Intro
L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
AI as Your Creative Partner · Introduction

Every Creative Era Begins With a Tool That Feels Like Cheating

Why AI in creative work is not unprecedented — and why that history is the most useful thing you can know right now.

In 1888, when George Eastman introduced the Kodak No. 1 camera at $25 and the slogan "You press the button, we do the rest," professional portrait photographers were furious. They had spent years mastering wet-plate collodion chemistry, exposure calculations, and darkroom technique. Here was a device that handed the act of image-making to anyone willing to wind a key. The Photographic Society of London debated whether photography done by machine could even be called art. Within two decades the argument had dissolved — not because the camera won, but because photographers discovered that the machine freed them from the mechanics of capture so they could concentrate entirely on light, composition, and meaning. Alfred Stieglitz was making photographs that hung in the Metropolitan Museum by 1910.

That pattern — a tool arrives, a profession panics, the tool gets absorbed and the work deepens — has repeated with the electric guitar in the 1930s, magnetic tape editing in the 1950s, the synthesizer in the 1970s, desktop publishing in the 1980s, and digital audio workstations in the 1990s. Each time, the objection was that the machine was doing what the artist should be doing. Each time, artists who engaged with the tool early set the terms for everyone who followed. Brian Eno built an entire compositional philosophy around the limitations of the early synthesizer. Trent Reznor used Pro Tools to make Nine Inch Nails records that could not have existed on tape.

This course is about AI in creative work — writing, visual art, music, design, and the creative decisions embedded in almost every professional role. We will not pretend AI is neutral, limitless, or simple. We will examine what it actually does technically, where it genuinely helps, where it reliably fails, and how working practitioners are integrating it today. By the end you will have a clear framework for deciding when to reach for an AI tool and when to put it down — and you will have practiced both.

If you finish every module, here's who you become:

  • You'll understand technically what AI tools actually do when they generate text, images, or audio — not magic, not mystery.
  • You will recognize the historical pattern of tool adoption well enough to place your own moment inside it.
  • You'll be able to use AI to break a creative block, draft, iterate, and still hear your own voice in the finished work.
  • You will apply a clear framework for deciding when an AI tool sharpens your work and when it flattens it into something generic.
  • You'll know how to protect your authorship, credit, and intellectual property in a landscape where those lines are actively contested.
  • You will ship a real creative project that used AI where it helped and didn't where it wouldn't — and you'll be able to explain every choice.
  • You are becoming the kind of practitioner who sets the terms for how AI enters your field, rather than waiting to inherit someone else's terms.
AI as Your Creative Partner · Lesson 1

The Prediction Engine: What AI Is Actually Doing

Understanding the mechanism behind the output — probability, not perception.
If AI has no understanding, no taste, and no intention — what exactly is it doing when it writes a poem or designs a logo?

In January 2023, The Atlantic published a piece by staff writer Stephen Marche documenting his experiment using GPT-4 to co-write a short story. Marche, a novelist, spent three weeks feeding the model fragments, revising its output, and steering it away from clichés. His conclusion was precise: the model was "an extraordinarily sophisticated autocomplete." It never surprised him the way a collaborator surprises you — by bringing something it had genuinely experienced. But it also never got tired, never needed encouragement, and never forgot a constraint he'd specified. What he had was not a co-author. It was something harder to name: a system that had processed more narrative structure than any human could read in a lifetime, and that surfaced patterns from that structure on demand.

That description — patterns surfaced on demand — is the most accurate plain-English account of what a large language model does. The technical name is next-token prediction. The creative implications take a full course to unpack. This lesson starts with the mechanism.

How Large Language Models Generate Text

A large language model (LLM) is trained on a massive corpus of text — in GPT-4's case, an estimated several hundred billion words drawn from books, websites, academic papers, code repositories, and digitized human writing going back decades. During training, the model learns one thing obsessively: given a sequence of words (or sub-word tokens), what token is most likely to come next?

This is not storage and retrieval. The model does not keep a copy of its training data. Instead it compresses statistical relationships between tokens into billions of numerical weights distributed across a neural network architecture called a transformer. When you prompt the model, it uses those weights to compute a probability distribution over its entire vocabulary — roughly 50,000 tokens for most current models — and samples from that distribution. It then appends the chosen token and repeats the process. A 500-word response involves approximately 650 individual prediction steps.

What this means for creative work: the model is not imagining, intending, or evaluating. It is sampling from learned distributions. High-probability outputs tend to be familiar and fluent. Low-probability samples (controlled by a parameter called "temperature") can be surprising but are also more likely to be incoherent. The creative practitioner's job is to understand this mechanism well enough to work with its tendencies rather than against them.

Technical Detail

The "temperature" parameter — available in most AI APIs and some consumer interfaces — controls how sharply the probability distribution is peaked. Temperature 0 always picks the highest-probability token (deterministic, repetitive). Temperature 1 samples proportionally (default behavior). Temperature above 1 flattens the distribution, increasing randomness. Most creative writing tasks benefit from temperatures between 0.7 and 1.1.

Image and Audio Models: The Same Logic, Different Domains

Text models predict the next token. Image generation models — Midjourney, DALL-E 3, Stable Diffusion — use a related but distinct mechanism called diffusion. During training, the model learns to reverse a process of adding random noise to images. At inference time, it starts from pure noise and iteratively denoises, guided by a text prompt, until a coherent image emerges. The guidance comes from a text encoder (often CLIP, developed by OpenAI and published in January 2021) that has learned relationships between text descriptions and image features across hundreds of millions of image-caption pairs.

Audio models like Suno (launched publicly in 2023) and MusicLM (Google, published February 2023) apply similar principles to spectrograms — visual representations of sound. The model learns what combinations of frequency, rhythm, and timbre patterns co-occur with text descriptions of musical genres, moods, and instruments.

In every case, the underlying operation is the same: learned statistical associations between human-generated outputs and the conditions that accompanied them. There is no listening, no seeing, no reading in any phenomenological sense. There is pattern matching at extraordinary scale.

Why This Matters Practically

Because these systems are pattern-matchers, they excel at tasks that have clear patterns in their training data — genre fiction conventions, standard logo layouts, common chord progressions — and struggle with tasks that require genuine novelty, factual accuracy outside their training window, or understanding of context that exists only in the real world. Knowing this lets you assign tasks intelligently.

Key Terms

TokenThe basic unit of text a language model processes — roughly 3/4 of a word on average. "Creativity" is one token; "antidisestablishmentarianism" is four.
TransformerThe neural network architecture introduced by Google researchers in the 2017 paper "Attention Is All You Need" that underlies virtually all current large language models.
Diffusion modelAn image generation architecture that learns to reverse the process of adding noise to images. Stable Diffusion (released August 2022), DALL-E 3, and Midjourney all use variants of this approach.
TemperatureA parameter controlling randomness in token sampling. Lower values produce more predictable output; higher values produce more varied (and potentially more creative or more incoherent) output.
HallucinationWhen a language model generates confident-sounding but factually incorrect output — a direct consequence of the prediction mechanism, which optimizes for fluency, not truth.

Lesson 1 Quiz

Five questions on the prediction engine and what AI is actually doing.
1. What is the core operation a large language model performs when generating text?
Correct. LLMs perform next-token prediction — sampling from a probability distribution over vocabulary tokens, one step at a time.
Not quite. LLMs do not store or retrieve sentences. They compress statistical relationships into numerical weights and sample from learned distributions at inference time.
2. What does a higher temperature setting do in a language model?
Correct. Higher temperature flattens the probability distribution so lower-probability tokens get sampled more often — producing more varied, sometimes more creative, sometimes more incoherent output.
Not quite. Temperature controls the shape of the probability distribution over tokens, not processing speed, data access, or restriction.
3. Image generation models like Stable Diffusion primarily use which mechanism?
Correct. Diffusion models learn to reverse a noise-adding process. At inference time they start from noise and iteratively denoise, steered by a text encoder that relates language to visual features.
Not quite. Stable Diffusion and similar models use diffusion — a process of iterative denoising from random noise, guided by learned text-image relationships.
4. "Hallucination" in AI language models refers to:
Correct. Hallucination is a direct consequence of the prediction mechanism: the model is trained to produce fluent, plausible-sounding text, not to verify factual accuracy.
Not quite. Hallucination refers specifically to factually incorrect but confidently stated output — a structural consequence of training on fluency rather than truth.
5. According to the lesson, AI creative tools are most reliable for tasks that:
Correct. Because AI systems are pattern-matchers, they perform best on tasks whose patterns appear frequently in training data — genre fiction conventions, standard design layouts, familiar musical forms.
Not quite. Pattern-matching systems perform best on well-patterned tasks. They struggle with genuine novelty, real-time facts, and context that exists only outside their training window.

Lab 1: Interrogating the Prediction Engine

Ask the AI to explain what it's doing — then probe whether that explanation holds up.

Your Task

In this lab you'll have a conversation with an AI assistant about the mechanics of AI text generation. The goal is not to receive a lecture — it is to test the boundaries of the explanation. Ask it to describe what it does when it generates a creative sentence. Then push back: ask whether it "understands" what it wrote, whether it could be wrong, and how temperature would change the output. Aim for at least 3 exchanges.

Suggested opener: "Can you walk me through what you're actually doing, mechanically, when you generate the first sentence of a short story? Be honest about what you don't do as well as what you do."
AI Lab Assistant
Lesson 1 · Prediction Engine
Welcome to Lab 1. I'm here to help you interrogate how AI text generation actually works — including my own limitations and the places where the standard explanation gets fuzzy. What would you like to start with?
AI as Your Creative Partner · Lesson 2

Where AI Genuinely Helps: The Documented Cases

What practitioners have actually found useful — drawn from published accounts, production records, and verifiable experiments.
What does the evidence actually show about where AI improves creative output, and where it demonstrably does not?

In 2022, Cosmopolitan magazine published what it described as the first AI-generated magazine cover, produced using DALL-E 2 in collaboration with digital artist Karen X. Cheng. The cover — depicting an astronaut on the moon — took Cheng about two days of iterative prompting and manual refinement in Photoshop. The image itself was striking; the process was labor-intensive in ways the press coverage largely glossed over. Cheng's own account, published on LinkedIn, described spending hours crafting precise prompts, discarding dozens of unsatisfactory generations, and spending significant time in post-production correcting anatomical errors in the astronaut's glove. The story of that cover is not "AI made a magazine cover." It is: a skilled artist used AI to access a visual style she couldn't have produced as quickly with traditional tools, while providing all the judgment the tool cannot.

Documented Creative Productivity Gains

A September 2023 study by researchers at MIT and Stanford (published in Science, lead author Shakked Noy) measured the effect of access to ChatGPT on professional writing tasks including cover letters, short reports, and analytical memos. Workers with AI access completed tasks 37% faster and received quality ratings 18% higher from blind evaluators. The gains were concentrated among workers who started at lower performance levels — the AI raised the floor of the distribution more than it raised the ceiling. High performers improved modestly; lower performers improved substantially.

A separate 2023 study from Boston Consulting Group (published in Harvard Business Review, September 2023) tested GPT-4 against consultants on creative and analytical tasks. On tasks within the model's "frontier" — brainstorming, writing, synthesizing — AI-assisted consultants outperformed unassisted consultants by 40% on quality. On tasks outside that frontier — causal reasoning, real-world judgment calls — AI assistance made performance worse, not better. Consultants who trusted the AI uncritically on out-of-frontier tasks performed below the control group.

What "Better" Actually Measured

In the BCG study, "quality" was evaluated by experienced consultants who did not know which documents were AI-assisted. They rated documents on clarity, persuasiveness, and originality of ideas. AI-assisted work scored higher on the first two dimensions. On originality, the gap narrowed. The evaluators noted that AI-assisted work tended toward "competent but predictable" framings on novel problems.

Where the Evidence Is Weaker

The enthusiasm in the productivity literature collides with a different body of evidence from creative practitioners. In a 2023 survey of 1,000 professional writers conducted by the Authors Guild, 90% said they were concerned about AI's impact on their livelihood; fewer than 10% reported that AI tools had improved the quality of their published work (as distinct from their speed of drafting). The discrepancy between "faster drafting" and "better published work" is significant: speed in early-stage drafting may not translate to quality in the finished artifact that readers encounter.

In music, Suno and Udio (both launched publicly in early 2024) can generate full songs from text prompts in seconds. What they cannot yet do is generate a song that grows from a specific lived experience — the structural quality that distinguishes a Townes Van Zandt lyric from a competent genre exercise. The pattern-matching mechanism produces music that sounds like the genre it was prompted toward. It cannot produce music that sounds like a specific human life.

The honest summary: AI reliably helps with volume, variation, and the mechanics of production. It is unreliable for tasks requiring originality in the deepest sense — work that could only come from a specific person's specific experience.

The Practitioner's Framework

Use AI for: first drafts, variation generation, style exploration, ideation at scale, mechanical production tasks (transcription, reformatting, summarization). Reserve human judgment for: final selection, voice consistency, factual accuracy, and any work where provenance — who made it, out of what experience — is part of the value.

Specific Documented Use Cases

Screenwriting: In 2023, the WGA strike foregrounded AI's specific threat to TV writers' rooms — particularly the practice of using AI to generate first-draft "outlines" that human writers would then be paid less to polish. The WGA contract reached in September 2023 prohibited studios from requiring writers to use AI and from using AI-generated material as a basis for setting compensation. This is a documented instance of AI affecting not just the work but the economic structure around the work.

Visual design: Adobe introduced Generative Fill in Photoshop in May 2023, allowing designers to extend images, remove objects, or generate background fills using Firefly, Adobe's in-house diffusion model trained on licensed images. Within six months, Adobe reported that Firefly had generated over 3 billion images. Professional designers reported using Generative Fill primarily for comp mockups and background extension — tasks previously requiring either stock photography licensing or manual retouching.

Marketing copy: Jasper AI, founded in 2021, raised $125 million in October 2022 at a $1.5 billion valuation based primarily on enterprise adoption for marketing copy generation. CMOs at companies including Levi's and HubSpot publicly described using AI to generate hundreds of copy variations for A/B testing — a use case where volume and variation matter more than the singular authorial voice that distinguishes literary writing.

Lesson 2 Quiz

Five questions on documented evidence for AI's creative applications.
1. The 2023 MIT/Stanford study (Noy et al.) found that AI access improved professional writing task completion time by approximately:
Correct. The Noy et al. study found tasks completed 37% faster, with quality ratings 18% higher from blind evaluators.
Not quite. The Noy et al. study measured approximately 37% faster task completion, with the greatest gains among lower-performing workers.
2. In the Boston Consulting Group study on GPT-4 and consultants, what happened when AI was applied to tasks "outside the frontier" of AI capability?
Correct. The BCG study found that on tasks requiring causal reasoning and real-world judgment, uncritical AI use produced worse results than no AI use at all.
Not quite. The BCG study found that AI assistance on out-of-frontier tasks actively degraded performance — consultants who trusted the AI scored below the unassisted control group.
3. The 2023 WGA contract (Writers Guild of America) addressed AI by:
Correct. The September 2023 WGA contract prohibited studios from requiring writers to use AI tools and from using AI-generated material as the baseline for compensation negotiations.
Not quite. The WGA contract specifically prohibited studios from requiring AI use and from leveraging AI output to depress writer pay — it did not ban AI entirely or create joint credits.
4. Adobe Firefly's Generative Fill, launched in Photoshop in May 2023, was reported by professional designers to be most useful for:
Correct. Designers reported using Generative Fill primarily for comping and background work — high-effort mechanical tasks — rather than for final creative output.
Not quite. Professional designers reported using Generative Fill most for background extension, comp mockups, and object removal — productive mechanical tasks, not autonomous final design.
5. Based on the documented evidence in Lesson 2, AI creative tools most reliably add value in which of the following scenarios?
Correct. Volume, variation, and mechanical production tasks are where AI adds documented value. Personal memoir, unique emotional expression, and causal reasoning are outside the frontier where AI reliably helps.
Not quite. The evidence shows AI is most reliable for high-volume, high-variation tasks like copy generation — not for work requiring personal provenance, unique emotional specificity, or causal reasoning.

Lab 2: Testing the Frontier

Give AI tasks on both sides of its capability frontier and compare the results.

Your Task

This lab asks you to test the BCG study's finding about the "frontier" — the boundary between tasks where AI helps and tasks where it doesn't. Have the AI assist you with a task clearly within its frontier (generating marketing headline variations for a product you describe), then give it a task outside the frontier (explaining why a specific real creative decision was made by a real artist). Compare the quality and confidence of responses.

Suggested opener: "I want to test where you're actually useful. First: give me 10 headline variations for a new brand of noise-canceling headphones aimed at remote workers. Then I'll give you the harder task."
AI Lab Assistant
Lesson 2 · The Frontier
Ready to test the frontier. Start with the in-frontier task — marketing headlines — and then we'll move to something more demanding. I'll try to be honest about where my confidence drops.
AI as Your Creative Partner · Lesson 3

Prompting as Creative Direction

The craft of writing prompts is the craft of communicating intent — and it has more in common with directing than with coding.
Why does the same AI produce wildly different outputs from prompts that seem to ask for the same thing — and what does that tell you about how to write prompts?

When Refik Anadol created "Unsupervised" for the Museum of Modern Art in New York — installed in January 2023 — he and his studio spent months engineering the prompts and parameters that would guide the piece's real-time AI generation of visual forms. The work runs on a custom machine learning system trained on MoMA's collection of 200 years of art. Anadol described his process in a February 2023 interview with The New York Times as closer to "conducting an orchestra" than to making images: he was specifying mood, rhythm, and transformation rules, not individual outputs. The installation required the same kind of sustained intentional work that any large-scale artwork requires. The fact that a machine was executing the individual frames did not reduce the directorial labor — it transferred it to the specification stage.

That transfer — from execution to specification — is what makes prompting a genuine creative skill, not a technical shortcut.

The Anatomy of an Effective Prompt

Research on prompt engineering converged on a set of components that consistently improve output quality. A 2022 paper from Google DeepMind (Wei et al., "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models") demonstrated that simply asking a model to "think step by step" before answering significantly improved performance on reasoning tasks. This was not a magic phrase — it was an instruction that changes what the model treats as the relevant prior context for each subsequent token.

For creative tasks, the components of an effective prompt are:

Role: What perspective or expertise should the model adopt? ("Write as a crime fiction editor reviewing a first chapter" produces different output than "Write as a crime fiction fan.")

Task: What is the specific output format and scope? ("Write a 200-word pitch paragraph for the following concept" is more useful than "Write about this concept.")

Constraints: What should be excluded or bounded? Constraints are often more valuable than positive specifications because they prevent the model from defaulting to high-probability generic patterns. ("Avoid clichés about isolation. Do not use the word 'journey'." is actionable in a way that "Be original" is not.)

Examples: Providing one or two examples of the target style — called "few-shot prompting" — is the single most reliable technique for aligning output with a specific voice or format. OpenAI's own documentation consistently ranks few-shot prompting as the highest-leverage prompting technique for style replication.

Evaluation criteria: Telling the model how the output will be judged ("The reader is a skeptical editor who has seen 500 pitches this year") shapes the probability distribution toward outputs that would satisfy that evaluator.

Documented Technique: Constraint-First Prompting

In a 2023 experiment documented by writer and educator Ethan Mollick (Wharton School, University of Pennsylvania) in his Substack "One Useful Thing," Mollick found that prompts that specified what NOT to do before specifying the task produced output rated more original by blind reviewers than prompts that only specified what to do. The mechanism: negative constraints push the model away from its highest-probability (most generic) outputs.

Iteration: The Prompt Is Not the Product

One of the most common errors in AI-assisted creative work is treating the first output as the output. In reality, the first generation is a starting position — a rough draft of a draft. The workflow that produces usable creative work almost always involves multiple rounds of prompting, each one informed by what the previous round revealed about the model's tendencies and limitations.

In September 2023, designer and researcher Linus Ekenstam published a detailed account of building an illustrated children's book using Midjourney v5.2 and GPT-4 for text. The project took approximately 80 hours across three weeks. Ekenstam documented 340 distinct image generations, of which 23 were used. The ratio — about 7% — is consistent with reports from other practitioners using diffusion models for professional work. The creative work was almost entirely in the curation and iteration, not in the initial prompts.

This is not a criticism of the tools. It is an accurate description of how creative work with any tool actually functions. A photographer on a professional shoot might take 800 frames and use 12. The ratio is not evidence of inefficiency; it is evidence of the selective judgment that distinguishes professional from amateur work.

Directing vs. Generating

The skill shift AI creates is from generating to directing. A skilled prompt engineer is doing what a film director does with a cinematographer — specifying intent, evaluating output, redirecting when the output misses the intent, and maintaining a vision across many iterations. The model is the cinematographer: highly capable at execution within a brief, unable to supply the vision itself.

Common Prompting Errors and How to Fix Them

Too vague"Write something creative about loss." Fix: Specify form, length, perspective, tone, and one concrete constraint. "Write a 150-word prose poem from the perspective of an empty parking space outside a hospital. Avoid sentimentality."
No examplesAsking for a specific style without showing it. Fix: Paste 2–3 sentences in the target voice before the instruction. "In the style of the following sentences: [examples]. Now write a product description for…"
Asking for quality"Make it better." or "Make it more creative." These are not instructions the model can act on. Fix: Be specific about what dimension to improve. "Tighten the opening sentence. Replace the abstract noun in the second paragraph with a concrete physical object."
Single-shot thinkingExpecting the first output to be usable. Fix: Plan for at least 3–5 iteration rounds. Treat generation 1 as raw material, not a draft.

Lesson 3 Quiz

Five questions on prompting as creative direction and the craft of effective prompts.
1. "Few-shot prompting" refers to:
Correct. Few-shot prompting means including examples of the desired output style within the prompt — ranked by OpenAI as the highest-leverage technique for style alignment.
Not quite. Few-shot prompting means including example outputs within the prompt itself to show the model what target style or format looks like.
2. Based on Ethan Mollick's documented experiment, what type of constraint in a prompt tends to produce output rated more original by blind reviewers?
Correct. Mollick's experiment found that constraint-first prompting — specifying what NOT to do — pushed the model away from high-probability generic outputs and produced work rated more original.
Not quite. Mollick found that negative constraints (what NOT to do) outperformed positive specifications for originality, because they push the model away from its most probable — most generic — outputs.
3. Refik Anadol's "Unsupervised" installation at MoMA (January 2023) illustrates which principle about AI creative work?
Correct. Anadol described his process as "conducting an orchestra" — sustained directorial work at the specification level, not reduced creative labor.
Not quite. Anadol's account emphasizes that AI-assisted work transferred the creative labor to specification and direction — months of work determining mood, rhythm, and transformation rules.
4. When a practitioner asks AI to "make it more creative," what is the core problem with that instruction?
Correct. Abstract quality instructions like "be more creative" give the model no concrete criteria to act on. Specific instructions about which element to change and in what direction are actionable.
Not quite. The problem is specificity: "more creative" doesn't tell the model which dimension to alter. Specific instructions — "replace the abstract noun with a concrete object" — are actionable; vague quality requests are not.
5. Linus Ekenstam's documented children's book project using Midjourney found that approximately what percentage of generated images were used in the final book?
Correct. Ekenstam used 23 of approximately 340 generated images — about 7% — consistent with professional photography ratios and illustrating that creative work with AI is primarily a curation and iteration process.
Not quite. Ekenstam used 23 of 340 generations — about 7%. This ratio illustrates that the skill in AI-assisted image work is almost entirely in iteration and curation, not in the initial generation.

Lab 3: Prompt Engineering Practice

Build and iterate a prompt using role, task, constraints, and examples.

Your Task

You'll practice the components of an effective prompt by building one iteratively. Start with a vague prompt for a short piece of creative writing. Then rebuild it with: a specific role, a concrete task, at least one negative constraint, and one example sentence. Compare the outputs. Aim for at least 3 exchanges.

Suggested opener: "First, here's my vague prompt — just so we have a baseline: 'Write something about a city at dawn.' Give me a result. Then I'll tell you how I'm going to upgrade the prompt."
AI Lab Assistant
Lesson 3 · Prompt Engineering
Let's build prompts together. Start with your vague baseline prompt and I'll show you what that produces — then we'll upgrade it step by step using role, task, constraints, and examples. I'll also explain what changed between versions.
AI as Your Creative Partner · Lesson 4

Authorship, Attribution, and What You Actually Own

The legal, ethical, and creative questions about who made what — and why they matter before you publish.
When AI generates a significant portion of a creative work, who is the author — and does the answer change anything about how you should work?

In February 2023, the US Copyright Office issued a partial registration for the graphic novel "Zarya of the Dawn" by Kristina Kashtanova — and simultaneously revoked copyright protection for the AI-generated images within it, retaining protection only for Kashtanova's text and the selection and arrangement of the images. The Office's reasoning was precise: copyright requires human authorship; the images were generated by Midjourney based on Kashtanova's prompts, but the Office found that prompting did not constitute "sufficient creative control" over the specific expressive elements in each image to qualify as authorship. This was not a judgment about whether the work was good. It was a determination about the legal structure of who made what.

That determination established a precedent that has shaped every subsequent copyright discussion in the US about AI-generated creative work: prompting alone does not establish authorship. The human contribution must be specific, intentional, and traceable to particular expressive elements in the output.

The Current Legal Landscape

As of 2024, the US Copyright Office's position is that works generated entirely by AI are not eligible for copyright protection. Works in which a human makes "sufficient creative contributions" may receive copyright protection for those human-contributed elements. The Office has not defined a clear threshold for what constitutes "sufficient" — it is evaluating cases individually.

The European Union's AI Act (passed March 2024) includes disclosure requirements for AI-generated content but does not directly address copyright ownership. The UK Intellectual Property Office concluded a consultation in 2023 and has not changed its existing position that computer-generated works (without a human author) receive a separate, shorter protection period of 50 years — a position that predates generative AI by decades, from the Copyright, Designs and Patents Act of 1988.

In the training data question — whether AI companies infringed copyright by training on human-created work without license — multiple lawsuits are pending as of this writing. The New York Times filed suit against OpenAI and Microsoft in December 2023, alleging that GPT models were trained on Times articles. Getty Images filed suit against Stability AI in January 2023. These cases have not reached final verdicts and the legal landscape is actively evolving.

Practical Implication

If you intend to register copyright for AI-assisted work, document your human creative decisions at each stage. The Copyright Office's guidance suggests that the more you can demonstrate specific creative choices — selecting, arranging, editing, and modifying AI output — the stronger your claim. Work that amounts to "I typed a prompt and used the first result" is unlikely to receive protection.

Ethical Dimensions Beyond the Legal Ones

The legal question of copyright protection is distinct from the ethical question of attribution. Several professional communities have established disclosure norms that go beyond what the law requires. The journal Science announced in January 2023 that AI-generated text cannot be listed as an author and must be disclosed in the methods section. Nature published a similar policy. The Associated Press style guide was updated in 2023 to require disclosure of AI use in news content.

In literary publishing, the situation is less settled. After a flood of AI-generated submissions in early 2023, the science fiction magazine Clarkesworld temporarily closed submissions in February 2023 — editor Neil Clarke documented receiving hundreds of submissions he identified as AI-generated within a short period, overwhelming the slush pile. Many literary magazines updated their submission guidelines to require disclosure or to prohibit AI-generated work. The science fiction professional organization SFWA (Science Fiction and Fantasy Writers of America) updated its membership requirements in 2023 to specify that AI-generated work cannot qualify toward membership eligibility.

The ethical principle underlying most of these policies: audiences, editors, and collaborators are making decisions based on an implicit understanding that the work represents human creative effort. Concealing AI generation violates that understanding regardless of whether it violates a law.

The Creative Case for Disclosure

Beyond legal and ethical compliance, there is a pragmatic creative argument for transparency: audiences who know a work is AI-assisted evaluate it on different terms — and that can be advantageous. Refik Anadol's MoMA installation was celebrated partly because its AI process was central to its meaning. Karen X. Cheng's Cosmopolitan cover generated significant coverage because it was openly described as an AI experiment. Concealment forecloses the possibility of the process being part of the work's meaning.

Developing a Personal Authorship Practice

The question "who is the author?" becomes most important when you answer it honestly for yourself, before any legal or ethical obligation compels you. Three questions worth asking before publishing AI-assisted creative work:

1. What did I contribute that the AI could not have contributed without me? Specific lived experience, a particular editorial perspective, a selection judgment rooted in knowledge the model doesn't have — these are human contributions. Rephrasing a prompt until you liked the output is a thin contribution.

2. Would the people encountering this work make different decisions if they knew AI was involved? If yes — if an editor would reconsider, if an audience would feel deceived, if a collaborator would object — that is a signal that disclosure is appropriate regardless of the legal floor.

3. Does the process reflect the values I want my creative work to represent? This is a question about craft identity, not legal compliance. Some practitioners are entirely comfortable with AI as a tool, disclosed or not. Others find that extensive AI generation undermines something they value about making work by hand. Both positions are coherent. What is not coherent is avoiding the question.

Lesson 4 Quiz

Five questions on authorship, copyright, and attribution in AI-assisted creative work.
1. In its February 2023 decision about "Zarya of the Dawn," the US Copyright Office determined that:
Correct. The Office gave a partial registration — protecting human contributions (text, selection, arrangement) while revoking protection for the Midjourney-generated images themselves.
Not quite. The Copyright Office issued a partial registration — protecting Kashtanova's text and arrangement choices but revoking protection for the AI-generated images, finding that prompting alone did not constitute sufficient authorship.
2. What is the US Copyright Office's current position on works generated entirely by AI?
Correct. The US Copyright Office's current position is that copyright requires human authorship — works generated entirely by AI without sufficient human creative contribution are ineligible.
Not quite. The US Copyright Office holds that copyright requires human authorship. Entirely AI-generated works do not qualify — no protection is granted, not even a reduced term.
3. Why did Clarkesworld science fiction magazine temporarily close submissions in February 2023?
Correct. Editor Neil Clarke documented receiving hundreds of submissions he identified as AI-generated within a short period — the volume made normal operations impossible.
Not quite. Neil Clarke, the editor, publicly documented that a surge of AI-generated submissions had overwhelmed the slush pile, forcing a temporary closure to reset the process.
4. The journals Science and Nature both announced in January 2023 that AI tools:
Correct. Both Science and Nature established that AI cannot be listed as an author and that AI-generated text requires explicit disclosure in the methods section.
Not quite. Both journals announced that AI cannot be listed as an author, and any use of AI-generated text must be disclosed — not banned entirely, but clearly attributed.
5. According to the lesson, which of the following represents the strongest human creative contribution to AI-assisted work?
Correct. The Copyright Office and the lesson both point to traceable, specific human decisions — rooted in knowledge or experience the model lacks — as the meaningful contribution that may support authorship claims.
Not quite. The strongest human contribution is specific, traceable creative judgment — lived experience, editorial selection, and intentional modification of particular expressive elements — not prompt length or iteration count.

Lab 4: Authorship Audit

Examine a piece of AI-assisted work and identify the human contributions that would support an authorship claim.

Your Task

In this lab you'll conduct an authorship audit — a structured examination of who contributed what to a creative work. Describe to the AI a hypothetical project: a short piece of writing or an image created with AI assistance. Then work through three questions: What did the human contribute that was specific and traceable? Would the audience feel deceived if they learned AI was involved? What disclosure, if any, would be appropriate? Aim for at least 3 exchanges.

Suggested opener: "I want to audit an AI-assisted project for authorship. Here's the scenario: a freelance copywriter used GPT-4 to generate 80% of a brand story for a client, then edited it for tone. Help me work through who actually authored what, and what should be disclosed."
AI Lab Assistant
Lesson 4 · Authorship Audit
Let's run an authorship audit. Describe your scenario — what was created, how AI was involved, and what the human did at each stage. I'll help you identify which contributions are specific and traceable, which are thin, and what disclosure standards from the lesson apply.

Module 1 Test

15 questions across all four lessons. Score 80% or higher to pass.
1. The neural network architecture underlying virtually all current large language models is called:
Correct. The transformer architecture, introduced by Vaswani et al. at Google in 2017, underlies GPT, Claude, Gemini, and virtually all major current LLMs.
The transformer architecture — introduced in "Attention Is All You Need" (2017) — is the foundation of virtually all current LLMs.
2. At temperature 0, a language model will:
Correct. Temperature 0 collapses the distribution to always select the single highest-probability token — deterministic, reproducible, and prone to repetition.
Temperature 0 means always picking the top-probability token — deterministic, reproducible output, often formulaic.
3. CLIP, developed by OpenAI and published in January 2021, enables image generation models to:
Correct. CLIP learned relationships between text and image features across hundreds of millions of image-caption pairs, providing the text guidance mechanism in most diffusion models.
CLIP learned text-image relationships across massive datasets, providing the mechanism by which text prompts guide image generation in diffusion models.
4. The Noy et al. MIT/Stanford 2023 study found that AI writing assistance improved quality ratings by approximately:
Correct. Blind evaluators rated AI-assisted writing 18% higher in quality, alongside the 37% speed improvement — with gains concentrated among lower-performing workers.
The Noy et al. study found quality ratings 18% higher from blind evaluators, alongside 37% faster task completion.
5. In the Boston Consulting Group study, what was the specific effect of AI assistance on tasks "outside the frontier" of AI capability?
Correct. Uncritical AI use on out-of-frontier tasks (causal reasoning, real-world judgment) produced results worse than no AI use — a key finding about appropriate tool boundaries.
The BCG study found that AI use on out-of-frontier tasks actively degraded performance below the control group — the most important finding about knowing when not to use AI.
6. Jasper AI, which raised $125 million at a $1.5 billion valuation in October 2022, demonstrated AI's primary commercial value in:
Correct. Jasper's valuation was built on enterprise adoption for high-volume marketing copy — a use case where variation and speed matter more than singular authorial voice.
Jasper's enterprise adoption was driven by marketing copy at scale — the high-volume, high-variation use case where AI's pattern-matching excels.
7. The WGA (Writers Guild of America) contract reached in September 2023 regarding AI:
Correct. The WGA contract focused on the economic threat: preventing studios from using AI to lower the floor of what they paid for human creative work.
The WGA contract prohibited studios from requiring AI use and from using AI-generated material to depress writer compensation — an economic protection, not a creative ban.
8. The "chain-of-thought" prompting technique, documented by Wei et al. at Google DeepMind in 2022, works by:
Correct. "Think step by step" changes the prior context for each token prediction — the intermediate reasoning steps become part of the context window, improving performance on multi-step problems.
Chain-of-thought prompting asks the model to reason step-by-step before answering, which alters the token-prediction context to include intermediate reasoning — improving multi-step performance.
9. Linus Ekenstam's documented children's book project used approximately what ratio of Midjourney generations to final images?
Correct. 23 of 340 generations — about 7% — were used, illustrating that professional AI-assisted image work is primarily a curation and iteration discipline.
Ekenstam used 23 of 340 generations — roughly 7% — consistent with professional photography selection ratios and illustrating that curation is the primary creative skill.
10. 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.
11. The US Copyright Office's February 2023 decision on "Zarya of the Dawn" established that:
Correct. The Office drew a precise line: prompting is not authorship. Specific, traceable human creative decisions about particular expressive elements may support a copyright claim.
The "Zarya" decision established that prompting alone doesn't create authorship — human contributions must be specific and traceable to particular expressive elements in the output.
12. The UK's Copyright, Designs and Patents Act of 1988 is relevant to AI because it:
Correct. The 1988 Act's provisions for computer-generated works — 50-year protection — pre-date generative AI and provide an existing framework the UK is applying to the current situation.
The UK's 1988 Act already addressed computer-generated works without human authors, granting 50-year protection — a pre-existing framework being applied to generative AI now.
13. According to the lesson, what is the key ethical principle underlying disclosure policies at Science, Nature, the AP, and literary magazines?
Correct. The common ethical principle is that the implicit contract between creator and audience assumes human authorship — concealing AI's role violates that contract regardless of legality.
The ethical principle is about the implicit contract: audiences make decisions assuming human creative effort. Concealing AI generation violates that contract regardless of legal requirements.
14. "Hallucination" in AI language models is a direct consequence of:
Correct. Because training rewards producing fluent, probable-sounding sequences — not truthful ones — the model can generate confident-sounding factual claims that are false.
Hallucination is structural: models are trained on fluency and probability, not truth. The result is confident-sounding output that can be factually wrong.
15. According to the lesson's framework, which of these tasks is LEAST suited to AI assistance?
Correct. Work whose value derives from specific lived experience that only one person has — memoir, confessional poetry, testimony — is the case where AI pattern-matching contributes least and can actively undermine the work's core quality.
Memoir rooted in specific lived experience is the least suited to AI assistance — the work's value is precisely its provenance from a particular human life, which AI cannot replicate or substitute.