Using AI as a digital artist and researcher.

Abstract

I use AI models as a digital artist and researcher, and this page states how. Models take part in ideation: the early, nondescript stage where a question is turned over, directions are sketched and a craft problem is worked through, with names, files and data left out. Everything after that stage is mine: the judgment, the making, and the words that carry my name. The paper sets out the ethics and the morals of that split, and what honest representation of self-expression asks of a maker who works with these tools. The video walks through one session, start to finish.

1. Where the models enter

Ideation is the first stage of every piece of work. The brief is questioned, the field is read, directions are sketched, and the craft problem is stated and restated until it is clear. At that stage I work with several models, each on its own terms: Claude, ChatGPT, Mistral, Aleph Alpha, Cohere, Qwen, Kimi K3 and DeepSeek. Which one depends on the question. A model is good at breadth: naming the ten ways a thing has been done, finding the paper I half remember, arguing the other side of a position I hold.

The prompts are nondescript. No client name, file, dataset, credential or unreleased work goes in. The question goes in, in general terms, and the answer comes back as material to test, never as a decision. The same models appear as tiles on the home page, beside the drawing and building tools, because that is what they are in this studio: tools, listed in the open.

2. Where they stop

Judgment stays with me. Which direction survives, what the work looks like, what it says and whether it is good are decisions with an owner, and the owner is a person. The studio's rule for automation applies to its own art and research as well: automate repetition, not judgment.

Nothing reaches a client or the public as mine unless I made it, or reviewed it and approved it. Code an AI helped write ships after an adversarial review and tests. A finding is mine to defend; a model's summary of a source is a lead to check against the source.

3. Ethics

The paper works through four questions.

1. Provenance. The models were trained on work by people who were not asked. Using them is a choice with that fact inside it. The paper takes the fact seriously and states what follows from it for a working artist.
2. Attribution. Where a direction came from a model, the paper argues for saying so at the level of the process, the way a researcher names a method, and for keeping the credit where the labour was.
3. Substitution. Whose work is displaced when a studio of one reaches for a model in place of a collaborator, and when that is defensible.
4. Sovereignty. Where the data goes and under whose law it sits is an ethical question before it is a compliance one. The studio's rule is the practical answer.

4. Morals

Ethics is the argument. Morals are what I hold myself to when no one is checking.

• I do not pass a model's output off as my hand.
• I do not put another person's material into a model without their agreement, and never into one that fails the four tests in the privacy policy.
• I say which tools touched a piece when asked, and on this site without being asked.
• I name the invisible labour, including the labour of the people whose work these models learned from, and I keep the credit where it was earned.

5. Honest representation of self-expression

A digital artist's work is a claim about the person who made it. A tool that can generate a plausible version of anyone's work puts that claim under pressure. The answer I hold to is disclosure and a stable line. Models take part in ideation, in the studio, under the rules above. The artefact, the taste and the voice are mine. If that is ever otherwise for a piece, the piece says so.

The same line runs through the research. A question may be sharpened with a model. The method, the reading of the evidence and the conclusion are mine, and they are written so that a reader can check them.

6. Client material

Client work runs under a stricter rule than the studio's own ideation: client material is never fed into an AI model that was not designed for data sovereignty. What that means, and which providers meet it today, is written out in the privacy policy. Each proposal names the tools, if any, that will touch a client's material, and a client may ask for none.

7. Video and paper

The paper carries the full argument with its sources. The video walks through one ideation session, from the first prompt to the point where the models are set aside and the work begins.