As AI-generated images and AI-written novels and poems spread rapidly online, the debate over whether these creations truly deserve to be called "AI art" shows no sign of settling. This piece lays out the core criterion at the heart of that debate, and offers a practical yardstick for judging the AI-made works in front of you today. The short answer: what separates AI art from art isn't the polish of the finished piece, but the nature of the activity that produced it.

What Separates AI Art From Human Art

The moment an AI image or an AI-written novel looks impressively polished, people tend to judge whether it counts as art by that polish alone. But the more fundamental question is what kind of activity produced it. AI, at its core, works by finding an answer to a given question. Typing a desired style or composition into a prompt is itself a question, and the AI simply returns the most plausible answer to it. Finding an answer to a question and making art operate on different levels. Art isn't about finding answers — it's closer to wrestling with something that hasn't even become a question yet. Understanding this distinction settles half the debate over AI art. The other half only becomes clear once you look at how AI actually works.

Where Does AI's Work Actually End?

There's one practical thing worth checking when judging AI art: exactly when the work comes to an end.

Where AI's Work EndsEnter a PromptGenerate Image/TextActivity Ends

The moment you enter a prompt and receive a result, the work is over. Even if the output falls short, the AI doesn't reconsider on its own, try a different direction, or question the original prompt. A new attempt only begins when a person goes back and rewrites the prompt. When a person paints or writes, by contrast, the activity continues even after a result is produced. If they're not satisfied, they ask themselves why, and that question reshapes the direction of the next attempt. AI work has no such question built in. Entering a prompt and receiving a result is a single, self-contained unit, and the next attempt only starts when a person poses a new question. That's why, when judging AI art, it's more useful to look not at a single output but at whether anything follows from it.

What Happens When AI Runs Into Something Unfamiliar

A common counterargument in the AI art debate is that AI, too, produces increasingly creative results as it learns. And it's true: AI trains on vast amounts of data to generate plausible patterns. But when it encounters something unfamiliar that its training data can't explain, instead of wrestling with it to find a new answer, the AI simply files it away as another "exception." Rather than reaching beyond the data to draw out something that isn't in it, the AI flags the case as unprocessed and moves on. Given that art often begins precisely where existing frameworks fail to explain something, this difference matters. Classifying something as an exception and moving past it is an entirely different activity from digging into that exception to generate a new question. An approach that treats the unfamiliar as a mere exception to be filed away rarely arrives at a question no one has asked before.

How to Judge the AI Work in Front of You Right Now

Here's a practical set of criteria to apply the next time you encounter AI art. First, ask whether the piece is a single answer to a single question, or part of a process that generates its own next question. Second, check whether the activity ended with one round of prompting and generation, or whether something followed from it. Third, look at whether the piece let an unfamiliar problem outside the existing framework pass by as a mere exception, or confronted that unfamiliarity head-on. These three criteria apply equally whether you're looking at an AI image or an AI novel. Whether or not to call something "AI art" is ultimately a judgment call each person makes using these standards, and keeping them in mind means that the next time a similar debate comes up, you won't be swayed by surface polish and can reach your own conclusion. The judgment starts with a simple question: does the AI work in front of you stop at an answer, or does it lead to a new question?