Chapter 38 Quiz

Closed book, tools closed, hype detector on. This quiz tests the durable layer of the chapter — the audit questions, the failure modes, the consent lines, the honest comparisons — not the names of products that may not exist by the time you read this. Multiple choice first, then true/false where the justification carries the points, then short answer, then one applied scenario shaped exactly like a message you will someday receive. Answers hide under each Verify fold; scoring guide at the end.

Section 1 — Multiple Choice (2 points each)

1. In this chapter's working definition, what distinguishes "AI" tools from conventional DSP?

A. AI tools run in the cloud; DSP runs on your computer B. AI tools' behavior was learned from training examples rather than written as explicit rules C. AI tools work on full mixes; DSP works on single tracks D. AI tools are newer than 2020

2. Which tool family does the chapter call "the genuinely solved one"?

A. AI mastering services B. Generative full-track models C. Stem separation D. Mixing assistants

3. The two axes every AI tool gets measured on in this chapter are:

A. Price and speed B. Whether you can evaluate the output, and whether it replaces tedium or judgment C. Audio quality and platform compatibility D. Training-data legality and disclosure status

4. Why do cymbals come out of stem separation sounding "watery"?

A. Cymbals are recorded with worse microphones B. High frequencies can't be sampled accurately C. Cymbal energy is wideband and diffuse, overlapping everything, so the learned mask must guess at ownership D. Separators deliberately filter cymbals to avoid copyright issues

5. A separated a-cappella from a commercial record is:

A. Cleared for release, since the AI created a new file B. Cleared if you credit the original artist C. Still the original master recording and composition — full Chapter 36 clearance applies D. Legal for streaming but not for sync

6. The chapter's verdict on mixing assistants hangs on one condition:

A. Only use them on vocals B. The suggested starting point must never silently become the ending point — bypass-test every chain at matched loudness C. Only use them in genres they were trained on D. Never read the suggested settings, only listen

7. Why does an automated mastering service tend to flatten a deliberate dynamic dip?

A. Limiters can't pass quiet audio B. The service compares your track to statistical averages of finished records, and the average record doesn't drop there — intent isn't in the audio C. Streaming platforms prohibit dynamic bridges D. The service assumes all tracks are EDM

8. The chapter's "diagnostic" use of an AI mastering service: if the service's master sounds dramatically better than yours at matched loudness, the correct conclusion is —

A. The service is more skilled than any human B. Your mix or master carries a real, findable problem — run Chapter 30 on it C. You should always outsource mastering D. Your monitors are broken

9. The deepest structural failure of generative full-track tools, per the chapter:

A. They can't make music in minor keys B. Revision — output arrives as printed audio with no stems, session, or faders, so "change one thing" means regenerate-and-hope C. They only output MP3s D. They can't match a tempo

10. The three consent lines for voice cloning, in order from safe to forbidden:

A. Live use → studio use → broadcast use B. Cloning yourself → cloning under license → cloning a real artist without consent C. Short clips → full songs → albums D. Demos → singles → albums

11. Which is the strongest rights-holder argument in the training-data debates, as the chapter steelmans it?

A. AI music sounds bad B. Computers shouldn't be allowed to listen to music C. Industrial-scale ingestion builds a product that competes directly with the catalog it consumed, in an industry where far smaller takings require licenses D. Training is too expensive

12. As of this chapter's writing, the US Copyright Office's position on generated output is best summarized:

A. All AI output is automatically copyrighted to the tool's developer B. Purely machine-generated material isn't registrable; protection requires meaningful human authorship C. AI output is copyrighted for 10 years instead of life-plus-70 D. The question has been fully resolved by the Supreme Court

13. The chapter's "what won't change" list rests on the observation that the scarce resource in music was never:

A. Audio — the world has been oversupplied with recorded music for decades; attention, meaning, and trust are the bottlenecks B. Talent C. Money D. Distribution

14. The Billie Eilish lesson "re-applied" to the AI era states:

A. Expensive studios are obsolete B. Tools democratize; knowledge differentiates — every wave of democratization raises the value of the differentiator C. Bedroom production guarantees Grammys D. AI tools should be banned from bedrooms

15. In the AI audit's decision grid, "one-button anything you can't yet audit" lands in which quadrant?

A. Adopt freely B. Use as second opinion C. Low risk, low need D. Do not delegate

Section 2 — True/False with Justification (3 points each: 1 for the call, 2 for the why)

16. True or false: This chapter is the first place in the book where machine listening appears.

17. True or false: Because automated masters usually arrive louder than your master, your first A/B impression of them is systematically biased in their favor.

18. True or false: The chapter argues that because previous tool panics (tape, drum machines, sampling, Auto-Tune) all turned out fine, generative AI is the same situation and there's nothing genuinely new.

19. True or false: In the chapter's framework, a producer who can't yet hear what an AI tool did is exactly the producer who benefits most from using it.

20. True or false: Stem separation used for private practice — muting the bass on a record to play along — sits in safe territory under the chapter's clearance reminder.

Section 3 — Short Answer (5 points each)

21. State the three questions of the AI audit and, for each, name the earlier chapter whose skills answer it.

22. Explain why the chapter calls the matched-loudness A/B an "antibody" in the AI era — what bias does it neutralize, and why do automated outputs trigger that bias so reliably?

23. A bandmate says: "AI mastering exists, so Chapter 31–33 was a waste of our time." Give the two-part rebuttal the chapter would make.

24. Describe the emerging disclosure landscape in category terms (no product names needed), and the "provenance hygiene" habit the chapter prescribes regardless of where the norms settle.

Section 4 — Applied Scenario (10 points)

25. A producer friend messages you: "Done deal — I ran our single through an AI master and it SLAPS, way better than the one you made. Shipping it to the distributor tonight. Also I pulled the vocal out of that 2003 hit with a separator for the bridge — sounds totally clean, no one will know. And I generated the B-side entirely with a prompt tool; saves us a week. You in?"

There are three separate problems in this message. Identify each, explain the chapter's reasoning, and write the plan you'd send back — concrete steps, not vibes. (Partial credit for two problems; full credit requires the verification protocol, the clearance issue, and the rights/ownership issue, plus a workable plan.)

Scoring

60–75: You can be trusted alone with a robot — you'll audit it, and it'll make you faster. 45–59: The frameworks are in; re-run the blind A/B (B1) and the audit worksheet (C1) to make them muscle. Below 45: Reread "The Craft Argument" and "The AI Audit," then redo the quiz before Chapter 39 — the capstone assumes you can render verdicts, because release day doesn't grade on intentions.