A hammer can put a nail in a wall. The same hammer can kill someone. Nobody walks into a hardware store and demands the hammer be banned. We blame the person holding it.
That’s the argument most people reach for when AI in design comes up. “It’s just a tool.” For a while I accepted it. I use AI in my own work. It helps me explore layouts, test copy, clean up code and get through the boring parts of a project faster.
But the more I thought about the hammer, the less the comparison held. Not because AI is more dangerous than a hammer. Because of where it came from.
This article looks at AI design ethics from three angles: the tool, the way it lets you use it, and how you actually use it. My argument is simple. Designing with AI can be completely ethical. The way many AI systems were built, and how they treat the work they learned from, often isn’t. Those are two different problems, and mixing them up is exactly why the whole conversation feels broken.
Three Places Blame Can Sit
When a technology causes harm, there are roughly three places to look.
The tool itself. What it is, what it’s made of, how it was produced.
The affordances. What the tool makes easy, what it makes hard, and what it quietly encourages.
The use. What a specific person chose to do with it.
The “it’s just a tool” argument puts all the weight on the third point. Philosophers call this instrumentalism: technology is neutral, and morality lives entirely in the user.
Plenty of thinkers disagree. The historian Melvin Kranzberg summed it up in his first law of technology: technology is neither good nor bad, nor is it neutral. Langdon Winner’s 1980 essay “Do Artifacts Have Politics?” argued that the design of a technology can carry values and consequences before anyone even picks it up.
Designers should find this familiar. We know defaults shape behaviour. We know a checkout flow can push people into decisions they didn’t mean to make. We call those dark patterns, and we don’t excuse them by saying the user could have clicked differently. So we can’t suddenly claim tools are neutral just because the tool happens to be AI.
Where the Hammer Analogy Breaks
Here’s the part the hammer comparison misses.
A hammer is made of steel and wood. Nobody’s livelihood went into it without their knowledge. The carpenter using it isn’t building on top of other carpenters’ work that was taken without asking.
Generative AI is different. An image model learned what “editorial illustration” looks like by processing enormous amounts of editorial illustration. A text model learned how a good article reads by processing millions of articles. That material came from people: illustrators, photographers, writers, type designers, developers. In many cases they were never asked, never credited and never paid.
So the ethical question doesn’t start when you type a prompt. It starts before you open the tool. It’s baked into the material the tool is made from.
Picture a hammer forged from nails pulled out of other people’s houses at night. You could use it to build something beautiful. Your use might be perfectly decent. But the hammer still has a problem, and it isn’t yours.
That’s the gap in the “it’s just a tool” argument. It treats AI like a finished object with no history. AI tools have a history, and the history is the problem.
Why Designing with AI Can Be Ethical
None of this means designers who use AI are doing something wrong. That conclusion would be just as lazy as “it’s just a tool.”
Look at how AI actually shows up in a normal design workflow:
- Generating ten rough layout variations so you can reject nine of them faster.
- Writing first-draft alt text for 200 product images, which you then check one by one.
- Debugging a CSS grid that refuses to behave at 3am.
- Summarising twelve user interviews so you can spot patterns before a workshop.
- Testing how a headline reads at different lengths before the copywriter polishes it.
- Turning a messy spreadsheet of client feedback into a clear list of revisions.
- Resizing one campaign visual into 30 social formats instead of doing it by hand.
In each case, the designer is still the author of the intent. You define the problem, you judge the output, you take responsibility for what ships. The AI speeds up exploration and removes grunt work. It doesn’t replace judgment.
This isn’t so different from using a grid system, a font library or a stock icon set. Design has always stood on shared tools and conventions. Using something that makes you faster isn’t unethical in itself. Refusing to use it doesn’t make anyone more virtuous either.
Where It Actually Breaks: The Sources
The problem sits at the source. Many artists and advocacy groups frame it around three words: consent, credit and compensation.
Consent. Most creators whose work ended up in training data never agreed to it. Opt-out systems arrived later, if at all, and opting out after your work has already been used is like locking the door after the house has been emptied.
Credit. When an AI produces an image in a recognisable style, the person who spent years developing that style disappears from the picture. Design has always had influence and imitation, but a human imitator at least knows who they’re copying, and the audience can often trace it back.
Compensation. The value created by these systems is huge. The people whose work made that value possible mostly see none of it.
This isn’t abstract. Artists filed a lawsuit against several image-generation companies in 2023, and Getty Images took legal action against Stability AI. Early image models were widely reported to produce garbled versions of stock photo watermarks, a strange little fingerprint showing exactly where some of the material came from. Regulators are moving too: the EU AI Act now requires providers of general-purpose AI models to publish summaries of the content used to train them.
When a watermark shows up in your generated image, you’re looking at a nail from someone else’s house.
Five everyday examples of the source problem
To make this concrete, here are situations designers run into all the time.
The logo that looks oddly familiar. A small business owner asks an AI tool for a logo. The result is clean and professional. It’s also uncomfortably close to a mark a real studio designed for a real brand. The owner has no idea. Neither does the tool. The original designer never will either, until the two logos end up on the same high street.
The typeface nobody drew. Type design takes months, sometimes years. Every curve, every kerning pair, every weight is a decision. When a tool generates “a geometric sans like the ones foundries sell,” it’s drawing on that labour. As someone who has designed typefaces, I can tell you the craft is in the thousand tiny choices. A generator that flattens them into a prompt doesn’t just copy a shape. It copies a practice.
The stock photographer’s lost income. A photographer spent a decade building a library of lifestyle shots that paid a steady licensing income. Now clients generate “a smiling team in a bright office” for free. Some of that photographer’s own images may well have helped teach the model what that scene looks like. They’re competing against a machine trained partly on their own work.
The “make it look like that famous director” brief. A client sends a brief: “Poster for our café opening, but make it look like a well-known filmmaker’s aesthetic.” Before AI, you’d design something inspired by that mood, filtered through your own eye. Now the client can get a near-copy in seconds. The line between inspiration and extraction gets very thin, very fast.
The illustrator who became a prompt. Some living illustrators have found their names used as prompt keywords thousands of times. Their style became a filter anyone can apply. They didn’t sell it, license it or agree to it. It simply stopped belonging to them.
In every one of these cases, the person using the tool might be acting in good faith. The harm happened upstream.
The Affordance Problem: When the Tool Makes the Wrong Path Easy
Between “the tool” and “the user” there’s a middle layer that gets ignored: what the tool invites you to do.
Take the “in the style of” prompt. One line of text, the name of a living illustrator, and you get work that imitates their decade of practice with no credit and no fee. The tool doesn’t warn you. It doesn’t suggest licensing. It doesn’t tell you whose style you’re borrowing. It just delivers.
That’s a design decision. Someone chose to make that path frictionless.
Other examples follow the same pattern:
- Face and voice tools that let you generate a realistic person for an ad without any check on whether that face resembles a real, identifiable human.
- “Remove watermark” or “upscale any image” features that make it trivial to strip ownership signals from someone else’s photo.
- Fake testimonial visuals. A few clicks produce a “happy customer” photo with a quote underneath. The tool doesn’t ask whether that customer exists.
Now compare these with tools that show provenance, train on licensed or opt-in material, block prompts naming living artists, or share revenue with contributors. Some of these exist. They’re usually slower or more limited. And that’s the point: ethics often shows up as friction, and friction is usually the first thing a product team removes.
As designers, we should be the first to recognise this. If we’d criticise a banking app for hiding the cancel button, we can criticise an AI tool for making imitation the easiest action on the screen.
The Use Problem: Where You Still Carry the Weight
Being honest means admitting the third layer matters too. A tool built on shaky ground doesn’t excuse every use of it.
Two designers can open the same AI tool on the same morning.
The first uses it to sketch twenty rough compositions for a charity campaign, picks one direction, then builds the final artwork by hand with a hired photographer and original illustration. AI helped them think. The final piece is theirs.
The second types a living illustrator’s name, generates a finished poster, sells it to a client as original work and says nothing about how it was made.
Same tool. Same training data. Very different ethics. The source problem belongs to the builders, but the second designer made a choice that added a fresh layer of harm on top of it.
So Whose Problem Is It?
Back to the hammer. Is it the tool, the way it allows you to use it, or how you use it?
The honest answer is all three, but not equally.
Builders carry the heaviest weight. They chose the training material, the licensing (or lack of it) and the defaults. The original ethical failure sits with them.
Tool designers carry the affordance problem. They decide what’s easy, what’s visible and what’s blocked.
Users carry the use. You can’t fix how a model was trained, but you can decide what you prompt, what you disclose and what you ship.
This is why the debate feels broken. People defending AI talk about use (“I’m using it responsibly”). People criticising AI talk about sources (“it was built on stolen work”). Both are right, and they’re talking about different layers. Until we separate those layers, we’ll keep arguing past each other.
A Practical Code for Designers Using AI
Since most of us can’t retrain the models, here’s what we can control:
- Don’t prompt living artists by name for commercial work. If you want a style, hire the person or develop your own direction.
- Tell clients when AI is part of the process. Transparency builds trust and avoids awkward ownership questions later.
- Prefer tools with licensed or opt-in training data where the option exists, even if they’re less impressive.
- Keep human authorship at the centre. Use AI for exploration and grunt work, not as a substitute for the thinking the client is paying for.
- Check outputs for close copies. Reverse image search anything that looks suspiciously polished or familiar, especially logos.
- Never fake people or proof. No generated “customers,” no invented testimonials, no faces that could pass for real individuals.
- Credit and pay humans where you can. If a project leans on a recognisable influence, acknowledge it or commission it.
None of this solves the source problem. But it stops you from adding to it.
Conclusion
The hammer comparison is useful because it shows us what we’d like to believe: that tools are neutral and only people are guilty. For a real hammer, that’s mostly true.
AI isn’t a hammer made of steel. It’s a hammer made of other people’s work. Designing with it can be thoughtful, creative and completely ethical. But the ethics of AI in design broke long before the first designer typed a prompt. It broke at the source, when the work of millions of creators was treated as raw material instead of as someone’s labour.
The fix won’t come from designers giving up AI. It will come from demanding tools that respect where their knowledge came from, and from using the tools we have with our eyes open.
Ask Yourself
Before your next AI-assisted project, run through these:
- Would I be comfortable showing the client exactly how this was made? If the answer is no, something’s off.
- Am I borrowing a style, or borrowing a person? Mood and influence are fair game. A living artist’s name in a prompt usually isn’t.
- Could someone recognise their own work in this output? If yes, you’re not done.
- Is AI doing my thinking, or helping it? Clients pay for judgment, not generation.
- Do I know anything about how this tool was trained? If not, is there a more transparent alternative?
- If a tool were trained on my portfolio without asking, how would I want it used? Hold yourself to that standard.
The tool won’t ask these questions for you. That’s still your job.