171. ChatGPT, the Chinese Room, and the future of human creativity

Could an AI write a story? Yes, they already exist. I the Road was published in 2018. Here is a list of some others. World Clock was published in 2013, Dinner Depression in 2019. The Day a Computer Writes a Novel was entered in 2015 for the third Hoshi Sinichi award, a Japanese sci-fi competition, and proceeded past the first judging round. There is also a collection of books entirely written by AIs.

None of these stories are perfect, and those that were not edited by humans tend to be rambling and incoherent. AI-generated fiction was still not very good. The plots tended to be prosaic and the characterisation shallow. But the field is advancing by leaps and bounds.

Chat GPT

The third generation of the language generating AI, Generative Pre-trained Transformer (GPT3), introduced in 2020, can hold remarkably human-like conversations and write passable fiction.  You can play with GPT3 and explore its abilities through ChatGPT (though you’ll have to surrender both your e-mail address and your telephone number) or through writing apps such as Sudowrite and Jasper.

The consensus of technical opinion is that GPT3 is “scary good” at tasks such as copywriting, composing essays, and holding human-like conversations. However, it does also make mistakes, so don’t rely on its output. Its makers admit it can create “plausible-sounding but incorrect or nonsensical answers.” This is, perhaps, because according to some critics, it’s very good at putting words into an order that makes sense from a statistical point of view, but with no awareness of the meaning or its correctness. This may be an overly harsh judgment. ChatGPT was good enough to score between B and B- in an MBA exam, though it made a fairly monumental arithmetical error.

Here’s where the field gets philosophically interesting.

The debate

There has been quite a debate about whether an AI might surpass the abilities of a human writer. Below is a flavour of some of the positions from a writing website to which I belong:

“Better than quite a few writers, for sure. I’ve literally seen worse. So, impossible that AI could someday be indistinguishable from a real human? It’s already on par with a lot of real people, if not better than the worst writers.”

“It also doesn’t matter to those looking for literary fiction if the genre readers are getting their books through a computer software program. What these readers want, the software cannot provide.”

“As writers, we are somehow biased by the ethical question – some of us see AI-generated text as a mere tool (as Photoshop is for visual artists), while others consider it cheating. For me, speaking solely as a reader, if the book is good, I wouldn’t mind that it’s written partly or entirely by an AI.”

“I think this is hella cool. It’s at least a basic foundation some writers can use upon which to flesh out their ideas, if they so choose. They’re still writing the story, executing the ideas in their own unique way.”

“I can’t think of any technology that has, or could, replace human creativity.”

“Can it ever deliver emotionally and philosophically illuminating stories in ways that skilled and experienced authors can? Personally, I doubt it, because story telling isn’t just plot. Or characters. Or subplots and twists and opening sentences and all the “rules” people like to clutter their imaginations with.”

“Compared to my very limited life, and the pathetically tiny amount of literature I have consumed, ChatGPT in its current form already has vastly more experience to drawn upon than me. Even with just a short time playing with it, it has written short stories with characters and settings that I could never have dreamed of writing, and come up with ideas that I could never have thought of.”

“I really can’t believe the people who are saying AI will never write better than humans. It literally writes better than me at this moment.”

“But what about meaning? What about the illuminating ideas of self and behaviour and memory and emotion and justice? Do you believe that personal expression – the epiphany of the author in the scenes they write and the meaning they are trying to share with others is something that software can create?”

There are several views here. One holds that AI is a tool for authors, much as dictionaries, thesauri, word processors, grammar and spelling checkers are tools. Another holds that a sufficiently complex AI should be able to write works that would satisfy readers. Still another holds that, while AI may be capable of writing formulaic genre fiction, only a human writer can be truly creative.

The second and third positions are philosophical arguments about what it is to be human. To be human, the third position argues, is to attach meaning to things and manipulate them symbolically to create new things. The second position implicitly denies there is anything particularly special about creativity: that it’s just a highly complex set of mental operations.

Brain and consciouness

Let’s explore these two positions about humanity. I acknowledge from the start that machines are not conscious (at least not yet) and do not “understand”. GPT3 is a language program trained on a huge data set of writing. There is a reason that understanding consciousness is labelled “the hard problem” by philosophers and neuro-scientists. We know quite a lot about what brains are and how they work, but consciousness has evaded scientific explanation (to date). So machine learning is not capable of understanding meaning. Instead, GPT works by detecting language patterns, following rules it has generated about what words are likely to follow other words.

I’m going to present three concepts here that may help in unravelling the problem. The first is the Turing Test; the second is the Chinese Room problem; and the third is the role of metaphor in creativity.

First, the Turing Test. Proposed in 1950 by the mathematician Alan Turing, the test assesses whether people can tell when they are conversing with a machine. If the evaluator cannot reliably tell the machine from the human, the machine would be said to have passed the test.

Second, the Chinese Room. This is a 1980 thought experiment by the philosopher John Searle in rebuttal of the Turing Test. He imagines he is a sealed room with access to the instructions used by a language computer which can answer questions in Chinese. Questions are fed in through a slot and he follows the instructions, enabling him to write out entirely correct answers without speaking a word of Chinese. This, Searle argues, is what artificial intelligence is doing. You will see that my position concurs with Searle that the machine does not “understand” anything.

The question is, does it matter that the machine understands nothing? Since we don’t know what consciousness is, we can’t measure it directly, but we can infer (though again without proof) that other people possess it. If our judgment of a respondent’s humanity is all we can rely on, we would have to conclude that the ability to perform as if conscious is indistinguishable from being conscious. In the case of creative writing, the reader’s response is the arbiter. The Turing Test becomes: could a sufficiently discerning reader tell that a piece of fiction was written by a computer? Already, this may be difficult and will certainly become more so as AI advances.

The nature of creativity

This brings me to the third element: the nature of creativity. The quotes from the writers’ discussion above contain the view that while a machine can follow the rules of a formula, it would be incapable of investing this with original meaning and creativity. Let us grant the fact many readers enjoy repetitions of formulae. That is what the strictures of genre mean. There is no shortage of formulae available to writers. The Hero’s Quest is among the most popular. So let’s consider only writing that possesses greater literary “depth” and that explores complex meaning.

Where does that depth and meaning come from? It would, in principle, be possible to write a set of rules for deep writing by specifying what the meaning behind the story is, and some recurring motifs to express this. But would a machine be able to use these effectively and creatively? What is creativity? The Cambridge Dictionary defines it as “the ability to produce or use original and unusual ideas”, which is good enough for my purpose here. Understanding creativity is, arguably, almost as difficult a problem as consciousness. But there are techniques and routines for developing the habit of creativity, such as Edward de Bono’s methods. If you’re stuck in thinking, try to add a wild card to free up creativity (for example, “how could you use spaghetti to solve this problem?”). One of the GPT apps, Sudowrite, offers a facility for “adding a twist” to a story.

Most spaghetti ideas don’t work, but a few do. And exploring them frees up creativity.

 I want to finish by suggesting a mechanistic answer to how creativity works, which is an extension of the spaghetti idea. It’s not my concept but one developed by Donald Schon in his book Invention and the Evolution of Ideas. He argues, and this pleases me as a writer, that metaphor is at the root of creativity, whether in the arts or the sciences. A metaphor or simile relates unlike things (“my love is like a red, red rose”). We know they’re unlike, but in conjoining them, our sense of each of them changes, the one illuminates our understanding of the other. James Clerk Maxwell used the well-understood properties of waves to explore the mathematics of electricity and magnetism and uncovered the physics of electro-magnetism. He used water waves as a metaphor for electrical waves.

If there are, indeed, “algorithms” for creativity, a machine should be programmable to replicate it.

 In the skies above the port, the neon lights and holographic advertisements flickered and pulsed like the synapses of some vast, artificial brain, the electric nerves of the city stretched taut against the darkness. The port itself was a glittering hive of activity, a mass of chrome and steel, the beating heart of the sprawling metropolis.

Was this passage written by a person or a machine? It has metaphor. As does this one:

Sometimes the scattered thoughts of their deaths run like a jagged red seam of fire inside me and I burn from the inside out, like a lightning-struck tree; the outside whole, the inside, that carried the lightning’s charge, a coal. At other times, I feel empty, transparent, a child of the wind…they are gone, I tell myself. Nothing comes back

One of these passages was written by ChatGPT. The other by a prize-winning human author. Can you tell the difference? It would be great to hear your decision and the reasons for it.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s