ADVERTISEMENT

Easter Giveaway – Enjoy 1 Year of the Best AI Image Enhancer for FREE Whether you’re restoring old family photos, enhancing downloaded images, or preparing pictures for printing, sharing, or preserving memories, Aiarty makes it effortless for everyone Why Aiarty Image Enhancer? Upscale images up to 800% to stunning 4K/8K/16K/32K quality AI models optimized for portraits, landscapes, and intricate textures 4-in-1 tool: Upscale, fix blur, remove noise, and enhance facial details Restore fine details in skin, fur, feathers, and more Get your FREE 1-Year License now and start enhancing your photos. No tech skills required.

MEDIA

By Catherine Barrett • Comment about this article

The horse may have five legs, but it’s already out of the barn.

AI-generated images are here to stay, and we need to learn how to recognize them and use them legitimately. They’re not authoritative depictions of how things look, but they are handy for illustrating ideas. In what follows, I’ll tell you how they work and address ethical and practical concerns.

Now that AI art is here, we are all going to need to become artists, or art critics, in a way that we haven’t been before. Just as people had to get used to photographs in the 1800s and TV special effects in the 1960s, the educated public needs to become aware of AI art. So let’s dive in.

Generative AI makes pictures the way ChatGPT makes English text: by predicting, based on its training data, what to place where. The training set of an image generator is a large collection of pictures and descriptions. AI tries to make a new picture to fit a description you provide.

Easy-to-use image generators include Midjourney, DALL-E 3 (embedded in ChatGPT), Leonardo.ai, and the tools in AI assistants such as Google Gemini and Microsoft Copilot. You can try them yourself. Put in a prompt (a description), and AI will try to draw what you described. In the best case, the result is like a good but simple stock photo (Figure 1).

Figure 1. “A red apple on a kitchen table” produced by Google Gemini.

More often, the result is less precise. Because it’s a neural network, the generator will produce an inexact picture, a whimsical combination of different possibilities — which is often factually inaccurate. But in visual arts, this “hallucination” can be a feature, not a bug. It may introduce elements that inspire the human operator’s creativity in new directions.

AI is not a good tool for fine art or technical illustration. The AI model is not an accurate encyclopedia of what things actually look like. Don’t use it to count the columns on the U.S. Capitol or to make a precise rendering of your latest invention. It translates descriptions into approximate, informal views.

AI does not always render exactly what you ask for. Occasionally, the picture comes out wildly disrupted, so take a good look at it before using it.

More subtly, neural networks are bad at counting, and pictures often contain mistakes with repeated elements (Figure 2).

Figure 2. “A red apple on a kitchen table” by Midjourney

Notice the mismatched spindles on the chair, the off-center drawer knobs, and hard-to-identify objects on the counter. The middle of the picture is fine; you could crop it with any picture editor.

Perhaps you’ve seen the widely circulated picture of President Trump praying, but with six fingers on each hand. Five-legged horses are passé by now, but count columns and fingers carefully.

AI pictures have gotten a bad reputation because too many people have distributed them without saying what they are, or even falsely claiming they are real photographs. Obviously, if you ask for something strange or deceptive, you’ll get it. That’s your responsibility, not the computer’s.

Picture generators are trained on huge collections of pictures and descriptions. Concerns have been raised about training them on copyrighted material. Is AI merely art theft on a grand scale?

In defense of AI, we need to consider three points. First, AI does not simply piece together parts of pictures it was trained on. Unless the training set was woefully small, the images and their structures are broken down more finely and abstractly, then recombined to fill your request. You are unlikely to see a recognizable part of a training image.

Second, real artists train themselves on everything they see — both in pictures and in the real world — and even imitate each other (e.g., in da Vinci’s studio, or the Impressionists).

Commercial art, in particular, is collaborative. While creating the publications and products we interact with every day, working artists purchase clip art and photographs to use as elements of larger projects. Although many entering the AI discussions are imagining a tormented painter in a garret squeezing out a masterpiece, that’s not what most art in our society is, and generic stock photos and simple cartoons are what AI can effectively replace.

Third, artists already use digital tools. The rest of us imagine that a brilliant artist does everything freehand. Actually, real artists use every tool that can help — pantographs and tracing paper in the old days, software to straighten lines and create textures today. Generative AI is just one more.

Still, some questions about copyright and ownership haven’t been properly addressed. Without hammering out legal technicalities, we can agree that AI shouldn’t be trained on images whose owners object. An artist using a distinctive design element might not want it replicated. What’s worse, an infringement might happen in a tool upstream from the artist who seems to have committed it. This doesn’t mean we shouldn’t have AI art. All these concerns can be addressed and fixed without discarding the technology.

Why do some artists not want their work used to train AI? Because once you get beyond the level of simple illustrations and cartoons, creating a work of art requires more human effort than just making the piece, and more than the years learning technique and skills. It requires study of yourself and the world to have an insight — a work of art is the fruit of a whole human life.

For that reason, it is widely recognized that artists have not only copyright (to control republication) but also moral rights: to be recognized as the creator (or not), to decide when to publish the art (or withdraw it), and to prevent work from being altered in a way that changes its meaning or destroys it. Current AI products step on these rights if they replicate something distinctive from a training source.

The Internet interferes with moral rights by blurring the distinction between viewing and copying, and by making it easy to distribute a picture much more widely than its creator intended. These are areas in which society needs to do some long, hard thinking. I am confident that in a few years, acceptable practices will be much clearer, just as people nowadays all recognize that breaking into computers online is wrong (some of us can remember a time when that was not the consensus!).

If the artist creates art by prompting an AI image generator, AI is a tool. The artist’s creativity is the feedback loop between prompt, generated picture, revised prompt, revised picture, and so on. The prompt by itself is not a work of art, merely a wish for a work of art, just as in patent law a stated wish for an invention is not an invention.

For anyone creating AI art, the big ethical guideline is that you are responsible for what you do with your tool. Just as weapons aren’t prosecuted for assault — people are — computers aren’t responsible for whether pictures are defamatory or indecent or violate the rights of the original artist. Responsibility falls on the humans who choose to make them and release them.

As Michael Covington explained last year, the second ethical guideline is that if it’s wrong to do something without a computer, it’s still wrong to do it with a computer. Deepfakes, for example, could be made, at great cost, by expert artists and actors in any era. It was wrong to use them deceptively, and it still is. Even if the technology is cheap and easy to use, the ethics haven’t changed.

As with any other art, depicting real people without their permission is unwise unless they are public figures and it is clear that your image is not a real photograph. That fact can easily be lost if others pass the image along. You can guard against this by asking for a cartoonish rather than photo-realistic style (Figure 3).

Figure 3. “A simple cartoon-like line drawing of an apple” by Midjourney. Simple clip art is a good use for AI.

In fact, that goes for all generated images. If it matters whether anyone thinks the picture is an accurate, actual photograph, be sure to explain that it isn’t! If you’re using AI for the right purposes, it will already be evident that your picture isn’t a photo and this won’t come up.

Contribute your thoughts in this article’s forum!

Catherine Barrett has a degree in
studio art from the University of Georgia and attended the University of Kentucky School of Law. She enjoys exploring new frontiers of computer-assisted creativity.

ADVERTISEMENT

Enjoying the newsletter? Become a PLUS member and get it all!
	Don’t miss any of our great content about Windows, Microsoft, Office, 365, PCs, hardware, software, privacy, security, safety, useful and safe freeware, important news, analysis, and Susan Bradley’s popular and sought-after patch advice. PLUS, these exclusive benefits: Every article, delivered to your inbox Four bonus issues per year, with original content MS-DEFCON Alerts, delivered to your inbox MS-DEFCON Alerts available via TEXT message Special Plus Alerts, delivered to your inbox Access to the complete archive of nearly two decades of newsletters Identification as a Plus member in our popular forums No ads We’re supported by donations — choose any amount of $6 or more for a one-year membership.