Notes on ‘Why A.I. Isn’t Going to Make Art’ by Ted Chiang

Whenever I read anything by Ted Chiang, even if it were just his grocery shopping list, I feel a sense of humble admiration. A restricted set of people, though I’m unsure of the exact statistics, possess an exceptional ability to express complex ideas. This is something I personally find quite challenging – not only in identifying those ideas but also in articulating them. Then there is an even more restricted subset of people who can express through storytelling something that is not even an idea yet, but you can still feel it and sense its significance. Chiang is one of those individuals in the subset of the subset.

I follow the New Yorker on LinkedIn, and it was there that I came across this quote from a Chiang article titled ‘Why A.I. Isn’t Going to Make Art’:

We are all products of what has come before us, but it’s by living our lives in interaction with others that we bring meaning into the world. That is something that an auto-complete algorithm can never do, and don’t let anyone tell you otherwise.

I find this quote insightful, as expected.

The purpose of this short essay is to reflect on the quote, or perhaps more accurately, to use it as a prompt to share my evolving thoughts on the interaction between art and machines. Hence, I’ll go about machines that create or produce art, where the choice of terminology here is particularly meaningful. To set the context, on one side we have people creating art using technologies like canvases, brushes, oil paints, and typewriters. On the other side, we have people “prompting” machines to produce images or pieces of writing. I don’t want to focus on the quality of the outcome. After all, it’s difficult to determine the quality of an artwork, and at the risk of stating something profoundly superficial, beauty and price are in the eye of the beholder. Instead, I want to focus specifically on the creative process.

My pretty weak thesis is that in the long run, there may not be much fundamental difference between the creative processes used by machines versus those employed by human artists. This viewpoint could be seen as somewhat mechanistic in its outlook on human creativity, and we know that ‘mechanicist’ is not a nice term because it gives the feelings of something cold like metal, while humans are warm. However, my aim is to at least begin to unpack and perhaps even justify this perspective, even if I’m unable to get even close here to the complex philosophical questions it raises.

The way I like to see art, without any veneer of originality, is as a process and its outcome. This process brings something into the physical world by translating or representing what the artist thinks and feels. Translating almost makes you think that art is somewhere already platonically incubated in the brain, but this is a detail. The important point is that art brings into the physical world something that was felt or thought by an individual or group of individuals. And in the comparisons with technology, technology is seen as the activity that bridges what can be thought and what can be built in terms of machines or procedures, while art bridges what can be felt and its representation. By the way, just as a parenthesis, that representation uses plenty of technology, from oil paint, to video installation, to a beautiful Olivetti Lettera 22 in the context of writing. But, again, that is not what I am interested in here.

I want to focus on the creative process itself. We still have a limited understanding of the nuances and unpredictability inherent in human creativity. Consider the vastly different styles and approaches used by different artists like Mark Rothko and Xu Beihong – they worked with different materials, subjects, and likely drew upon divergent emotional experiences. This suggests that artistic creation is heavily influenced by both cultural context and personal experiences. Obviously. Artists do not work in isolation; their work evolves through interactions with the world around them, including exposure to other artworks and observations of daily life. An artist who has gone through a personal tragedy, for instance, may channel feelings of grief rather than joy into their creative output during that period. This notion that “we are all products of what has come before us” ties closely to the idea that the artist’s work is the outcome of their individual journey, shaped by both their internal experiences and external interactions. The creative process does not occur in a vacuum, but rather emerges from this complex interplay of influences.

Still, what is the dynamic in the creative process that transforms a small idea, a glimpse of something, into an artwork? Certainly, there are commissioned artworks where a patron or customer requests their family portrait or Leonardo’s Last Supper. Then there’s the poem you write to your partner when you realize they’re in love with someone else, hoping that this swan song might help you metabolize the pain and perhaps win them back – recall Fernando Pessoa:

I write to free myself from the fever of feeling.

Whatever drives the genesis of art, it seems the creative process is built on two phases: accumulation and action. You’ve seen many sunsets, and now you’re going to write about one. Maybe it’s a real sunset you’re trying to describe faithfully, or perhaps an imaginary one filtered through particularly significant sunsets in your mind. But can you trust your mind? Maybe these sunsets are from movies, or your idea of a sunset has little to do with what you’ve actually seen. Fine. Or perhaps you’re imagining a sunset while staring at the ceiling of your room, inspired by a game of shadows cast by streetlights. What matters is that the sunset you’re going to write about or draw is the product of your experience and interaction with the world of objects and people. This is the accumulation phase, whether you’re conscious of it or not. Another suitable term is “storing side information”, which closely reminds us of Shannon’s information theory jargon.

The action phase is the production of the artwork itself. It’s the moment the pen touches the paper. When you paint or write, there’s a good chance you’re not entirely sure what you’re doing or why, but the accumulation drives your hand. You may make many choices, but possibly these are forced choices resulting from the accumulation. And even if you have some good predictions and a list of to-do’s, these actions may be induced by other actions during the process itself, in ways that are hard to predict and fully control. In the end, the human mind is beautifully on automatic pilot quite often. So, what’s the difference between you and a machine that has a lot of information in memory and maybe flips a coin to make choices based on its information and create connections that lead to choices?

In full honesty, do we really need to care at this stage? Futurology can be fun but it’s self-referential. Who will live will see. Still, there are points that are emerging in the context of these discussions. The first one is: how much are machines going to replace human beings in artistic production? The question could possibly be rephrased: how much will people work together with machines in the agency of creativity? “Action” here is connected with what I mentioned above about accumulation and action.

I had a chat with Luca Lorenzini, a highly imaginative person, being the co-founder of the creative agency Small (see, for example, Assume that I Can). He thinks that

one day in advertising, human beings will be like tailors and machines will be like producers of mass-produced clothes,

a metaphor that I quite like. In other words, in the future of advertising, humans will be artisans crafting bespoke campaigns, while AI will mass-produce generic content. Again, who will live will see and it will depend on how much AI will advance and any form of comparison will need to be done on the basis of our future understanding of biological brains.

The second point is: there’s a lot of confusion today between the notion of “something created by a computer” and “something associated with the physical world”. While in the realm of language this distinction is really hard, and its philosophical nature eludes this short discussion, in figurative art, photography and video this is now super common. You hear people saying “look at how real this AI-generated photo looks!”. What people tend to forget is that the ontological weight of objects in the real world is very different from imagined objects. This is an absolutely crucial point. For the sake of representation, like in theatre, a machine-generated image or text can work pretty well, as long as there is a label stating that it was machine-generated and its subject didn’t actually exist in the physical world.

Fixing with a camera or telling a story coming from the real world seems much more actionable than pure imagination, even if this imagination is put together from a patchwork of different things. If you tell me that in that forest I will find wonderful berries, I can go there and eat the berries. They are real and this allows me to take actions. This is why we should be very careful not to deform reality with what we may call bullshit, intended as a technical term (H. G. Frankfurt, On Bullshit. Princeton, NJ: Princeton University Press 2005). Bullshit introduced into reality by machines, even in the form of art, can lead to the wrong actions – again, unless the artistic intent is clearly stated.

To conclude, let’s give a quick look at the photo above. I took it this July at the Palio di Siena. Based on the density and size of the crowd in the image, I’d estimate that there are at least several thousand people present, but it’s difficult to provide an exact number. Can you recreate this with AI? I don’t know, it’s probably pretty simple, and if it’s not pretty simple today, it will be at some point. One may then say “look, the AI-generated image is better / more interesting / more aesthetically pleasant than the photo taken by a person”. Nothing wrong with that.

So, yes, AI’s going to make art, assuming we agree on a definition of art and creative process. This sentence is in neat contradiction with the title of Chiang’s essay. However the difference between the photo generated by a computer and the one taken by a person is that in an exact moment in time, there were a few thousand people physically in a square built in the XIII century or so, all looking in one direction a few moments before the beginning of the Palio, an event some of these folks care about with an intensity that ‘defies imagination’. We can easily disregard the meaning behind the photo, whose imprecise description would fill pages of text anyhow, but the most important point here is in a precise moment of the history of the universe these people were there and the photo is a witness of that, in the same way I can describe my walk this morning to buy bread in the shop. It was a piece of reality, even if a personal one.

There’s also a good chance that indeed AI won’t really be able to produce art until it can interact with the world with a two-way channel and form perception. In the end,

Computers are useless. They can only give you answers

is a quote attributed to none other than Pablo Picasso, someone who we can safely say had a pretty strong connection with art. Maybe to get to the point where computers will produce art, they should start asking questions, but let's agree on the definitions first.