← Back to Academy
Module 1 · Stories & Creativity with AI — Basic | AESOP AI Academy Module 4
Color
Basic
Module Test
Lesson 1

AI Can Tell Stories

Story generation, prompting for narrative, and the mechanics of AI creativity.

In 2023, the Writers Guild of America went on strike — one of their core demands was protection against studios using AI to generate scripts or to rewrite human-written scripts at lower cost. The writers weren't saying AI couldn't write. They were saying it was writing, and that raised questions about credit, pay, and authorship that nobody had answered yet.

How AI Generates Narrative

Language models generate stories by predicting the statistically likely next token given prior context. They have no intent, no lived experience, and no emotional investment in the narrative. But they have processed enormous amounts of human storytelling — and that shows in the output.

  • Coherence: AI maintains local coherence well (sentence to sentence) but often loses long-range narrative logic.
  • Voice: AI can mimic styles convincingly but produces an average of many voices rather than a distinctive one.
  • Surprise: AI generates statistically likely next steps — which can feel formulaic. Breaking formula requires explicit prompting.
Prompting for Narrative

The quality of AI storytelling correlates directly with prompt quality. Effective narrative prompts specify: protagonist with a concrete goal, an obstacle that creates genuine tension, a constraint that limits easy solutions, and a specific tone or voice.

Key Principle

Vague prompt: "Write a story about a girl." Specific prompt: "Write a story in the voice of a tired 1940s detective about a girl who discovers her school librarian is running a counterfeit book operation. Dry humor, noir tone, first person." The second produces something distinctly different.

Quiz 1

AI Can Tell Stories

4 questions — free, untracked, retake anytime.

is the main limitation of AI-generated narrative compared to human writing?

✓ Correct — ✅ AI maintains local coherence well but often loses long-range narrative logic and defaults to statistically likely (formulaic) choices.
❌ AI generates statistically likely text — which often feels formulaic. It also struggles to maintain long-range story logic.

did the WGA strike include AI as a core concern?

✓ Correct — ✅ Writers weren't saying AI couldn't write — they were saying it was writing, and that raised unresolved questions about authorship, credit, and pay.
❌ The concern was economic and legal: studios could use AI to avoid paying writers, and existing contracts didn't address AI authorship.

makes a narrative prompt more effective?

✓ Correct — ✅ Effective narrative prompts give the AI a protagonist with a concrete goal, an obstacle, a constraint that limits easy solutions, and a specific tone.
❌ The most effective narrative prompts specify: protagonist goal, obstacle, constraint that limits easy solutions, and a specific tone or voice.

voice in storytelling is best described as:

✓ Correct — ✅ AI produces a statistical average of the voices in its training data — convincingly mimicking styles but lacking a distinctive voice of its own.
❌ AI voice is a statistical average of its training data — it can mimic styles but produces something averaged rather than distinctive.
Lab 1

Narrative Prompt Analysis

Analyze and strengthen a story prompt.

Lab 1 — Narrative Prompt Analysis

The AI guide will help you analyze what makes a narrative prompt work and workshop one of your own.

  1. The AI opens with a prompt pair — one vague, one specific — and asks you to analyze the difference.
  2. Write your own narrative prompt for a story idea you have.
  3. The AI will give feedback on how to strengthen it.
Focus on: protagonist goal, obstacle, constraint, tone. These four levers control almost everything.
🔬 AI GuideLab 1
Lesson 2

We Tell Stories Together

Co-authorship models, improv with AI, and the dynamics of turn-taking narrative.

Improv theater has a golden rule: "yes, and..." Whatever your scene partner adds, you accept it as real and build on it. You don't block, redirect, or deny — you expand. Writers working with AI report that the most interesting results come from applying the same principle: accepting what the AI generates and asking yourself, "where could this go?"

Co-Authorship Models

Human-AI creative collaboration takes several forms:

  • Generative expansion: You write an outline or premise; AI expands it into full prose.
  • Turn-taking: You alternate contributions — each party builds on what the other adds.
  • Editing assistant: You write; AI suggests improvements, variations, or alternatives.
  • Constraint-based: You set strict constraints; AI works within them.
What You Bring That AI Can't

AI excels at execution within established patterns. Humans bring what isn't in the training data: lived experience, emotional specificity, cultural particularity, genuine surprise. The most interesting co-authored work comes from human directions that force the AI off its most statistically likely path.

The Creative Division

Use AI for bulk, pattern, and execution. Use yourself for direction, surprise, and meaning. The story is better when each party does what it's actually good at.

Quiz 2

We Tell Stories Together

4 questions — free, untracked, retake anytime.

the 'yes, and...' improv rule, what does 'yes' mean?

✓ Correct — ✅ 'Yes' means you accept what was established as real within the story — you don't block or deny it. 'And' means you build on it.
❌ 'Yes' in improv means accepting what your partner added as real within the story world — then 'and' means building on it.

type of co-authorship is happening when you write an outline and AI expands it into full prose?

✓ Correct — ✅ Generative expansion: you provide the structure and direction; AI generates the detailed prose within that framework.
❌ This is generative expansion — you provide structure and direction; AI generates detailed prose within your framework.

do humans bring to co-authorship that AI genuinely cannot?

✓ Correct — ✅ AI can't access lived experience, emotional specificity, or genuine surprise — those come from you. That's the human contribution that makes co-authored work interesting.
❌ Humans bring what isn't in training data: lived experience, emotional specificity, cultural particularity, and genuine surprise.

does AI produce the most interesting collaborative output?

✓ Correct — ✅ The most interesting results come when humans push AI toward the unexpected — forcing it off the statistically likely choices it defaults to.
❌ AI defaults to statistically likely choices. The most interesting output comes when human direction pushes it toward the unexpected.
Lab 2

Improv Narrative

Build a turn-taking story using yes, and... thinking.

Lab 2 — Improv Narrative

Build a story using turn-taking with your AI guide. Apply "yes, and..." thinking throughout.

  1. The AI opens with one story sentence.
  2. You add one sentence, accepting what the AI established and building on it.
  3. Continue for at least 3 full exchanges, then discuss: what surprised you?
Try to push the story somewhere unexpected on your turn. Force the AI off the obvious path.
🔬 AI GuideLab 2
Lesson 3

AI Makes Pictures and Music

Image and audio generation — how it works, what it produces, and what it means.

Midjourney, DALL-E, and Stable Diffusion entered public use in 2022-23 and immediately won art competitions, illustrated books, and generated album covers — sometimes without disclosure. In August 2022, "Théâtre D'Opéra Spatial" by Jason Allen, generated with Midjourney, won first place at the Colorado State Fair fine arts competition. Allen disclosed he used AI. The result sparked a global debate about whether AI-generated images constitute art.

How Diffusion Models Work

Image generation models (like Stable Diffusion) use a process called diffusion: they start with random noise and gradually remove that noise in a direction guided by a text prompt, until a coherent image emerges. They were trained on billions of image-caption pairs from the internet.

  • Text-to-image: A written prompt guides the noise-removal process toward a matching image.
  • Style transfer: The model can apply a specified artistic style to new content.
  • Music generation: Audio models work similarly — generating waveforms guided by text or style descriptions.
The Training Data Question

These models were trained on images created by human artists — often without consent or compensation. This has led to lawsuits, licensing debates, and ongoing questions about what "learning" from creative work means legally and ethically.

Open Question

If a model trained on ten thousand paintings by a specific artist can generate images "in the style of" that artist, and sells those images commercially — what does the original artist deserve?

Quiz 3

AI Makes Pictures and Music

4 questions — free, untracked, retake anytime.

do diffusion models generate images?

✓ Correct — ✅ Diffusion models start with noise and progressively remove it in a direction guided by a text prompt, until a coherent image emerges.
❌ Diffusion: start with noise, progressively refine toward the prompt, until a coherent image emerges. No copying — generation.

Colorado State Fair AI art controversy raised what core question?

✓ Correct — ✅ The controversy was about authorship and competition: should AI-generated images, regardless of technical quality, compete in fine arts categories against human-created work?
❌ The core question: should AI-generated images count as art for competition purposes? Who deserves credit — the prompter, the AI, the training data creators?

is the key ethical concern about training data for image generation models?

✓ Correct — ✅ Image generation models were trained on billions of human-created images — often without the artists' consent or any compensation to them.
❌ The ethical issue: human artists' work trained the models without consent or compensation. The models can now generate work 'in their style.'

transfer in AI image models means:

✓ Correct — ✅ Style transfer: the model can apply a recognized artistic style — impressionism, a specific artist's technique — to new generated content.
❌ Style transfer applies a learned artistic style (from training data) to new content — generating something 'in the style of' a specified artist or movement.
Lab 3

Image Generation Ethics

Analyze the ethics of training on artists' work.

Lab 3 — Image Generation Ethics

Discuss the training data ethics question with your AI guide.

  1. The AI opens with the artist compensation question.
  2. Develop your position on whether training on artists' work without consent is acceptable.
  3. Address: does your answer change if the model is used commercially?
Consider: what rights do artists have over their style? Over their training data contribution? Is style even copyrightable?
🔬 AI GuideLab 3
Lesson 4

Is It Real?

Deepfakes, authenticity, and synthetic media literacy.

In 2023, a synthetic image of an explosion near the Pentagon went viral, briefly sending stock markets down before being identified as AI-generated. No explosion had occurred. The image had been shared by verified Twitter accounts. It took less than an hour to cause measurable financial impact before fact-checkers caught up.

Deepfakes and Synthetic Media

Deepfakes use AI (specifically GANs — Generative Adversarial Networks) to create convincing synthetic video or audio of real people. The technology started as a research curiosity and is now widely accessible.

  • Face-swap deepfakes: Place one person's face on another's body in video.
  • Voice deepfakes: Synthesize a person's voice from a few minutes of audio.
  • Synthetic news images: Fabricated photographs of events that didn't happen.
Synthetic Media Literacy

Detecting synthetic media is increasingly difficult as tools improve. More reliable than technical detection is source verification:

  • Check the original source — who published this, and when?
  • Reverse image search — does this image appear elsewhere in a different context?
  • Check for corroboration — are multiple credible independent sources reporting this?
  • Consider motive — who would benefit from this being believed?
Key Shift

Synthetic media has effectively ended the era in which "seeing is believing." Verification now requires sourcing, not just perception.

Quiz 4

Is It Real?

4 questions — free, untracked, retake anytime.

made the synthetic Pentagon explosion image particularly dangerous?

✓ Correct — ✅ The image caused measurable stock market movement before being debunked — demonstrating that synthetic media can have real-world consequences faster than fact-checking can respond.
❌ The danger: it caused measurable real-world impact (stock market movement) before fact-checkers identified it as AI-generated.

'voice deepfake' can be created from:

✓ Correct — ✅ Modern voice synthesis tools can clone a voice convincingly from just a few minutes of audio — a phone call, a speech, a podcast appearance.
❌ A few minutes of audio is often enough for a convincing voice deepfake — a phone call, speech, or podcast appearance.

is source verification more reliable than technical detection for synthetic media?

✓ Correct — ✅ Generation quality improves faster than detection tools can adapt. Source verification — who published it, when, with what corroboration — is a more durable strategy.
❌ Detection tools are an arms race that keeps losing. Source verification — who published this, when, what corroborates it — is more durable.

key question does synthetic media raise about the phrase 'seeing is believing'?

✓ Correct — ✅ Synthetic media has effectively ended the era in which visual perception alone was sufficient. Verification now requires sourcing, not just seeing.
❌ 'Seeing is believing' no longer works. Synthetic media means you must verify source and corroboration — not just trust what you see.
Lab 4

Synthetic Media Literacy

Build your personal verification protocol.

Lab 4 — Synthetic Media Literacy

Build your personal verification protocol for synthetic media.

  1. The AI presents the Pentagon image case and asks how you would have identified it as fake.
  2. Build a step-by-step verification protocol.
  3. Address: what responsibilities do platforms have for synthetic media?
Consider: source, corroboration, motive, and technical signals. Which matters most?
🔬 AI GuideLab 4
Lesson 5

Who Owns the Story?

Copyright, authorship, and creative credit in the age of AI.

In 2023, the US Copyright Office ruled that a graphic novel created with Midjourney images could not be copyrighted — because copyright requires human authorship. The comic's creator, Kristina Kashtanova, retained copyright for the text and arrangement, but not the AI-generated images themselves. Meanwhile, authors including George R.R. Martin and John Grisham sued OpenAI, alleging their books were used to train ChatGPT without permission.

Copyright and AI Outputs

Current US copyright law requires human authorship. AI-generated content — images, text, music — cannot be copyrighted as of 2023 rulings. This creates a paradox: anyone can use, copy, and sell AI-generated content freely.

  • Content you write, edit, or substantially shape can still be copyrighted.
  • Pure AI output (minimal human input) currently has no copyright protection.
  • The law is rapidly evolving across different countries.
Training Data and Authorship

The second legal battleground: were AI models trained on copyrighted works without permission? Current lawsuits challenge whether training on copyrighted text and images constitutes infringement — or fair use. No definitive ruling has been made in most jurisdictions.

Open Question

If an AI model trained on your creative work can produce outputs that compete commercially with your own work, what legal and ethical obligations does the model's creator have to you?

Quiz 5

Who Owns the Story?

4 questions — free, untracked, retake anytime.

can't pure AI-generated images be copyrighted under current US law?

✓ Correct — ✅ Current US copyright law requires human authorship. Content generated purely by AI, with minimal human creative input, currently cannot be copyrighted.
❌ Copyright requires human authorship. AI-generated content without substantial human creative input has no copyright protection under current US law.

did the Copyright Office allow Kristina Kashtanova to copyright in her AI-illustrated comic?

✓ Correct — ✅ She retained copyright for the text and the arrangement of elements — the human creative decisions. The AI-generated images themselves were not copyrightable.
❌ Kashtanova retained copyright for her text and creative arrangement — the human authorship portions. The AI images themselves couldn't be copyrighted.

is the core legal claim in author lawsuits against AI companies like OpenAI?

✓ Correct — ✅ The core claim: training on copyrighted works without permission may constitute infringement — and authors deserve compensation for their work being used to train AI.
❌ The lawsuits claim that using copyrighted books to train AI without permission or compensation may constitute copyright infringement.

is the practical consequence of AI outputs having no copyright protection?

✓ Correct — ✅ Without copyright protection, AI outputs enter a kind of public domain by default — anyone can use, copy, and sell them freely.
❌ No copyright = no exclusive ownership. Anyone can use, copy, and sell AI-generated content without restriction under current law.
Lab 5

Copyright in the AI Age

Analyze authorship and copyright for AI creative work.

Lab 5 — Copyright in the AI Age

Analyze the authorship and training data questions with your AI guide.

  1. The AI opens with the Kashtanova case and asks where you think the line between human and AI authorship should be drawn.
  2. Develop your position on copyright for AI-assisted creative work.
  3. Address the training data question: what do authors deserve?
Consider: what percentage of human creative input makes something 'yours'? What does an author deserve if their work trained a model?
🔬 AI GuideLab 5
Lesson 6

AI and Human Creativity Together

Collaboration models, augmentation vs. replacement, and the future of creative work.

The band Radiohead's producer Nigel Godrich, the composer Hans Zimmer, and novelist Kazuo Ishiguro have all spoken publicly about using AI in their creative process — not to replace creativity, but to get unstuck, generate raw material, or hear unexpected possibilities. At the same time, entry-level illustrators, stock photo contributors, and junior copywriters have seen their markets collapse as AI tools perform similar work at near-zero cost.

Augmentation vs. Replacement

AI creative tools can augment human creativity — expanding what's possible for a skilled human — or replace it, substituting AI output for human work at lower cost. The same technology does both simultaneously, for different people:

  • An established novelist using AI to draft rough scenes they then refine: augmentation.
  • A company replacing junior copywriters with AI output: replacement.
  • A solo filmmaker using AI to generate concept art they couldn't afford to commission: augmentation.
  • A stock photo site replacing human photographers' libraries with AI-generated images: replacement.
The Creative Division of Labor

The most productive human-AI creative relationships typically follow a pattern: humans provide direction, surprise, meaning, and specificity; AI provides scale, variation, and execution. The challenge is maintaining the human contribution rather than deferring to AI defaults.

The Risk

When humans consistently accept AI's first suggestions rather than pushing back, the creative output converges on the statistical average of the training data. The distinctive voice gets averaged out.

Quiz 6

AI and Human Creativity Together

4 questions — free, untracked, retake anytime.

is the difference between AI augmenting and AI replacing creative work?

✓ Correct — ✅ Augmentation: AI expands human creative capability. Replacement: AI substitutes for human labor. The same tool does both depending on how it's deployed.
❌ Augmentation = AI expands what a skilled human can do. Replacement = AI substitutes for human work, reducing the need for human labor.

of these is an example of AI augmentation (not replacement)?

✓ Correct — ✅ The filmmaker is using AI to do something they couldn't otherwise afford — expanding their creative capacity. That's augmentation.
❌ The filmmaker using AI for concept art they couldn't otherwise afford is augmentation — AI expanding what's possible, not replacing a paid human worker.

is the main risk when humans consistently accept AI's first creative suggestions?

✓ Correct — ✅ When humans defer to AI defaults, the output reflects the statistical average of training data. The distinctive, specific human voice gets diluted.
❌ Consistently accepting AI defaults means the output converges on statistical averages — the distinctive, specific human voice gets averaged away.

do humans bring to AI collaboration that AI cannot replicate?

✓ Correct — ✅ Humans bring direction, genuine surprise, meaning, and specificity from lived experience — the parts of creativity that aren't in the training data.
❌ Humans bring direction, surprise, meaning, and specificity from lived experience. These are the parts of creativity that genuinely aren't in the training data.
Lab 6

The Creative Division of Labor

Develop your framework for human-AI creative collaboration.

Lab 6 — The Creative Division of Labor

Develop your own framework for human-AI creative collaboration.

  1. The AI presents the augmentation vs replacement distinction and asks where you see the dividing line.
  2. Develop your framework for what humans should contribute vs. defer to AI.
  3. Address: how do you protect your distinctive voice in AI-assisted creative work?
Consider: which parts of your creative process would you use AI for, and which would you never hand off?
🔬 AI GuideLab 6

Module 4 Test

6 questions covering all lessons. Free, untracked, retake anytime.

makes AI story generation potentially formulaic?

✓ Correct — ✅ AI generates statistically likely text — which means it defaults to common patterns. Breaking formula requires explicit prompting.
❌ AI generates statistically likely text, which defaults to common patterns and can feel formulaic without explicit prompting to do otherwise.

human-AI creative collaboration, the most interesting output comes from:

✓ Correct — ✅ Human directions that force AI toward the unexpected — off the statistical average — produce the most distinctive collaborative output.
❌ The most interesting output comes when human direction pushes AI off its statistically likely defaults toward the unexpected.

models generate images by:

✓ Correct — ✅ Diffusion: start with noise, progressively remove it in the direction of the prompt, until a coherent image emerges. Generation, not retrieval.
❌ Diffusion models start with random noise and progressively refine it toward the prompt — they generate, not retrieve.

can't pure AI-generated images be copyrighted under current US law?

✓ Correct — ✅ Current US law requires human authorship. AI-generated content without substantial human creative input cannot be copyrighted.
❌ Copyright requires human authorship. Without sufficient human creative input, AI output has no copyright protection.

key difference between augmentation and replacement in AI creative work is:

✓ Correct — ✅ Augmentation = AI expands what a skilled human can do. Replacement = AI substitutes for human labor. The same tool can do both.
❌ Augmentation expands human creative capability. Replacement substitutes AI output for human labor at lower cost. Same tool, different deployment.

media literacy primarily means:

✓ Correct — ✅ Since generation quality outpaces detection tools, synthetic media literacy focuses on source verification: who published it, when, and what corroborates it.
❌ Synthetic media literacy = source verification over visual detection. Check who published it, when, what corroborates it, and who would benefit from it being believed.