this course is sufficient on its own, and by now that shouldn’t be a disappointing conclusion — it should be the whole point. Textual tells decay. Watermarks get stripped. Provenance standards only cover what was signed in the first place. Detectors get gamed the moment they matter. What holds up is the practice underneath all of it: treat a claim as a case, gather more than one kind of evidence, know exactly how confident each piece of evidence entitles you to be, and never let the accusation outrun the proof.

That practice was never really about AI. It’s the same discipline Caulfield taught for misinformation and Bergstrom and West taught for statistical bullshit, aimed at a harder adversary because the adversary got faster. It will need to be re-aimed again in a year, at whatever comes after this. That’s not a flaw in the method. It’s the reason the method, and not any single tool inside it, is what you’re actually taking with you.

Outline#

• what generalised across the twelve weeks, and what was specific to today’s tools
• final case presentations
• where this discipline is likely headed next, and how you’d know if you were wrong about that

Further reading#

AI Snake Oil is where the throughline this course has followed keeps going after the semester ends — a full book-length version of the argument weeks 1 and 3 only had time to sketch.