AI, creativity and copyright

What you'll learn

• What copyright actually protects, and what it has never protected
• Why style and expression are treated differently, and why that distinction does most of the work
• What the dispute over training data is really about
• Who owns the output of an AI tool — and why "nobody" is a real possible answer
• Why a licence matters more than ownership for most practical purposes
• Where likeness, voice and performance raise issues copyright does not cover
• What creativity might mean when generating an image takes seconds
• How to use AI in creative work without producing something you cannot use

Key terms and definitions

Core concepts

What copyright protects

Copyright arises automatically when someone creates an original work in a fixed form — writing it down, recording it, saving the file. No registration, no symbol, no notice required. A sketch in a school book is protected from the moment it exists.

What it protects is narrower than people assume. It covers expression: the particular arrangement of words, notes, lines or pixels that the creator produced. It does not cover:

• Ideas — a story about a school on a flooded island
• Facts — the boiling point of water, last year's results
• Methods — a way of solving an equation
• Styles — painting in thick, visible brushstrokes

This is the idea–expression distinction, and it carries more weight in AI questions than any other single principle. Most arguments about AI and copyright turn out to be arguments about which side of that line something falls on.

One clarification worth making early. Copyright is about legal and economic rights — who may copy and reuse a work. That is a different question from honesty about your own work in an assessment, which is a matter of integrity rather than law. The two overlap but answer different questions, and an exam answer should keep them apart.

Style is not protected, and that is not an accident

If you ask a model for an image "in the style of" a living illustrator, you may get something recognisable. Many people's first instinct is that this must be theft.

Legally, in most systems, imitating a style is not infringement. And that rule long predates AI, for good reason: every artist learns by absorbing the work of others, and genres only exist because styles spread. A rule protecting style would make ordinary artistic development impossible.

What would be infringement is reproducing protected expression — a recognisable character, a specific composition, a logo, a distinctive figure that is itself the protected work.

But "it is legal" is not the end of the matter. Two further points stand:

• Scale changes the ethical picture. One student imitating a style is how art has always worked. A system that can produce thousands of convincing imitations, competing directly with the person whose work made the style recognisable, is a different situation in practice even where the legal rule is unchanged.
• Naming a living artist in a prompt is a choice. You are deliberately trading on a reputation they built. That may be legal and still be something you would not want to defend.

The training-data dispute

Models are built from very large amounts of existing material, much of it under copyright. Whether using it that way required permission is genuinely contested, is being litigated in several countries, and differs between jurisdictions. An honest answer says so rather than picking a side.

The shapes of the argument are clear enough to learn:

The case that permission was needed

• Copying the material to train is itself a reproduction
• Creators were not asked, credited or paid
• The output can compete commercially with the work it learned from

The case that it was not

• Training extracts statistical patterns rather than storing works
• The output is usually new expression, not a copy
• Many systems allow limited use of protected work for analysis and research

Two things you can say confidently whatever the law settles on. Training is not the same as storing — a model is not a database of its training material. But models can and sometimes do reproduce memorised fragments, especially text that appeared very often, so "it cannot copy" is too strong a claim.

Who owns what a model produces

Here is the part that surprises people. In several countries, including the UK and the US, copyright requires human authorship. On that basis, text or an image produced by a model from a short prompt may have no copyright owner at all — not you, not the provider, not the creators of the training data.

That has consequences students rarely anticipate:

• Anyone else may be able to use your generated image freely
• You may not be able to stop someone copying it
• A competition, publisher or client may refuse it for exactly that reason

Where a human contributes enough creative work — substantial editing, arrangement, a generated element composed into a larger original piece — protection may attach to that human contribution, not to the raw generated material.

Ownership and permission are different questions

For most real decisions, the licence matters more than the copyright. Providers' terms set out what you may do with output: whether you may use it commercially, whether attribution is required, whether some uses are excluded. Those terms bind you by agreement, regardless of who owns what.

So the practical sequence for any creative project is:

1. What do the terms of the tool permit for this use?
2. Does the output contain anyone's protected expression — a character, a logo, a recognisable figure?
3. Does it rely on a real person's likeness or voice?
4. Does the context require disclosure — a competition, a client, a publication, a course?

Likeness, voice and performance

Copyright is not the only right in play, and this is where AI has created the sharpest new problems.

A person's image, name and voice are protected by rights separate from copyright, which vary considerably by country. Generating a convincing likeness or a cloned voice can be unlawful, or actionable, even where no copyrighted work was copied at all.

The serious cases are easy to state: making someone appear to say something they never said; cloning a performer's voice to replace paid work; producing material that places a real, identifiable person — including a classmate — in a situation that never happened. Harm here does not depend on copyright, and consent is the governing idea rather than ownership.

What creativity means now

Worth separating two claims that get run together.

"AI is creative" usually rests on novelty: the output has not existed before. That is true and fairly weak — a random generator also produces novel output.

The stronger view is that creative work involves having something to say, judgement about what is worth making, and the taste to tell a good version from a mediocre one. A model has no intention and no preferences; it produces plausible work on request.

Which suggests where the human contribution actually sits in AI-assisted creative work: not in producing material, which is now cheap, but in deciding what to make, choosing between options, and knowing when something is not good enough. If you generate forty images and pick one, the judgement was yours — and the judgement is the part that was always scarce.

There is also a cost worth naming. Skill comes from doing the difficult part. Generating finished work instead of attempting it means the ability to make it yourself never develops — which matters most to the person who most wants to be good at it.

Worked examples

Example 1: Style imitation (4 marks)

A student generates artwork "in the style of" a living illustrator for a personal project. Discuss whether this is acceptable.

• Copyright protects expression, not style, so imitating a style is generally not infringement (1 mark)
• It would be infringement if a recognisable character or specific composition were reproduced (1 mark)
• Naming a living artist deliberately trades on a reputation they built, which is an ethical question separate from the legal one (1 mark)
• Scale matters: mass imitation competing with that artist's own work differs in effect from one student learning (1 mark)

Example 2: Ownership of output (3 marks)

A student generates a logo from a one-line prompt and assumes they own it. Explain why this may be wrong.

• Copyright in many countries requires human authorship (1 mark)
• Output produced from a short prompt may therefore have no copyright owner at all (1 mark)
• So others may be free to use the same logo, and the student could not prevent it (1 mark)

Example 3: Evaluating a claim (4 marks)

"Training a model on copyrighted work is obviously theft." Evaluate this.

• Training copies the material, and creators were not asked, credited or paid (1 mark)
• But training extracts statistical patterns rather than storing works, and the output is usually new expression (1 mark)
• Models can still reproduce memorised fragments, so a claim that copying is impossible is too strong (1 mark)
• The legal position is contested and differs between countries, so a confident verdict either way is unjustified (1 mark)

Common mistakes and how to avoid them

• Thinking copyright protects ideas. It protects expression — the particular form given to an idea.
• Assuming style imitation is infringement. It generally is not, which is why the ethical argument has to be made on its own terms.
• Assuming you own what you generated. Human authorship may be required, so there may be no owner.
• Confusing ownership with permission. The tool's licence terms govern what you may do.
• Forgetting likeness and voice. These are separate rights, and consent is the issue rather than copyright.
• Claiming models store their training data. They learn patterns, though fragments can be reproduced.
• Treating the law as settled. It is contested and varies by country.
• Equating novelty with creativity. Judgement and intention are the harder part.
• Mixing up copyright and academic honesty. Related, but different questions.

Using this in practice

Before using AI-generated material in anything that leaves your desk:

1. What do the tool's terms allow for this particular use?
2. Is anyone's protected expression in here — a character, a logo, a recognisable work?
3. Is a real person's likeness or voice involved? If so, is there consent?
4. Do I need to be able to own this? If so, generated material alone may not be enough.
5. Does this context require disclosure — competition, client, publication, course?
6. Is my contribution the judgement, or did I just accept the first result?
7. Am I avoiding the difficult part of something I actually want to get good at?

Quick revision summary

• Copyright arises automatically and protects expression, never ideas, facts, methods or style
• Style imitation is generally lawful; reproducing a character, logo or specific composition is not
• The ethical case against style imitation rests on scale and on trading on a named creator's reputation
• Whether training on copyrighted work needed permission is contested and varies by country
• Training extracts patterns rather than storing works, but memorised fragments can still appear in output
• Many systems require human authorship, so generated output may have no copyright owner — including no owner you can enforce
• A licence governs what you may do, and matters more in practice than ownership
• Likeness and voice are separate rights where consent, not copyright, is the issue
• Novelty is cheap; the human contribution is judgement — deciding what to make and when it is not good enough

A note on legal advice

Nothing in this topic is legal advice, and it should not be used as a substitute for it.

Laws differ considerably between countries, they change, and how a law applies depends on facts this material cannot know. Where this topic says something is unlawful or an offence, it is describing the general position in many jurisdictions so that you understand why the rules exist — not telling you what the law is where you live.

What this material is for is helping young people use technology, including AI, appropriately, lawfully and safely, and knowing where to go for help. For anything that has actual consequences, ask somebody qualified: a teacher, your school's safeguarding lead, a solicitor or attorney, your exam board, or the police. If something is happening to you now, do not wait for advice before telling a trusted adult.