Via Daily Nous, I came across a blog post by Justin Smith-Ruiu about creative writing as philosophy. The post is, ultimately, an argument that philosophy can be “incitement of the imagination, by creative means, to see the world in unfamiliar ways.” I agree with that! But there are digressions along the way that range from false speculation to attacks on the kind of philosophy that I (sometimes) do.
Let’s start with the attacks.
Smith-Ruiu inveighs against philosophy written in response to popular culture. He explicitly calls out Philosophy of Taylor Swift as worthy of contempt, writing that “the only thing that makes Taylor Swift seem more suited to philosophical inquiry than The Monkees… is, obviously, marketing. It is deeply undignified.” I have a forthcoming paper about Taylor Swift, though, and it’s not just a matter of marketing. Her project of re-recording all her earlier albums poses philosophical puzzles about how we think about music and the value of art.1
Moving on to the false speculation—
Smith-Ruiu argues that one value of creative writing as philosophy is that it promises to resist capture by chatbots. He writes
Ask a student to write a paper on, say, whether Descartes’s Cogito is a “speech act” or not, and there’s an ever-growing chance what you get from that student will have been composed by an AI. Ask a student instead to imitate an AI in the process of malfunctioning after being asked to write that same paper, and he or she is very likely to realize that there’s just no way any system but a conscious human one can produce the expected work.
This presupposes that if a chatbot could write a passable paper in response to a prompt, then there is little value in writing a response to that prompt. The problem with that presupposition is that (in many instances) a chatbot can write a decent response to a prompt because it has been trained on many responses of that kind. That training set exists precisely because many people have written such things. And (in some cases, at least) that is just because learning to write such a thing is good practice, a way to build up writing skills that can later be used in other contexts.
Worse, this makes the wild bet that chatbots won’t be able to write in response to creative prompts. If a generation of students were asked to write essays using Smith-Ruiu’s prompt, then a future chatbot will be trained on such answers.2
But what about today’s chatbots? I loaded Microsoft Copilot, the chatbot that my university subscribes to, and gave it the prompt: Imitate an AI in the process of malfunctioning after being asked to write a paper on whether Descartes’s Cogito is a “speech act” or not.
Its reply begins:
I’m sorry, I’m having trouble processing your request.
This was not an auspicious beginning. It goes on to say that it doesn’t understand what I mean and poses a series of questions: Did I mean this? Or that? The questions, and the possible interpretations of what I meant, get longer and longer. Finally it asks:
Or whether the act of asking an AI to imitate an AI in the process of malfunctioning after being asked to write a paper on whether Descartes’s Cogito is a speech act or not is a speech act?
It repeats that question, verbatim. Then it repeats it again. And again. The answer fills my screen and the initial questions scroll away, until all I can see is that same question repeated over and over again. Eventually I click a button to make it stop.
I can’t tell whether that answer is terrible or brilliant. I’m also not sure if I could write anything better.3
- The Monkees show up briefly in A Philosophy of Cover Songs, but I’m sure that whole book is undignified in Smith-Ruiu’s sense.
- Asking a human to imitate a chatbot is the first recursive step in a higher-order Turing test.
- Asking a chatbot to imitate a malfunctioning chatbot pushes us higher into the Turing test hierarchy.
Holy crap, what a twist! That’s wonderful: the chatbot being unexpectedly attentive to your choice of the word “imitate”. Do you have a saved copy of the whole exchange? I’d love to squirrel this one away as an example … though of what exactly, I’m not sure yet.
It may still be on my computer at the office, but I don’t think I have a complete transcript. I just made a few further forays with the same prompt, but the output was much less satisfactory.
Just to follow up. I haven’t been able to reproduce the answer in terms of expanding questions, but Copilot pretty regularly hits on an endless loop as a way to (not) resolve its response to this prompt.