AI and Art

Can AI create art?

First, art is a means to express thoughts. The present day AI cannot have thoughts. This may or may not change in the future. But this means, just like the camera or the brush, AI is currently a tool to express art. Efficient in some scenarios and in some others, not so much.

Camera changed art. It became a more efficient tool to capture live subjects. It pushed some painters to more abstract art forms, or focus on painting ‘thoughts’, something camera is not efficient to do.

AI changed this. AI can now be used to express thoughts as well. How will it change painting as an artform now? This brings me to the second point.

Words and AI

AI will push painters to create art that are not easily captured in words, and hence cannot be created with AI using word prompts.

The parameters space of images that we can draw efficiently is much larger than the parameter space of words that we can type efficiently. Hence, using word prompts to generate art is not efficient. It is not possible to efficiently access/explore most regions of images using words.

Let me justify through a calculation. Consider a 1024 x 1024 pixel image, where each pixel is just black or white. Size of this image would be 1/8 megabyte. This is roughly $10^5$ characters, or about 20 thousand words.

You will need to write a novella to be able to generate all possible low res black and white pixel art. Best of luck. This is the reason why going blind is so devastating. Words cannot replace paintings. Paintings cannot replace words.

The reason why AI art is so beautiful because they are used to generate low effort paintings that can be easily described in terms of words. For example, if i prompt, make a ghibli style painting, the output will look good. But this is because Miyazaki did all the heavy lifting and defined what the term ‘Ghibli art’ means. My prompt is low effort.

And most importantly, in AI art, the map between the input and output is poor. The output is dashing, but is it what you had intended? Try having a clear image in mind and have ai recreate it, and you will need to sweat to achieve it. Not just because the AI tools are not advanced yet. Imagine seeing an image and trying to recreate it by describing it to an expert human artist vs the artist getting to see the image himself. The objective here is to recreate the image as closely as possible and clearly, the first method is less efficient and not because the artist is not intelligent or talented.

Image to image generation using AI

Since the parameter space of images is not efficiently accessible with words, the future is image prompt to generate images. But the best case scenario, of course, is to be able to communicate directly to the AI through neurons, the way our brain communicates with our hand. Something like neuralink. Use of any intermediatory, like word or image prompt, will lead to loss on the control over the output.

Conclusion

Using brush, camera, word or image prompt to create images are all different artforms. Different tools. Some tools may become redundant over time.