Wordle, AI and Writing as Gaming

Recently on Facebook, my friend Ken Smith, who retired a little while ago as an English professor from Indiana University at South Bend, shared his customary opening on Wordle, which generated other suggestions. David Hamilton, long-ago director of my undergrad thesis and past editor of The Iowa Review, suggested “adieu,” which knocks out four vowels. Now that Wordle reports have displaced food pictures on social media, strategy sharing (already distilled into articles) is frequent, which got me speculating about where gaming the game was headed—and why people play language games (pace Wittgenstein), including the fundamental one of writing.

For the unfamiliar, Wordle is an online word-guessing game. Players encounter a blank grid, five spaces across, six rows down. The goal is to guess a specific word in as few tries as possible. For their first attempt, players enter a five-letter word into the top row and immediately get feedback on their guess: a green square for a correct letter from the target word in the correct position, yellow for a correct letter in the wrong position, gray for a wrong letter. Using that information, players enter a new word in the second line, again get feedback, and so on, until they’ve guessed the right word before a sixth and final try.

I recognized that Wordle is primed for artificial intelligence solutions. One might easily create software that draws upon the frequency of letters in English words, the frequency of phoneme clusters (“ck” or “qu” or “st,” for example) and a dictionary of all five-letter words to create optimized guesses: if the second letter is an “r” and the fifth is a “k,” then possible words are “break” or “creek” or so on. With a little more fussing, the program could shake hands with the Wordle site and play automatically, leaving the player nothing to do but watch and paste the results on Facebook. It would be like watching WOPR play tic-tac-toe or Global Thermonuclear War in that old Matthew Broderick movie War Games, though with lower production values and without Ally Sheedy.

Of course, I learned that programmers were way ahead of me, offering numerous solving tools, including this one at the aptly named site Not Fun at Parties. Now, I completely respect the pleasure of developing software. Further, I can see the appeal of using hints, as with an obstreperous crossword or a baffling bridge hand. And I suppose, at some level, I can even understand watching an occasional autoplay in hopes of discerning strategies.

But I can’t imagine why typical Wordlers would embrace algorithms that fully solved puzzles. Sure, there’s an element of social media competition, my wanting to show how my three-move solution beats your four. But the experience of actually playing the game—the process—is why people play, isn’t it?

That brings me to writing. For a decade I’ve been watching the evolution of “writing” by AI programs. Enter some parameters, perhaps some data, and let the AI generate a text. I started watching one company that I caught in 2012, Narrative Science, which then advertised products that helped “companies leverage their data by creating easy to use, consistent narrative reporting—automatically through our proprietary artificial intelligence technology platform.” Narrative Science has apparently turned a good business in a decade, having been acquired by Salesforce (and united with Tableau) in December 2021.

Among more promising AI writers is OpenAI’s GPT-3, which The Guardian in fall 2020 famously had write an article to convince readers that the robots come in peace. A 2019 New Yorker article by John Seabrook reviewed the state of AI writing for a lay audience, explaining how GPT-2 was “designed to write prose as well as, or better than, most people can.” GPT-3 is now the engine behind commercial products such as AI Article Writer 2.0, which promises to “write full-length high-quality blog posts and other long-form content within seconds” with an input of “a minimum of two to five words.” And, of course, if writing all that stuff strikes some as onerous, so must be reading it. To the rescue come other species of AI to do the reading for you. For example, TLDR This promises to cut wasted time by summarizing long texts. When I entered Seabrook’s article, the program spit out a six-paragraph version and bragged, “Time saved: 35 mins.” Simply extracting a few sentences and generally missing the point (let alone the experience), the summary was terrible.

But my interest is less critiquing the quality of AI writing—or reading—than wondering at the motivation for it. To be sure, I’d be more than happy to avoid certain kinds of writing, such as my annual report or literature reviews for grant applications. Some of the earlier AIs were built to handle routine and mundane tasks like generating stories of Little League baseball games from box scores, and while I could argue that such writing is good for you in the way that planking is—building writing muscles—I withhold blanket condemnation. In a 2021 Inside Higher Ed blog post, John Warner pointed out an intriguing use of GPT-3 to critique writing assignments, suggesting that if an AI can perform it well, an assignment is most likely not worth making.

Two Categories of Writing

Several of us in the discipline of writing studies cleave the world into two broad categories: obliged and self-sponsored. Obliged writing has consequences if you don’t do it: you fail an assignment, don’t get tenure, don’t get paid, lose a health insurance appeal or so on. Self-sponsored writing, in contrast, originates because you want to write, and you have control. Such writing may be creative or expressive or simply interesting. Its genres may range from family histories to journals, stories to letters, memoirs to op-eds, and include essays like the one you’re reading. No one obliged (or even invited) me to write it, and beyond a transitory bit of attention, there’s no extrinsic reward for my hours spent drafting. Incomprehensible as it may seem to many people, my reward is the pleasure of playing the game of writing.

While teaching people how to succeed in the kinds of obliged writing crucial to economic and personal success, those of us in writing studies are also interested in aspects of self-sponsored writing that can be part of personal and civic life. Obviously, not everyone considers writing a fun way to spend free time, just as not everyone plays Wordle or golf. But it’s clear from the sales of blank journals and from interest in writing groups that many do. For them, technologies that would play the game or do the writing for them would be anathema. Absolutely, writing is a skill that in countless situations has consequences if done well or done poorly. But writing is also a liberal and liberating art whose value lies in personal or intellectual growth, in sustaining relationships, in civic participation. The late Joan Didion’s thoughtful essay “On Keeping a Notebook” meditates on the value of some personal writing not as “patently for public consumption” but as “an indiscriminate and erratic assemblage with meaning only for its maker.”

I think Wordle is fine and fun, though I’m not compelled to play daily. My own strategy is to throw out a random first word—“demon,” “stone,” “gavel,” “light,” “timid”—to see where it leads. I try to win, but within my own odd processes. I appreciate others who devise automated strategies grounded in probabilistic efficiencies, but they’re playing a different game.