Artificial Intelligence and the Linguistic Infinity

Why AI can write a novel, but it still can’t say anything new

Photo by freddie marriage on Unsplash

One of the most fascinating components of language is the sheer variety of means that people have for getting their ideas across. Perhaps one of the best analogies for thinking about our options at any given moment is to compare them to a chessboard.

For the first two moves of the game, there are 400 possibilities for the different positions that the figures on the board could take. But by the time each player has moved 4 times each, that number expands to a staggering 202.5 billion.

Once each player moves 5 times, the number of permutations reaches 182.25 trillion. It can be hard to truly appreciate exponential growth, and understand the way that even powers of 2 can quickly turn into a quintillion.

But few pursuits give us a clearer glimpse into the inconceivable sums we traffic in throughout our daily lives than when we combine words on a page.

According to some estimates, the average person’s vocabulary extends to a command of about 20,000–35,000 words. Going off of even the lower range, there are already 400,000,000 possibilities at our disposal in even a two-word sentence.

Once the sentence reaches a mere 4 words long, there are already 160 quadrillion distinct roads for us to have taken. And that’s without even getting into different punctuation marks.

Considering that most sentences contain a healthy fifteen to twenty words, a period, and a comma or two in the mix, it’s clear that each of our essays commands some pretty remarkable infinities. There are roughly 3.28×10 to the seventieth power possibilities for the fifteen-word sentence.

To reach a 1 with a hundred zeroes (a googol), it would only take an additional eight words. In other words, there are more possible ways to formulate one single sentence than there are atoms in the observable universe.

The mind can’t fathom the sheer multitude of possibilities for how each novel or page could be composed, let alone every individual paragraph encased within each. Perhaps this is one of the reasons why artificial intelligence still vehemently struggles to mimic the human voice.

Even while these AI services are trained on over a trillion parameters, as is the case with ChatGPT, they still struggle to parse the weight of each word embedded within each sentence. The nuance of language is lost. Even as these services have grown by leaps and bounds with each passing iteration, they still can’t effectively imitate a writer’s style. There are just too many components to each author’s voice.

When there’s hardly a sentence that boasts less than a quadrillion routes for its possible construction, ChatGPT’s abilities begin to suddenly seem finite.

In addition to having its limits, the way that it processes language is also far different than a person would. While the trillions of data points that these AI services have access to far surpass the amount of knowledge any singular human mind can house, it takes that vast composite and dilutes it down into cold and binary averages.

When generative AIs try to write something new, they don’t pull from the separate experience they’ve had with every book and website they’ve ever visited. They don’t visit websites. They don’t have experience. They compile every bit of text they’ve processed into patterns. They determine the roads most likely to be taken with algorithmic precision. But in their precision, they falter.

Everything they produce is based squarely on what’s already been written and created.

Without those personal experiences coloring its words, there’s a hollow and stilted air that over-arches everything AI manufactures. Whether it’s prompted to write poems, essays, or diatribes, the results rarely come across as more than mere robotic drivel. There’s a predictability to the tones used in its explorations of nearly every subject. It “dives” into topics and “weaves intricate tapestries” like a million theses combined into one. It has no voice of its own. The voice it does have, it owes to a vast reservoir of anonymous others.

There’s a universal sound to the singing audience. As with the twenty colors of paint that blend into a monotonous brown when mixed together, there’s a consistency to the sound that emerges when ten-thousand voices function as one. The male combines with female, the amateur with the expert, and the old with the young. The result is a tonal conglomeration that remains largely the same across states and borders.

But each time the crowd functions as one, the individual voice is lost in the noise. The idiosyncrasies in each vocal cord become one with every other.

If ten-thousand separate voices went into crafting a single novel, each individual writer would lose what makes their use of language unique. Their vocabulary, experiences, tones, preferences, and character would each meld into a bland monotony. Prosaic language would cancel out the poetic. The terse and laconic would cancel the dense and long-winded. The safe would negate the adventurous. Formulaic would supplant creative.

That’s precisely what happens when AI tries to recreate our writing. The infinities buried in each voice — in each and every sentence — blur into a colorless whole. The impossibly large amount of roads before us are diminished into only the most traveled boulevards. Souls are lost to the droning mass of words already written.

The essence of human creativity lies in choosing the path less taken. It’s the dusty and unexplored back roads that lead to groundbreaking ideas. When creation is left to the hands of detached algorithms and averages, individuality falls by the wayside. What emerges in its wake can scarcely even be called creation.

---

This article was originally published on Medium.

---

If you enjoyed this article, you can support my work on here for under $2.00 a month. It would make an enormous difference in helping me to bring you the quality writing you deserve during these times when journalism is under attack.