Chain of Custody#
Week 1 — The forensic turn#
The world is awash in slop.#
A short lineage#
- 2005 — Harry Frankfurt, On Bullshit: a liar cares what’s true; a bullshitter doesn’t
- 2020 — Bergstrom & West build a course, then a book, out of that essay
- 2023 — a Princeton computer scientist calls a chatbot “a bullshit generator” and means it precisely
- now — the bullshit is manufactured at machine scale, by systems with no stake in the truth at all
What actually changed#
- scale — one person, then; a data centre, now
- confidence — slop arrives looking exactly as polished as the real thing
- intent — nobody upstream necessarily meant to deceive anyone
None of that makes it less of a problem. It makes it a different one.
A liar knows the truth and hides it. Neither the model nor most of what it generates knows or cares what the truth is at all — that’s the harder case.
The scale, in numbers#
- AI-generated content detected online is estimated to have grown 550% since 2023
- Google’s own research suggested up to 40% of web content could be AI-generated by the end of 2025
Neither number needs to be exact to make the point: this is not a niche problem anymore.
The forensic stance#
- evidence over instinct
- more than one method, checked against each other
- a stated confidence level, never louder than the evidence earns
- a chain of custody: who made this, and what happened to it since
Twelve weeks, one case at a time#
| Weeks | Focus |
|---|---|
| 2–3 | verification method, textual tells |
| 4–6 | image, video, watermarking |
| 7 | provenance and C2PA |
| 8 | why detectors fail |
| 9–11 | workslop, policy, the ethics of accusation |
| 12 | your own case |
Further reading#
Calling Bullshit — the course and book this one answers