AI art authorship raises a simple but uncomfortable question: if a machine does most of the visible work, are you still the creator?
This question is personal for me. I spent more than thirty years as a software engineer, including work at Apple and GE. I hold patents, studied architectural design, and have spent much of my life thinking in systems. I also paint, draw, and write songs with a guitar or piano.
Those experiences pull me in two directions. I know what a pencil feels like against paper. I also know that nobody builds a modern software system by controlling every transistor. We work through layers of tools and abstraction. So I do not believe the answer is as simple as “AI is only a tool” or “anything made with AI is fake.”
For me, the real issue is human intent. How many meaningful choices did the person make, and how much of the result did the machine decide?
Why this question feels personal
When I paint or draw, my mind does not work only through color and emotion. It maps relationships. It looks for structure. It asks why one line belongs here and another does not. Even in music, I look for patterns. That is why I have explored ideas such as chord progressions as design patterns.
I have sometimes thought of myself as a passive artist but an active designer. I may not always have the fastest hand or the strongest performance skills, but I care deeply about the idea, the structure, and the reason each part exists.
AI makes that distinction harder to see. A person can type one sentence and receive a polished image in seconds. Another person may spend weeks sketching, masking, rejecting, adjusting, combining, and rebuilding. Both may say, “I made this with AI,” but they did very different work.
Generative art did not begin with generative AI
The word generative existed in art long before today’s AI boom. Artists and programmers used code to create images, animation, sound, and interactive experiences. Tools such as Processing and p5.js made this kind of creative coding more accessible.
In traditional generative art, the artist writes the rules that define the system. The program may still use randomness. In fact, Processing shows that a sketch can produce a different result each time it runs. When the artist wants repeatable results, a fixed random seed can reproduce the same sequence. The important point is not that every result is deterministic. It is that the artist designed the rules, inputs, and boundaries.
Modern generative AI works differently. The user usually does not define every visual rule. A trained model supplies much of the learned capability. It has detected statistical patterns connected to objects, lighting, perspective, texture, and style. The user asks for a result through prompts or other controls, and the system fills in many details.
This difference matters. We should not use one word—“generative”—as if creative coding and prompting a trained model were the same process.
A prompt is not a paintbrush
People often defend AI art by saying that it is just another tool. That comparison is partly true, but it is incomplete.
A traditional brush is passive. It does not know what a castle looks like. It does not suggest the lighting, repair the perspective, or add stones to an empty wall. The person supplies both the idea and the execution.
A text-to-image system contributes much more. If I type “a medieval castle at sunset” and accept the first result, the system has made thousands of decisions for me. I am closer to a client requesting an image than an artist drawing one.
That does not mean the result has no value. It may communicate an idea, solve a practical need, or inspire something better. But I should be honest about where the craft came from.
The training behind these systems also creates real ethical and legal questions. Some models were trained using large collections that included copyrighted material. Whether a particular use is lawful depends on the facts and the jurisdiction; it should not be reduced to a slogan. The World Intellectual Property Organization describes this as an active area of dispute, not a settled question.
When the user becomes the designer
Now consider a different process.
A creator begins with an original idea and hand-drawn composition. They set strict requirements, provide their own reference material, reject weak results, mask specific areas, revise the lighting, combine multiple outputs, and manually edit the final work. They may repeat this process hundreds of times.
At that point, the person is no longer pulling a prompt lever. They are designing a system of constraints. The AI still performs some execution, but the human supplies the direction and makes the important expressive choices.
This feels familiar to me because architecture and software work in similar ways. An architect does not personally cut every board or pour every section of concrete. A software architect does not manually control every bit moving through a processor. We use abstraction so we can work at the level of systems, behavior, and intent.
Abstraction does not remove authorship. But it does create a responsibility: the higher the level of abstraction, the more clearly we must show what the human actually contributed.
The density of choice: a practical test
I have arrived at a simple principle:
Creative authorship is measured by the density of deliberate human choices.
If you are unsure whether an AI-assisted work is truly yours, ask five questions:
- Who supplied the central idea? Did it begin with your experience and point of view, or with a random suggestion from the system?
- Who made the important choices? Did you decide the structure, composition, tone, and meaning?
- What original material did you contribute? This could include sketches, photographs, music, writing, code, or carefully designed constraints.
- How much did you challenge the output? Did you accept the first result, or did you reject, revise, combine, and edit?
- Can you explain why the major elements are there? If you cannot explain the choices, the system may have made more of the work than you did.
This test is not a legal rule. It is a way to think honestly about creative control. In the United States, the Copyright Office also focuses on human expressive choices. It has said that prompts alone generally do not provide enough control, while human selection, arrangement, and modification may qualify for protection. Its guidance on AI-assisted works is useful reading for anyone publishing this kind of work.
AI at work changes the pressure, too
This is not only a debate about art. AI has become part of professional life. Engineers, writers, designers, musicians, and architects are being asked to work faster. Once a task can be completed in hours instead of days, that faster schedule can become the new expectation.
That creates a danger. The time saved by AI may not be returned to the creator for deeper thinking. It may simply become a demand for more output. Speed can replace judgment, and “good enough” can crowd out work that carries a real point of view.
My answer is not to reject AI. It is to use it deliberately. Begin with your own idea. Use your own material when possible. Treat generated output as raw material, not as a finished truth. Check the facts. Respect licensing and privacy. Spend the saved time improving the concept and the details.
So, who really makes the art?
If the machine chooses the composition, lighting, details, and meaning while you accept the first result, then the machine performed most of the creative work.
If you bring the original vision, define the constraints, make the expressive decisions, and reshape the result until it reflects your intent, then you are doing more than prompting. You are designing.
The pixels may be rendered by an algorithm you did not write. The notes may be played by software. The code may rely on frameworks built by other people. That has always been true of creative tools at some level.
What matters is whether the work contains a dense trail of choices that only you would have made. That trail is where I find the human author.