BoI - AI Content: Genius or Garbage

Let me begin with the strongest argument my opponents will make. Yes, there is an enormous amount of AI-generated garbage. But so what? Most writing is garbage. Most photographs are unremarkable. Most music is forgettable. Most things posted on social media are rubbish. Give everyone an iPhone and you don't get Cartier-Bresson; you get thousands of photographs of someone's lunch. Nobody argues that the camera is a bad technology because most photographs are bad. The answer, my opponents will say, is to learn how to use AI properly. Don't type “write me a good speech”. Ask better questions. Give it better prompts. Iterate. Challenge the answer. Edit it. Use the machine as a collaborator rather than a vending machine. There is a lot of truth in that. But I think it misses something important. The problem isn't simply that AI makes it easy to produce bad content. The problem is that AI makes it extraordinarily easy to produce content that looks good without necessarily having a good reason to exist. Those are not the same thing. Consider photography. The fact that most photographs are uninteresting doesn't diminish Cartier-Bresson's Behind the Gare Saint-Lazare. We know why that photograph exists. A particular photographer was in a particular place, saw a particular moment and decided that it was worth capturing. The photograph embodies a human act of attention. The same is true of music. The existence of millions of people playing badly doesn't diminish a performance by Glenn Gould. The imperfections of human culture aren't the interesting thing about human culture. The interesting thing is that somebody cared enough to make a particular choice. And this is where I think generative AI is fundamentally different. An AI can help a human make a choice. It can even make choices on our behalf. But it doesn't begin with a life from which those choices arise. Consider Anna Ridler's Myriad (Tulips). Ridler created thousands of photographs of tulips herself and used them as the basis for a machine-learning system. The resulting work is interesting not simply because a machine generated images of flowers. It is interesting because there is a human idea behind the entire process — questions about data, beauty, value, scarcity and the extraordinary history of the tulip market. The machine is part of the work. But it isn't the reason the work exists. The artist supplied the question. That distinction matters because the strongest defence of AI art often turns out to be: Yes, but you have to use a human being properly. The more interesting the question, the more purposeful the selection, the more rigorous the editing, the more interesting the resulting work can become. But notice what that means. The strongest defence of AI creativity often depends upon restoring precisely those things that AI itself does not possess: human intention, taste, judgement and a reason for making something. And there is another problem: scale. Of course humans have always produced rubbish. But humans have also been constrained by time, money, skill and attention. A novelist cannot produce 10,000 novels before breakfast. A photographer cannot take a million carefully composed photographs in an afternoon. A musician cannot record 50,000 songs simultaneously. AI can get much closer to doing exactly that. That changes the cultural environment. When the cost of producing competent-looking content approaches zero, the problem isn't simply that we get more bad art. We get an enormous quantity of perfectly competent material competing for our limited attention. And “slop” is a useful word because it captures this distinction. Slop doesn't necessarily mean something technically terrible. It means something produced without sufficient reason, thought or discrimination and then pushed into the world anyway. That is why “just learn better prompting” isn't quite an adequate answer. Better prompting can improve the output. Human editing can improve it further. But neither automatically gives the machine an experience, an obsession, a point of view or something at stake. A clever prompt can ask an AI to write about grief. It cannot give the AI someone to grieve. It can ask for a painting about mortality. It cannot make the machine aware that it is mortal. And perhaps that is why AI-generated prose can be so uncanny. It can be eloquent about almost anything because it doesn't need to have a reason to care about any of it. Now, I am not saying AI cannot help produce great art. It plainly can. Nor am I saying human-made content is inherently good. Most of it isn't. I'm saying that the existence of human garbage doesn't tell us that human creativity and machine generation are culturally equivalent. The fact that most photographs taken on an iPhone are boring doesn't mean the photographer and the camera are interchangeable. The camera is a tool. The photographer is the person with something to photograph. AI changes that relationship because it can generate the photograph, the story, the argument and increasingly even the apparent reason for making them. So perhaps the real danger isn't that AI will fill the world with terrible art. It is that it will fill the world with something much harder to escape: art, writing and images that are good enough to consume, but not important enough to have been made.