1. Refik Anadol's "Unsupervised" installation at MoMA was trained on:
Correct. Anadol's studio trained a custom ML system on MoMA's own collection — making the institution's art history the training data for a work displayed within it.
"Unsupervised" used a machine learning system trained on MoMA's collection of 200 years of art, generating real-time visual forms from that training in the museum's atrium.
2. The lesson on guardrails notes that applying guardrails too rigidly is itself a problem because:
Correct. Over-applying guardrails is an overcorrection — it prevents real leverage gains. The goal is intentional use that stays in the leverage zone, not complete avoidance.
Rigid guardrails become their own limitation — preventing the genuine leverage AI provides. The goal is intentional use in the leverage zone, not abstinence.
3. The "material assistance threshold" for AI disclosure means that disclosure is generally expected when:
Correct. The material assistance threshold focuses on whether AI substituted for compensable human creative labor — not on minor workflow tools or precise percentage calculations.
Not quite. The threshold is about whether AI did work a human would have been paid to do. Grammar checking and thesaurus use generally don't meet this bar; generating images, copy, or music drafts generally does.
4. The AP's published AI guidelines specify that AI-assisted content requires what specific editorial mechanism?
Correct. The AP's standard requires named human editorial accountability — not just a review window or a label — as the mechanism ensuring factual verification actually occurs.
The AP's standard requires a named human editor who accepts full editorial responsibility — the accountability mechanism that forces rigorous factual review to happen.
5. What is the primary argument AI companies make in their fair use defense for using copyrighted works as training data?
Correct. Transformativeness is the central fair use argument, with Authors Guild v. Google (2015) as the key precedent. The counterargument focuses on market effect: unlike Google's index, AI tools generate commercial outputs competing with original creators.
Not quite. The primary argument is transformativeness — that training use creates something fundamentally different from the original works, similar to how Google's digitization created a searchable index. Market effect (factor 4) is where this argument faces its greatest challenge.
6. The PLOS ONE 2024 study estimated that what percentage of peer-review text submitted in late 2023 showed AI-generation markers?
Correct. The Stanford/PLOS ONE study estimated 6–16.9% of peer-review text had likely been AI-generated.
The study found 6–16.9%, detected by vocabulary clustering around words AI models statistically favor.
7. What did the partial ruling in Andersen v. Stability AI (November 2023) establish?
Correct. The ruling's significance was that it kept the door open for direct copyright infringement claims — the court didn't dismiss all claims, signaling that certain theories of AI-related infringement remain viable.
Not quite. The ruling was mixed: most claims were dismissed, but Andersen's direct infringement claim survived. The key takeaway is that direct infringement claims against AI companies are not categorically barred at the pleading stage.
8. Andreas Gursky's description of digital editing — using it to realize visions he had before sitting at the computer — illustrates:
Correct. Gursky's practice is the career-length causal chain principle: creative roots predate and are independent of the tools used to realize them.
Gursky illustrates the career-length causal chain principle: vision comes first, tools realize it. AI and digital tools are downstream of creative identity, not the source of it.
9. What is a "digital replica" as defined by the SAG-AFTRA 2023 contract?
Correct. In the SAG-AFTRA framework, digital replicas are AI-generated likenesses or voices that realistically reproduce a specific performer — the consent and compensation requirements apply specifically to these AI-created replicas.
Not quite. A digital replica in the SAG-AFTRA context means an AI-generated reproduction of a performer's likeness, voice, or performance — the kind created by training AI on that performer's work and then generating new content in their image or voice.
10. Why should long pieces be drafted section by section rather than in a single prompt?
Correct. Token drift — the degradation of brief fidelity in later sections — is the primary reason to draft section by section and re-supply context with each new prompt.
The reason is token drift: in long single-prompt drafts, AI models progressively lose track of the tone and argument established in the brief, leading to inconsistent output.
11. A LoRA (Low-Rank Adaptation) in the Stable Diffusion ecosystem is:
Correct. LoRA is a fine-tuning technique enabling users to train a small supplementary model on their own images and attach it to a base Stable Diffusion model for style-consistent or subject-consistent outputs.
Incorrect. LoRA is a fine-tuning technique — a small supplementary model trained on custom images and attached to a base Stable Diffusion model.
12. The 2024 Science Advances study of Amazon reviews found that AI-patterned reviews were rated lower on which dimension specifically?
Correct. Grammatical correctness went up while consumer helpfulness ratings went down — readers sensed the absence of specific experience.
Helpfulness ratings went down even as grammatical correctness improved — demonstrating that readers sense the absence of real specific experience.
13. Which of the following is identified as a "healthy signal" — not a warning sign — of a well-maintained creative identity under sustained AI use?
Correct. Making choices that go against AI's tendencies indicates your judgment is still operating independently — a healthy signal that AI hasn't normalized your aesthetic baseline.
The other three are all warning signals. Making choices that would surprise the AI — going against its statistical tendencies — is the healthy signal of an intact independent creative identity.
14. In Robin Sloan's documented workflow for Moonbound, what was the AI's role?
Correct. Sloan used AI output as a pressure to write against — defining his choices by opposition rather than by adoption.
Sloan's method was the opposite of adoption: AI produced sentences he wrote against, using none verbatim — a "tuning fork" not a drafter.
15. Screenwriter John August, on his podcast Scriptnotes in March 2023, described using AI to draft "bad first scenes" primarily to:
Correct. August described the deliberate bad draft as a blank-page elimination tool — the AI gets him past the paralysis, but his own voice takes over in the rewrite.
Incorrect. August's reason was specific: the AI bad draft removes the blank page without replacing his voice. He then rewrites — the AI gets him started, not finished.
16. The "Heart on My Sleeve" incident featured an AI track that convincingly mimicked which two artists?
Correct. "Heart on My Sleeve" by Ghostwriter977 convincingly mimicked Drake and The Weeknd, triggering DMCA takedowns by Universal Music Group and a congressional inquiry.
"Heart on My Sleeve" mimicked Drake and The Weeknd — convincingly enough to accumulate millions of streams before Universal issued DMCA takedowns.
17. The Recording Academy's 2023 AI task force principle on production chain position states:
Correct. This is the formalized professional version of the dirty first draft rule: establish musical identity through human decisions first, then AI can extend and process without erasing the foundational act.
The principle is about causal chain position: AI enters after human creative decisions have established musical identity — not before.
18. When registering copyright on AI-assisted work with the U.S. Copyright Office, what is the applicant's obligation regarding AI-generated content?
Correct. Full disclosure of AI elements is required, and registration covers only the human-authored portions. Concealing AI involvement can invalidate the registration.
Not quite. The Copyright Office does accept registrations for AI-assisted works. The requirements are: disclose AI-generated elements and claim protection only for the human-authored portions of the work.
19. In Getty Images v. Stability AI, what was the significance of watermark artifacts appearing in model outputs?
Correct. Watermark artifacts were striking evidence of training data memorization — specific visual features from Getty images being reproduced in model outputs, central to Getty's infringement claim.
Incorrect. The watermark artifacts evidenced training data memorization — specific features of Getty images being reproduced in outputs.
20. What did the January 2024 Biden robocall incident lead the FCC to propose?
Correct. The FCC responded to the Biden robocall by proposing rules that would require disclosure when AI-generated audio is used in political advertising — a direct regulatory response to the incident.
The FCC's response to the Biden robocall was to propose disclosure requirements for AI-generated audio in political advertising — not a ban, but a transparency mandate.