The AI Art That Speaks for Itself

Every new tool starts out producing the old things. The first automobiles were ‘horseless carriages.’ The first things Gutenberg printed were copies of handwritten religious texts. Most early photographs of people were staged like painted portraits. 

We’re at that stage with AI now. While everyone is complaining about the quality of the product (It’s slop! It’s soulless! It’s completely inaccurate!), I think we miss the deeper issue, which is this: Almost everyone is using AI to do the same thing they did before, just faster/cheaper/easier. 

This will change. Remember that the first few years of television looked very much like filmed stage plays. The Honeymooners could have easily been a Broadway production with essentially no changes. Eventually, though, people figured out a whole new language for TV shows, one that took advantage of the possibilities of that medium. Right now people are using AI to write their emails, create their songs, paint their Ghibli images. But better, more interesting things will emerge. 

What would an “AI-native” piece of art look like? What things would you do to take advantage of a tool that has infinite capacity and minimal constraints? Will art start to look like, say, the modern Web page, which is assembled dynamically and on demand—that is, every time someone loads a web page it is built anew drawing from dozens of different data sources to create the single, mutable thing that lives at the end of most modern URLs?

One thing I keep coming back to is this: Art that—literally—speaks for itself. Consider the “Artist Statement,” an explanation of the work written by the artist. What would it be like if I as a viewer could engage in a conversation with an object of art—a painting, a piece of music, a book, a film, doesn’t matter what form it takes—about the author's intent? A piece of art that can answer questions about itself, that could point out things the viewer might have missed, that could suggest interpretations that other people have made, would be a really interesting use of LLMs. 

It’s something we’re noodling on as we build; would love to know your thoughts!