We Are in the Uncanny Valley  🔗 **Source:** [We Are in the Uncanny Valley](https://next.jazzsequence.com/posts/we-are-in-the-uncanny-valley) > [!note]+ Document Notes > tags: [generative-ai, ai-music, creative-authenticity, tabletop-rpg, ai-ethics, creative-labor] > author: Chris Reynolds > source: jazzsequence.com > reference: https://next.jazzsequence.com/posts/we-are-in-the-uncanny-valley > date created: 2026-09-22 > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > > [!warning] Disclosure. This analysis was written by Claude, one of the two AI tools the author > > describes using to make the album. Weigh its reading of the author's AI-use argument with that in mind. > > # 🏷️ Title: We are in the Uncanny Valley > <!-- deep-read:sections start --> ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [The PARIS album on Suno](https://suno.com/s/GrPOH6Il9JQ0ROlg) — linked in the post - [This Is What It Sounds Like (Goodreads)](https://www.goodreads.com/book/show/60165381-this-is-what-it-sounds-like) — the book the author cites on authenticity ## 📝 Summary - **What happened.** Reynolds is running a *Mage: the Ascension* campaign set in the 1990s, with *Severance* as the thematic base. Over a weekend they made a full album for an in-game NPC, PARIS, a Lestat-like vampire shock-rock star. Claude wrote the tracklist, lyrics, per-song style notes and track order. Suno (v6) generated the music. Earlier sessions' lyrics came from ChatGPT. They plan to burn it to a physical CD as an in-world prop. - **Why it matters.** The unsettling part, in the author's words, is that "the album is really good." The author has listened all weekend and sings along. The post is an honest attempt to explain liking something they have ethical objections to. - **Key stakeholders.** Hobbyist creators, working musicians and film composers, AI music companies (Suno, which faces training-data lawsuits per the author), and studios weighing AI for scores. - **Risks and tradeoffs.** Personal use with no lost commission versus normalizing tools built on uncompensated training data. The author's own "I wouldn't have paid an artist" defense stops holding at studio scale, and the author says so. - **Bottom line.** A candid first-person case study. Generative music now clears the "good enough to love" bar for personal, context-heavy uses. The author argues that decisions about the technology have to happen now, before creative skill atrophies, but doesn't propose what those decisions should be. ## 📖 Main Content - **The workflow.** An LLM does creative direction (tracklist, lyrics, style prompts, liner-note context, sequencing). A music model renders audio from the lyrics plus a style field. The human curates character, references and taste. This is a three-layer creative stack where the human authors the brief, not the artifact. - **The ethics frame the author sets up first.** Suno trained on copyrighted music without artist consent or compensation (lawsuits are ongoing, per the author). The author also objects to the Suno founder's view that no one wants to do the hard part of making music. The author states a belief in paying artists, then argues this use takes no paycheck because no commission would have happened. The counterfactual would have been a mismatched mixtape, or an image found online "whether or not that usage was strictly allowed." - **Five explanations for liking it.** (1) The style influences trace back to the author's own curated references. (2) It works because of the campaign story, and outsiders wouldn't care. (3) The lyrics are the weakest part: little cadence, little rhyme, lots of spoken word, which the author attributes to AI not "getting" poetry. (4) Creator bias from having prompted it into being. (5) The music is "risk-averse," tuned to a median of enjoyment and "designed for me to like it." - **The authenticity problem.** Citing *This Is What It Sounds Like* (Susan Rogers and Ogi Ogas), the author notes authenticity as one of the things listeners connect to. They find it unsettling to react as strongly to something with "no authenticity" as to human work. The post equates authenticity with human authorship. The book's own definition may be narrower, and the post doesn't quote it. - **The scale-up fear.** The personal-use defense doesn't survive the studio case. A studio can generate a score to "exacting standards" and skip John Williams or Danny Elfman, which means skipping the conversations and having to "accept their ideas and creativity." What's lost is friction with another person's judgment, not only a paycheck. - **Closing argument.** "We're here." Without decisions now, creativity gets "leeched from us" as people hand the muscle over to machines that deliver exactly what they want. - **Contested or uncited.** The lawsuit reference and the founder's views carry no citation in the post. Its two links go to the album and the book. Both claims are widely reported elsewhere but aren't sourced here. The lyrics-don't-rhyme diagnosis ("it doesn't get poetry") is also contestable, since style prompts that ask for spoken-word or shock-rock delivery could produce the same result. ## 🗒️ Notes - Campaign: *Mage: the Ascension* (the author's first-ever game system), 1990s setting, *Severance* thematic base. Planned for almost a year, started "Friday." - NPC PARIS: a Lestat-like vampire rock star who defected from the clans and is "too big to take down," singing about real events. A second NPC decodes the lyrics to reach the player characters. - Tools: ChatGPT (earlier lyrics), Claude (tracklist, lyrics, style notes, context, sequencing), Suno v6 (music, using the "style" input for non-lyric notes). - Output: a full album, planned as a physical burned CD prop. - Stated ethics: Suno's library is "stolen" (no consent or compensation), there are lawsuits, and the author disagrees with the Suno founder's stance on artists. - Counterfactual defense: wouldn't have commissioned an album, might have made it personally, probably a mixtape instead. - The five self-explanations: curated influences, story dependence, weak lyrics, creator bias, median taste. - Book: *This Is What It Sounds Like*, on authenticity in musical connection. - Studio scenario: AI scores replacing composers, and losing the human back-and-forth. - Close: decide now, or creativity atrophies. ## 📚 References - [The PARIS album (Suno share link)](https://suno.com/s/GrPOH6Il9JQ0ROlg) - [This Is What It Sounds Like — Susan Rogers & Ogi Ogas (Goodreads)](https://www.goodreads.com/book/show/60165381-this-is-what-it-sounds-like) - Not linked in the post, worth pairing: the RIAA-backed label suits against Suno and Udio (filed June 2024). Verify their current status before citing. ## 🗣️ Paraphrase - Something I did this weekend still unsettles me. - I'm running a new Mage campaign with a *Severance* theme. I invented a vampire rock star, PARIS, whose songs describe real events, and an NPC who uses the music to reach the players. - I'd been drafting PARIS songs with ChatGPT and Suno, and this weekend I decided to make the whole album. - Claude planned the tracklist and wrote each song's lyrics, style and backstory. Suno made the music, and Claude sequenced the CD. I'll burn a real disc for the table. - About framing: Suno trained on music without paying or asking artists, and I dislike its founder's views on artists, even though I'm using the tool. - I believe in paying artists, but I wouldn't have hired one here. I'd have grabbed something off the internet instead. - For this project that's even more true. At best I'd have made a mediocre mixtape. - The disturbing part is that the album is good. I love it, and it works as a campaign artifact. - I've tried to explain why. Maybe the influences are really mine. Maybe it only works inside the story. The lyrics are weak, I'm biased because I prompted it, and it's engineered toward median taste. - The book I just read says authenticity is part of why music connects, yet I reacted to something inauthentic just as strongly. - Worse, studios could do this at scale: generate scores to spec and skip composers, and skip their input too. - We're at the decision point. If we don't choose deliberately, we'll hand over our creative muscle to machines that feed us exactly what we want. ## 📰 Newsletter **When the fake band is better than you'd like.** Chris Reynolds made a full concept album for a *Mage: the Ascension* NPC over one weekend, with Claude writing and Suno performing. The [result](https://next.jazzsequence.com/posts/we-are-in-the-uncanny-valley) is good enough to be unsettling. The post works because it doesn't pretend: it names the training-data problem, gives five theories for why the album lands (the best is "it's literally *designed* for me to like it"), and then admits the personal-use defense falls apart at Hollywood scale. For WordPress folks, the same question hits closer than it looks. AI-generated images and copy are already flowing into block themes, pattern libraries and Openverse-adjacent workflows. "I wouldn't have paid anyone anyway" is a real argument for a hobby project. It doesn't hold for a product you ship. If you want the music-psychology backdrop, the author points to [*This Is What It Sounds Like*](https://www.goodreads.com/book/show/60165381-this-is-what-it-sounds-like). ## ✅ Action Items #### DevRel - Developer advocacy: use this post as a discussion case in any talk on AI in creative tooling. It's a practitioner being honest, not a vendor pitch. - Education: build a workshop exercise, "Where does your counterfactual defense stop holding?", covering personal, community, and commercial use. - Developer experience: when demoing AI content features, show provenance and licensing metadata next to the output. - Developer marketing: avoid "it's so good it's scary" framing in AI launches, since it borrows the unease without addressing it. - Community management: when members share AI art, have a lightweight disclosure norm ready so the conversation stays about substance. - Internal/external/cross-project: not applicable. The source raises no cross-project coordination issue. #### OSPO - Policy & Governance: define when AI-generated assets (images, audio, copy) are allowed in project repos and marketing. - License Compliance: treat generated media as having unclear provenance. Record the tool and its terms of service in the asset metadata. - Security Management: not applicable to this source. - Software Inventory: track AI-generated assets in the same manifest as third-party assets. - Developer Support: publish a short guide on disclosure wording for AI-assisted contributions. - Community Engagement: not applicable. - Strategic Alignment: not applicable. #### Community - Maintainers: decide and document a stance on AI-generated media in contributions before the first dispute. - Contributors: disclose AI assistance in PR descriptions when it produced creative assets. - Sponsors: fund human-made assets for flagship surfaces as a deliberate choice, and say so. - Ecosystem: not applicable. #### WordPress - Core: not applicable. - Core Performance: not applicable. - Core AI: the author's own counterfactual test is a useful frame for AI feature guidelines. Check whether a feature replaces a hire that would otherwise happen, or fills a gap nobody would have paid for. - **Core Program Team:** not applicable. - Mobile: not applicable. - Playground: not applicable. - Community: a WordCamp session idea, "AI in hobby vs. production creative work," with this post as the opener. - Training: include a Learn module segment on provenance and disclosure for AI-generated media on sites. - Marketing: set a policy for AI-generated imagery and audio in WordPress.org marketing assets. - Support: not applicable. - Design: consider how AI-generated imagery in default themes or patterns would be disclosed. - Documentation: document how to mark AI-generated media in the media library (alt text, caption, or custom field). - Photos: the Photos directory already bars AI images, so link this post as a plain-language rationale. - TV: not applicable. - Polyglots: not applicable. - Accessibility: AI-generated audio embeds need transcripts and lyrics text like any other media. - Meta: not applicable. - CLI: not applicable. - Hosting: not applicable. - Tide: not applicable. - Openverse: keep AI-generated works clearly labeled or excluded, since the author's concern is training on uncompensated work. - Themes: add review guidance for bundled AI-generated media and its licensing. - Plugins: plugins that generate media should surface the provider's terms and training-data stance to site owners. #### The WPOCC - Programs: consider a small commission fund that pays human artists for community event assets. - Projects: not applicable. - Funding experiments: not applicable. ### FAIR - Technical independence: not applicable. - Policies and covenants: provenance metadata for media assets fits FAIR's provenance-first approach. - Productized governance artifacts: not applicable. ## 🧪 Hypotheses - H1: Listeners who know a song is AI-generated rate it lower on "authenticity" but not on "enjoyment," compared with listeners who don't know. - H2: Creators rate AI outputs they prompted higher than the same outputs prompted by others (the author's "creator bias" explanation). - H3: AI music tuned for broad appeal scores higher on first-listen liking but lower on liking after repeated listens than human-made music in the same genre (the "median taste" explanation). - H4: Context-dependent AI artifacts, like campaign props, get higher ratings from people inside the context than from outsiders (the "it works because of the story" explanation). - H5: In personal creative projects, AI use mostly replaces unlicensed found material or nothing, not paid commissions (the author's counterfactual claim). ## 📊 Methodology - H1: between-subjects listening study with blind vs. labeled conditions. Measure enjoyment and perceived authenticity on validated scales. - H2: within-subjects study where each participant prompts and rates their own tracks alongside tracks from others made with the same prompts. - H3: longitudinal listening diary over 4 weeks, with play counts and liking ratings per track. - H4: rate the same album with campaign players vs. naive listeners. - H5: survey of hobbyist creators (TTRPG, indie games) on what they would have used before AI tools, with a willingness-to-pay question. ## ❓ Questions - Is "authenticity" in Rogers's book about human authorship, or about a performer's credible self-expression, which a fictional persona already complicates? - Does the physical CD change the ethics, since it moves from private listening to an object other people handle? - If the lyrics are the weakest part, would human-written lyrics over Suno music change how the author feels? - Where exactly does the counterfactual defense stop: paid streams of the campaign, a published module, a Kickstarter? - What would "making decisions about it now" look like for a hobbyist, as opposed to a studio or a policymaker? ## 🔮 Future Research - The psychology of emotional response to known-AI art, and whether disclosure changes the response over time. - Skill atrophy: does regular use of generative tools reduce people's own creative output or confidence? - Displacement data for film and game scoring budgets since AI music tools arrived. - Outcomes of the label suits against AI music companies, and their effect on training practices. ## 💡 Implications - **Practice.** The creative skill shifts from making to briefing and curating. That's a real skill, but a different one, and the author's worry is losing the first. - **Policy.** Personal, non-commercial use makes the easiest ethical case. Policy fights will center on commercial substitution, as in the studio scenario. - **Security.** Not applicable to this source. - **Education.** Media literacy should include "why do I like this?" introspection like the author's five-point list. - **Funding.** If commissions shrink at the high end (film scores), funding for human creative work may need explicit carve-outs rather than market demand. ## 🧠 Learning Styles Assessment - Reading level estimate: Flesch-Kincaid grade ~9–10, estimated from conversational sentences with some long compound ones (1,262 words). #### 👀 Visual learners - Present: a header image (paris-uncanny-valley.png). - Improve: a diagram of the Claude → Suno → CD workflow. #### 🔊 Aural learners - Present: the linked Suno album is the primary aural artifact. - Improve: an embedded player next to the discussion of specific tracks. #### 🗣️ Verbal learners - Present: conversational first-person prose with clear framing ("let's talk for a minute about the framing"). - Improve: define TTRPG, NPC and PC for non-gamers. #### 🧩 Kinesthetic learners - Present: describes the workflow in enough detail to replicate. - Improve: the exact prompts used for one song. #### 🧠 Logical learners - Present: a five-item list of explanations. - Improve: weigh the explanations against each other rather than just listing them. #### 🧑‍🤝‍🧑 Social learners - Present: none. There's no comment prompt in the captured text. - Improve: an explicit question to readers. #### 🧘‍♀️ Solitary learners - Present: the whole post is reflective self-examination, which models the practice. - Improve: not applicable. It already works as a reflection model. #### 🌿 Naturalistic learners - Present: none. - Improve: not applicable to the subject. ## 📊 Accessibility Assessment #### Screen reader semantics - Present: the five explanations are a real list (`<li>`), and emphasis uses `<em>`/`<strong>`. - Improve: the header image's alt text wasn't captured, so it can't be verified. #### Keyboard navigation - Not assessable from captured content. #### Contrast and readability - Not assessable from captured content. #### Text resizing and reflow - Not assessable from captured content. #### Media and captions - The album is an external Suno link. No lyrics text or transcript sits alongside it in the post. - Improve: publish lyrics for at least the referenced tracks. #### Forms and interactions - Not applicable. #### Standards reference - WCAG 2.2 AA: 1.1.1 Non-text Content (header image alt text, unverified). 1.2.1 Audio-only (Prerecorded) applies if the album is ever embedded without a text alternative. ## 🌍 DEIB Assessment #### Cultural breadth - Present: references are US/UK pop culture (Marilyn Manson, Lestat, *Severance*, John Williams, Danny Elfman). - Improve: not applicable. This is a personal post and doesn't need to claim breadth. #### Inclusive language - Present: plain language with light profanity ("pretty shitty ideas"). - Improve: define TTRPG jargon for outsiders. #### Neurodiversity support - Present: short paragraphs and a list break up the argument. - Improve: a one-line takeaway near the top. #### Representation - Present: only the author's own perspective. No musician or composer voice appears. - Improve: a working musician's response would test the counterfactual argument. #### Psychological safety - Present: the author models vulnerability, admitting enjoyment and discomfort together, which invites honest discussion rather than camp-taking. - Improve: not applicable. ## ✅ Logical Validity #### Strong reasoning - **Self-limiting argument (studio paragraph).** The author tests their own "no lost commission" defense against a case where it fails, and concedes it. That's strong: the conclusion is scoped to where the premise holds. - **Competing explanations (the five-item list).** The author offers several alternative causes for the liking instead of one comforting story, which is sound abductive practice. - **Precise diagnosis of loss.** Identifying the lost *conversation* with a composer, not just lost money, sharpens the harm beyond economics. #### Fallacies - **Relative privation / counterfactual excuse (paragraph on paying artists).** "I wouldn't have paid an artist either… So, for me, AI isn't stealing someone's paycheck." That answers the payment harm but sets aside the training-data harm raised one paragraph earlier. Fix: address the two harms separately. The use takes no commission, but it does use and pay for a tool built on unconsented work. - **Hasty generalization (lyrics diagnosis).** "I'm pretty sure it's because it doesn't get poetry." That infers a general incapacity from one album made with style prompts that may have asked for spoken-word delivery. Fix: test with prompts that request strict meter and rhyme. - **Slippery slope, mild (closing).** Creativity will be "leeched from us" unless decisions happen now. The mechanism (muscle atrophy) is plausible, but the post asserts it without evidence. Fix: name a decision or a measure of atrophy. #### Tone of voice - Candid, conversational, self-critical ("I can't say made it, but prompted the music into being"), escalating to alarm in the close ("a very scary thought"). ## 🛡️ DARVO Analysis - Deny: none against others. The counterfactual defense partly minimizes the author's own role in the harm the post names, but the author then undercuts it. - Attack: a brief criticism of the Suno founder's views on artists. That's criticism of a position, not an attack on an accuser. - Reverse victim/offender: none. - Not applicable in the DARVO sense. There's no accusation or conflict between parties. ## 🎯 Conclusion - Bottom line: an honest practitioner account showing that consumer AI music now clears the "I love this" bar for personal, context-rich uses. The author's own defense ("I wouldn't have paid anyone") is sound for a hobby and fails for studios, and the post says so. The unresolved part is the call to "make decisions now," which names no decision. - Next steps: listen to the linked album with the post in mind. Read Rogers and Ogas on what they mean by authenticity. Draft your own line between acceptable hobby use and commercial substitution for AI-generated media on your sites. ## 🔄 Refactor Appendix — We are in the Uncanny Valley > Density: 2–5 Literature atoms (H3), 0–3 Permanent candidates (H4). > Defaults: 3 Literature, 1 Permanent. Resonance filter: if you can't > articulate in one line why a claim matters, drop it. ### 📚 Literature Atoms (H3) #### [H1] AI music targets median taste by design > [!quote] ID:rw_01m34ft1cc2a69sskynz272whq > It’s picking the happy path, the middle road, the consensus of what is considered “good”. **Context & use**: - Fifth item in the list of explanations for liking the album. - Explains why generated output feels good without surprising you. **Tags**: #xref #aphorisms **Highlight**: [view highlight](https://read.readwise.io/read/01m34ft1cc2a69sskynz272whq) <!-- atomicity: 1 idea ✓ --> #### [H2] Personal-use AI defense fails at studio scale > [!quote] ID:rw_01m34ft5zdt25rqfte9b52vx51 > if I am a Hollywood studio and I can use an app to compose a movie score **Context & use**: - Second-to-last paragraph. The author tests their own counterfactual defense. - Useful boundary for any AI media policy: hobby versus commercial substitution. **Tags**: #xref #articleseed **Highlight**: [view highlight](https://read.readwise.io/read/01m34ft5zdt25rqfte9b52vx51) <!-- atomicity: 1 idea ✓ --> #### [H3] Emotional response survives the absence of authenticity > [!quote] ID:rw_01m34ft2rygbw8dkg960vhee88 > it is unsettling that I can have as strong an emotional reaction to something with no authenticity **Context & use**: - Paragraph citing *This Is What It Sounds Like*. - Challenges the assumption that knowing something is AI-made dampens the feeling. **Tags**: #xref **Highlight**: [view highlight](https://read.readwise.io/read/01m34ft2rygbw8dkg960vhee88) <!-- atomicity: 1 idea ✓ --> ### 🧠 Permanent Candidates (H4) #### [P1] "I wouldn't have paid anyone" answers only the payment harm **Claim** (one sentence, Courtney's voice): The no-lost-commission defense for AI media settles whether anyone lost a paycheck, but it leaves the question of whose work trained the tool untouched. **Evidence pointer**: [[#H2]] **Scope & conditions**: Holds wherever the tool trained on unconsented work. It weakens for models trained on licensed or public-domain data. **Links** (≥1 with annotation): - [[AI training data consent]] — because extends - [[Counterfactual harm arguments]] — because applies-in AI ethics debates - [[WordPress Photos AI policy]] — because applies-in community media rules <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Check how Rogers & Ogas actually define "authenticity" in *This Is What It Sounds Like* — Goodreads link in the post #fleeting #inbox - [F2] Possible blog angle: hobby vs. production lines for AI media on WordPress sites #fleeting #inbox ### → Things 3 (max 2 actions) - [ ] Listen to the PARIS album alongside the post — https://suno.com/s/GrPOH6Il9JQ0ROlg #things - [ ] Draft personal AI-media use line (hobby vs. shipped) for courtneyr.dev — blog policy #things <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/We Are in the Uncanny Valley.md|We Are in the Uncanny Valley by Chris Reynolds]] ## Highlights - I also want to say that I believe in paying artists what they are worth, and I understand using AI for creative endeavors is potentially taking that away. I’ve always said about my own use of AI for art for my TTRPG games that *I wouldn’t have paid an artist either*, I probably would have tried to find some existing image on the internet to use, whether or not *that* usage was strictly allowed. So, for me, AI isn’t stealing someone’s paycheck. ([View Highlight](https://read.readwise.io/read/01m34ft46n0wsyz3yq4rhgq6p2)) - Note: Counterfactual-substitution argument: no lost sale, so no harm. It sets the payment question aside, though, not the training-data question the previous paragraph raised. - Tags: atom/literature/h3, fallacy/relative-privation - So here’s the part that’s disturbing: **the [album](https://suno.com/s/GrPOH6Il9JQ0ROlg) is really good.** Like, I’ve been listening to it all weekend and singing along with the lyrics and grooving to it. It also is really good from the perspective of an artifact for the campaign — it tells a different story to players who are listening to the lyrics and enhances the world setting. ([View Highlight](https://read.readwise.io/read/01m34ft01bgjy2nvkmxhrr0hab)) - Note: The turn of the essay: the output's quality is the problem, not a defect. - Tags: atom/literature/h1 - - that the music itself is risk-averse. It’s creating songs based on some sort of median for enjoyment. It’s picking the happy path, the middle road, the consensus of what is considered “good”. It’s literally *designed* for me to like it. ([View Highlight](https://read.readwise.io/read/01m34ft1cc2a69sskynz272whq)) - Note: Median-taste thesis: generative output optimizes for consensus enjoyment, which explains why it lands without being surprising. - Tags: atom/literature/h1, atom/zettel/z1 - Having recently read *[This is What it Sounds Like](https://www.goodreads.com/book/show/60165381-this-is-what-it-sounds-like)* about the specific things that we connect to in music, and one of those things being *authenticity*, it is unsettling that I can have as strong an emotional reaction to something with *no authenticity* (because it was not created by a human) as I do with something fully authentic. And I’ve been trying to figure that out. But the other thing that occurs to me is a lot starker and more disturbing and upsetting. ([View Highlight](https://read.readwise.io/read/01m34ft2rygbw8dkg960vhee88)) - Note: Authenticity as a listed ingredient of musical connection, tested against the author's own reaction. The author treats "authentic" as meaning "human-made", which the book may define differently. - Tags: atom/literature/h2 - I said that I wouldn’t commission an artist to write an album for me. That’s true, I wouldn’t. As a consumer-grade product, I can create music for me, that I like, and maybe I will choose to listen to that over real artists sometimes. But I immediately think of the trend in Hollywood to use AI whenever possible — for actors, for scripts, for everything. And for the most part, you can’t replace the humans. But if I am a Hollywood studio and I can use an app to compose a *movie score* based on the very exacting standards that I have for sound and mood and tone and not have to commission John Williams or Danny Elfman and work with them individually as humans (and have *conversations* with them and have to accept their ideas and creativity) and save a *ton* of money…I can see a world where that could happen. ([View Highlight](https://read.readwise.io/read/01m34ft5zdt25rqfte9b52vx51)) - Note: The author's own counterfactual argument breaks at studio scale, where there was a budget line for a composer. - Tags: atom/literature/h2 - We’re here. This is it. This is where we figure out what we’re doing with this technology. Because if we don’t make decisions about it now, our own creativity is going to be leeched from us individually as we just hand over that muscle to machines that can, apparently, do it faster and better and feed us stuff that’s specifically tailored to be exactly what we want. And that’s a very scary thought. ([View Highlight](https://read.readwise.io/read/01m34ft7cn61sw7tmamm35f3ha)) - Note: Creativity-as-muscle: atrophy through disuse. Conclusion is a call to decide, with no proposed decision. - Tags: atom/literature/h3