Art, intelligence, and purpose

I can say that I am a frequent user of AI tools, particularly chatbots and various related APIs. This is largely a consequence of my work as a research software engineer. I subscribe to at least one “Professional” plan from OpenAI or Claude, although I could do without it given the rather high cost for a monthly subscription. However, I’m curious and I like to experiment. I consider them potentially useful tools, but sometimes they’re more like interesting toys because we don’t know what to expect when we use them. Or we might have an intuition, but all the consequences aren’t very clear to us.

A few weeks ago, I was chatting with a colleague who is always very (perhaps too) enthusiastic about AI, language models, and the privacy of data that made it possible to build these tools. At one point he said (I’m paraphrasing, I don’t remember his exact words): my written English is terrible, but I don’t care that much. I write what I need to write, then I feed it to ChatGPT or Claude, and ask “fix my English prose.” So I no longer need somebody who knows English better than I do to review what I write. I replied him: are you sure you’re actually learning to write better this way? Maybe he doesn’t care because writing doesn’t stimulate him that much or it’s not among his duties.

Another example. I wrote fiction (mostly short stories that we might call “speculative fiction”) for several years, and I still like to try. I have dozens of ideas (everyone does), few of which are truly worth exploring, and even fewer that I seriously start working on. But I’m not a writer—I don’t practice writing fiction as often as I say I want to, so my narrative muscles are somewhat atrophied. I know well that there’s only one solution to this problem: write. But where to start? How to effectively structure a potential story that starts from a genuinely interesting idea? An idea is just an idea, perhaps accounting for five percent of the finished work (if one ever gets there). So I turned to Claude, the responsible, honest, and harmless AI, as Anthropic claims, one of OpenAI’s many competitors. I even created a dedicated project called “Fiction Writing,” where I could collect documents to discuss with Claude.

Beyond the obvious question “what results did you get?”, I think there’s an even more interesting one: why is it so difficult to write something? Here I’m talking about fiction, but the question can extend to journaling or blog posts (with some differences). Ignoring for a moment more tedious writing tasks, the heart of this question is: writing is an act of artistic creation. A very vague and debatable phrase unless we define “art.” Ted Chiang comes to my aid:

Art is notoriously hard to define, and so are the differences between good art and bad art. But let me offer a generalization: art is something that results from making a lot of choices.

This definition is broad enough to include even writing emails or other forms that have little to do with what we call “creative writing.” Chiang offers an interesting and profound reflection on what it means to create something in an artistic sense; he tries to argue why no currently available AI is capable of doing such a thing.

When I start writing something, I continuously have to make choices: from the larger ones pertaining to the architecture of the story, to the smaller and almost invisible ones, such as choosing one word rather than another or where to place certain punctuation. And each of these choices matters; none exists as more important than another because all contribute in some way to what will be the final result. Convincing oneself that to write something meaningful it’s enough to be skilled at formulating an effective prompt for any of these language models (or LLMs) is a dangerous assumption. Any prompt, even a very detailed one, will have about a few dozens words, rarely hundreds, practically never thousands or more. The first draft of the beginning of a recent story of mine contains more than two thousands. And I’m talking about words, not characters: where to place periods, commas, or colons matters just as much as choosing the right adjective. It’s an enormous number of choices even for an incomplete, initial text that will be rewritten who knows how many times.

No matter how incredible and completely original an LLM’s response to the most refined prompt may seem to us, it takes away from us the fundamental toil common to any artistic endeavor: the effort of making choices and all the reasons behind each of them. Delegating this effort to a machine that has been programmed to optimize certain well-quantifiable objectives gives us the illusion of having found a valid solution to the problem, but it is almost always the wrong one. Chiang continues:

Any writing that deserves your attention as a reader is the result of effort expended by the person who wrote it. Effort during the writing process doesn’t guarantee the end product is worth reading, but worthwhile work cannot be made without it.

But effort oriented toward what purpose? Why does writing feel exhausting? We can discuss this endlessly, but one of the purposes of writing is to communicate something; this purpose is actually common to all kinds of art. We humans have invented various languages to communicate with each other: verbal, non-verbal, written, visual, musical. They all have the same purpose: to convey something, whether it’s an idea, an emotion, a problem, a worldview. Here’s Chiang’s central question: if the intention to communicate is missing, can we really say we are using a language? His answer is no. The very name “Large Language Model” is inappropriate. No AI has the intention to communicate or to do anything. The lack of originality in a text is not the fact it’s composed of (statistically) predictable words, arranged in an order already seen or adhering to a particular style. The absence of originality is the absence of intention. Language, by definition, requires an intention to communicate.

It takes little now to ask another question: does the use of language presuppose a form of intelligence? In other words: is the ability to use language an inherent characteristic of intelligent beings (or agents)? The discussion becomes even more complicated because defining intelligence is way more difficult than defining what art is¹.

The computer scientist François Chollet has proposed the following distinction: skill is how well you perform at a task, while intelligence is how efficiently you gain new skills.

I would add another one that I found at the beginning of Nello Cristianini’s book “La scorciatoia” (the translation is mine):

We will informally define intelligence of such agents as “the ability to behave effectively in new situations.”

By agent Cristianini means “any system capable of acting in its environment, using sensory information to make decisions.” I would say that the two definitions are similar. Chiang adds that Chollet’s definition is much more accurate than that produced by various IQ tests because it applies to non-human forms too; for example, animals, but we could also include plants.

If intelligence means being able to learn new skills quickly, then not even extremely advanced software like AlphaZero (or AlphaFold), two of the most famous products developed by DeepMind, can truly be considered intelligent. They are highly skilled in the area they were programmed for, but nothing more. This doesn’t negate how enormously useful they have been and will be—for example AlphaFold’s in helping biologists and other scientists understand and design the function of certain proteins.

Whether you are creating a novel or a painting or a film, you are engaged in an act of communication between you and your audience. What you create doesn’t have to be utterly unlike every prior piece of art in human history to be valuable; the fact that you’re the one who is saying it, the fact that it derives from your unique life experience and arrives at a particular moment in the life of whoever is seeing your work, is what makes it new.

If intention is missing, then any effort is in vain. And there’s more: when we don’t have a true intention, it becomes impossible to justify any effort at all. Why should I continue to endure the torture of walking twenty miles a day if there is nothing worth walking toward? It might be enough for me to simply arrive there, wherever that place may be, or I might need a deeper motivation. Whatever it is, it cannot be absent. If we are talking about any form of art, the primary intention can only be the desire to communicate.

To my knowledge, no algorithm or AI tool is able to express a genuine desire or intention to do anything. It’s possible that this may change in a more or less distant future, but today no such thing exists. And as Chiang reminds us, don’t let anyone tell you otherwise.