Creative AI

Introduction¶

You are reading the open textbook for Creative AI, a bachelor-level course at the University of Oslo (UiO). The course runs for twelve weeks, with three 45-minute blocks each week (a 45-minute lecture and a 90-minute lab), and it is open to students from all faculties.

There are no prerequisites. You need no programming, no machine learning, and no training in any particular art form; basic digital literacy is assumed, and curiosity from your own discipline matters more than any specific background. The book is also written for self-study, and as a public resource for anyone curious about creative uses of AI.

The aim of the course is twofold. The first aim is to give you a working understanding of how today’s generative AI systems are built and why they behave the way they do. The second is to give you hands-on experience of using these systems for creative work, together with the critical vocabulary needed to discuss their cultural, ethical, and environmental consequences. Neither aim works without the other: understanding without practice stays abstract, and practice without understanding leaves you at the mercy of whatever the tool happens to do.

Three concepts thread through every week: intentionality, aesthetic control, and ethical authorship. The interesting question is not can the model do this, but what do you want, how do you guide the system there, and on whose terms.

Course at a glance¶

Course description (for the course catalogue)¶

Artificial intelligence has moved from a back-office technology into a tool that shapes everyday creative work: writing, drawing, photography, music, video, design, code, journalism, scholarship, and teaching. Creative AI introduces students from across the University of Oslo to the concepts, tools, and ethics of working creatively with contemporary AI systems.

The course assumes no prior programming or specialised art background. Through a combination of short lectures and weekly hands-on labs, students learn how today’s text, image, sound, video, and multimodal models work. They also learn how to direct and refine their outputs, how to integrate them responsibly into their own discipline, and how to reflect critically on the cultural, legal, environmental, and political stakes of the technology.

Every lab follows the same three phases, in the order Explore, Reflect, Create. Students leave the course with a small portfolio of creative AI artefacts, a documented prompt and decisions log, and a semester project performed or installed at The Synthetic Gallery, a public showcase held at UiO in the exam period. The semester project is the exam. The activities that lead up to it are obligatory and assessed pass or fail.

Five layers¶

Any Creative AI system can be described at five layers. They run from what goes into the system to what it changes once it reaches the world, and each layer comes with its own set of questions.

Every chapter opens by naming the layer or layers it works at, so that you always know which questions are in play. The five layers give a shared vocabulary for describing very different systems at the right level of detail. They earn their place for one reason: many disagreements about AI turn out to be disagreements about which layer the two parties are standing on.

Course schedule¶

The twelve-week schedule maps the lectures and labs. The labs deliberately mix modes. Some weeks you work with tools in a browser; other weeks you read a model card, sketch a pipeline on paper, or run a few lines of Python in a notebook on your laptop or in UiO Educloud.

Five pages sit outside the weekly rhythm. This overview is the first of them. Tips and tricks gathers the craft that carries across tools: prompt patterns, decision logs, how to write and cite AI-made work, accounts and privacy at UiO, and how to prepare for the gallery. The glossary collects the key terms defined in the chapters, as a lookup list for revision. The tools page lists the tool categories used in the course, with current examples, an open alternative in each category, and the small web apps built for this book. The gallery page describes the Synthetic Gallery: the format, the requirements, the page template, and the archive of past cohorts.

Explore, Reflect, Create¶

Every lab runs in the same order, and every chapter uses the same three subheadings for it. These are not separate assignments, but three modes of engagement that together make up a Creative AI practice, and the order matters: exploring comes before deciding, and deciding comes before making.

1. Explore (about 30 minutes). Controlled experiments with the week’s tool: vary one thing, compare two tools, break something on purpose. This is where the model surprises you.
2. Reflect (about 15 minutes). A structured discussion in pairs or in plenary, not a writing block. It ends with each student stating one intention for what they will make. This is where you decide.
3. Create (about 45 minutes). Make the artefact you stand behind, and carry it over into your portfolio at home. This is where you exert your will.
4. Log (at home). The three-paragraph weekly entry, one paragraph each on Explore, Reflect, and Create, closes the loop as the retrospective.

The cycle maps onto the two questions you will answer in every process memo. Explore is where you meet surprise, the moment a system does something you did not predict. Create is where you exert your will, the moment you decide what the work should be and bend the output towards it. Reflection sits between them, turning a surprise into an intention.

Weeks 3 and 12 bend the timings, since the ethics week is reflection-heavy and the final week reserves time for project rehearsal, but the order stays the same.

Each phase also names a disciplinary way into the subject, so that whatever you studied before, one of the three is already familiar ground.

Explore¶

Use AI-based systems in creative practice, and see how creative methods can be applied in other domains. This is the applied and behavioural track: psychology, therapy, educational sciences, cultural heritage. Explore activities investigate how AI can enhance creativity, foster innovation, and support learning and well-being, by trying things and documenting what the tools actually afford.

Typical outputs: controlled experiments (same prompt, vary one knob); prompt comparisons across two tools; hallucination hunts; model-card analyses; tool-to-task mappings; failure-mode catalogues.

Reflect¶

Critically study and discuss the impacts of AI on humans, human creativity, cultures, and society at large. This is the humanities and social sciences track: ethics, history, aesthetics, politics, law, learning. Reflect activities ask why the work is being done, who pays for it, who benefits, and who is left out.

Typical outputs: in-class debates; ethical audits; comparative readings of two tools or two cases; honest captions and provenance notes for your own work; the middle paragraph of your weekly log.

Create¶

Make AI-based systems, tools, artworks, frameworks, and policies. This is the making track: computer science, engineering, art, design, with an explicit emphasis on co-creative AI systems that prioritise human agency, environmental sustainability, and the democratisation of AI technologies. Each lab leaves you with a concrete artefact for your portfolio.

Typical outputs: images, songs, short videos, code sketches, design pipelines, prototype agents, policy briefs, exhibition pieces.

How the three connect¶

The three modes reinforce one another. Explore asks what the tool actually does; Reflect asks why you are doing this and at whose cost; Create asks what you can make, with that understanding, that you are willing to stand behind. A good Creative AI practitioner never separates them for long.

The rhythm also maps onto the disciplinary breadth of the course, and of the co-creative AI research at RITMO and the fourMs Lab that gave the course its shape. Applied and behavioural sciences sit on the Explore side, humanities and social sciences on the Reflect side, and computer science, engineering, art, and design on the Create side. Students from every background contribute in every mode.

Pedagogical strategy¶

A course for everyone at UiO¶

Creative AI is not a specialist course in computer science, art, or media. It is designed as a general education course: students arrive from law, medicine, musicology, design, theology, mathematics, dentistry, education, literature, biology, and many other places. That diversity is the point: the classroom is itself a small interdisciplinary laboratory, where lawyers and artists, medics and historians can ask hard questions of the same AI tool and watch each other’s answers.

No programming background is assumed. Where technical depth would help, it sits in a collapsible Dig deeper note that you can open or skip without losing the thread, and any code is in notebooks you can run in a browser without installing anything.

Active learning and a flipped classroom¶

The course is built around active learning and a flipped classroom model. You read the chapter before the lecture; the lecture is the place to argue, demonstrate, and answer the questions you bring in; the lab is where you make things.

Expect roughly six hours of self-study and project work per week on top of the three 45-minute teaching blocks, including the reading, the follow-on experiments, and the assignments.

Studio labs and process over polish¶

The labs run studio-style: short briefs, fast iteration, peer feedback, and a teacher or teaching assistant on hand. The assessment philosophy follows from this. We grade process, reflection, and deliberate decisions rather than technical perfection. Risk-taking, honesty about failure, and originality are explicitly rewarded, and a failed experiment you can explain is worth more than a polished result you cannot account for.

Research-based and research-led¶

This is a research-based course: the content rests on current research in machine learning, human-computer interaction, media studies, the humanities, and the arts. It is also research-led, because the people teaching it are doing that research now.

The course grew out of three environments at UiO. RITMO is the Centre for Interdisciplinary Studies in Rhythm, Time and Motion, where musicologists, psychologists, and informaticians study how people move, listen, and synchronise. The fourMs Lab is its laboratory for music, mind, motion, and machines, equipped for motion capture, sensors, sound, and robots, and it is the home lab of this course. The MishMash Centre for AI and Creativity began as a collaboration between RITMO and the CreaTeME centre at the University of Agder, and is now a national consortium led by UiO. It aims to create, explore, and reflect on AI for, through, and in creative practices.

Every chapter carries a Research spotlight box on one project from these groups, with a link, a citation, and a sentence on how you could connect to the work yourself. Several of those studies run while the course is taught, and students regularly take part, as participants in an experiment, as performers in a lab session, or by carrying a semester project into an ongoing study. Ask in the lab if something in a spotlight catches your interest.

Reading claims about AI¶

Claims about what AI can and cannot do arrive faster than anyone can check them, and many of them are made by people with something to sell. Use the same four questions every time, whatever the source:

1. The claim. What exactly is being asserted, in one sentence, and about which system and which version?
2. The evidence. What was actually measured, on what material, and by whom?
3. The method. How was it measured, who chose the comparison, and could the result have come out otherwise?
4. The limits. What does the evidence not cover, and what would have to be true for the claim to generalise to your own work?

The checklist applies to three genres you will meet constantly. A model card documents a model’s intended use, training data, and known weaknesses; read what it declines to say as carefully as what it states. A benchmark reduces a capability to a score on a fixed set of tasks; ask what the tasks are and whether they resemble anything you do. A company announcement is a marketing document with numbers in it; look for the demo conditions and the phrase “in our tests”. Every chapter puts the checklist to work in its A critical look section, where one popular claim about Creative AI is taken through the four moves.

Guest lecturers¶

Where possible, each week’s lecture features a guest from a UiO department or research centre whose work meets the week’s theme. RITMO and the fourMs Lab cover the sound, body, and rhythm weeks, the Department of Informatics the model weeks, and the Department of Media and Communication the ethics and media weeks. Design and architecture colleagues take the spatial week, the Faculty of Law the copyright discussions, and the National Library of Norway’s AI lab Norwegian language and audio.

Open education¶

The textbook follows the principles of open education and open research: the material is openly licensed (CC-BY-4.0), the source is on GitHub, and we point to open tools and datasets where possible. Where a tool requires a paid account, we say so and try to offer an open alternative.

This is also a political stance. Creative AI is being built and deployed mainly by a handful of large companies. Treating the study of Creative AI as an open, collaborative project is one small way to push back.

A note on AI tools used to write this book¶

This textbook is itself an example of AI-supported authorship. Drafts of every chapter have been written collaboratively with large language models, then revised, fact-checked, and re-organised by human editors. Where AI tools have produced figures or examples, we say so. We treat the book as a living document, so please open an issue or a pull request on the GitHub repository when you spot errors or omissions.

Assessment¶

There is no written school exam. The expectation is that you can talk and write coherently about what you made, with what tools, and why. Assessment has two parts: the semester project, which is the exam, and the obligatory activities that lead up to it.

The semester project¶

The exam is your semester project, performed or installed at The Synthetic Gallery, the public showcase held in the exam period. It is graded A to F, and you deliver three things:

• the work itself, performed live or installed for visitors to come to;
• a short critical reflection of 1 000 to 1 500 words;
• your full prompt log, the file you have been keeping all semester.

You can work alone or in a group of two or three. A group delivers one project and one reflection, and the reflection names what each person contributed. A group receives one grade for the whole group, unless the course leader decides otherwise.

Two requirements carry through from the rest of the semester. The work must use at least two data types (text, image, audio, video, 3D, motion), for example text and image, or audio and code. And the reflection is structured by the two process-memo questions below, on surprise and on will.

Every obligatory activity must be approved before you can present. The gallery page has the format, what counts as a performance and what counts as an installation, the full delivery list, the page template, the consent options, and an archive of projects from earlier cohorts. Read it early, in week 9 at the latest, when you start writing your proposal.

Obligatory activities¶

Everything else in the course is an obligatory activity, assessed pass or fail. Together they form a ladder, and each rung is meant to make the next one easier. A failed activity can be revised once.

Process memo¶

Each activity apart from the weekly log is submitted with a short process memo answering, at minimum, two questions that recur all semester (adapted from the practice-based tradition at RITMO and from earlier creative-AI courses):

1. Surprise. Where did the AI surprise you, pleasantly or unpleasantly, and what did that teach you about the tool?
2. Will. Where did you exert your own creative will over the output, through prompt, edit, refusal, selection, or composition?

These two questions run through the whole course, and they are also the spine of the reflection you hand in with the exam project. Answering them honestly, across everything you make, is what the assessment is looking for.

Reading list¶

The textbook itself is meant to carry the conceptual load of the course. The readings below are complements and deepenings. They include short essays you can read in an evening, books you can dip into around the chapter that needs them, and a few classics that anyone working with AI in cultural production should meet at least once. Many of them are openly available; the rest are in the UiO library.

You are not expected to read all of them. Pick one core title to read alongside the textbook over the semester, and dip into the supplementary list around the weekly topics that pull you.

Core curriculum¶

A short, deliberately broad list. Pick one to read in parallel with the textbook:

• Mitchell, M. (2019). Artificial Intelligence: A Guide for Thinking Humans. Farrar, Straus and Giroux Mitchell, 2019. — the most readable contemporary introduction to AI for a general audience, by an AI researcher who refuses to oversell. Best technical anchor.
• Hertzmann, A. (2018). “Can Computers Create Art?” Arts 7(2):18 Hertzmann, 2018. — short, open-access essay covering the philosophical question that hovers over the whole course. Best creative-AI anchor.
• Crawford, K. (2021). Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press Crawford, 2021. — structural critique of where the data, the labour, and the energy come from. Best critical anchor.
• Bridle, J. (2022). Ways of Being. Allen Lane Bridle, 2022. — contemporary, accessible reframe of “intelligence” beyond the human-vs-machine binary. Best forward-looking anchor.

If you read only one of these, pick Mitchell for the technical side or Hertzmann for the creative side. If you read two, pair one of those with Crawford or Bridle.

Supplementary reading by theme¶

Each weekly chapter ends with a Further reading section that adds chapter-specific suggestions on top of this list.

How to read this book¶

The chapters are roughly linear, but each is also self-contained. If you are new to the field, read them in order, since every chapter assumes the vocabulary of the ones before it. If you already have some background, skim chapter 2 and jump to the chapters on the medium you work in. Read the week’s chapter before the lecture, not instead of it.

Every chapter has the same eleven parts, so you always know where to look:

1. A title and short description at the top, saying what the chapter covers.
2. An opening paragraph naming the layer or layers the chapter works at, linked back to the five layers.
3. Prose sections that explain ideas rather than document tools, with occasional Question boxes and short exercises where a two-minute experiment helps more than another paragraph.
4. Dig deeper notes, in the chapters that have them, collapsed by default and holding the optional technical depth: equations, model details, and code listings. You can skip every one of them and still pass the course.
5. A Research spotlight on one project from RITMO, the fourMs Lab, or the MishMash Centre, with a link and a sentence on how you could connect to it.
6. This week’s lab: Explore, Reflect, Create, the plan for the 90-minute lab, with the timings described above.
7. A critical look at one popular claim about Creative AI, worked through the claim, the evidence, the method, and the limits.
8. A chapter summary in one paragraph, for revision.
9. Five questions to test whether the chapter landed, worth answering before the lecture.
10. Further reading, at least three items, every one with a link or a DOI.
11. Explore interactively, links to the small web apps built for that chapter.

You are ready to begin.

Learn more¶

Two courses at the Department of Musicology sit close to this one, and all three can be taken in any order. None of them is a prerequisite for the others.