What It’s Like to Write Articles with Artificial Intelligence

Image:  The Writing Lesson (1855) by James Collinson/Artvee

I see GPT-3 both as a threat to the conventional notion of writing, but also as a great new tool for authors. Perhaps an analogy is useful here: I’m a calligrapher, and I learn about the Gutenberg press. It makes sense to learn how to work together with GPT-3 as a collaborator, especially when cost isn’t the issue. I’ve written about this before, and have been meaning to make more time and energy to keep at it.

It’s not accessible to everyone yet (check out CopyAI, Snazzy AI, and ShortlyAI), but undeterred, author Vauhini Vara emailed GPT-3’s parent company OpenAI CEO for access to GPT-3, which she got shortly after. (There’s no reason to think you can’t do the same.) She writes at Believer:

I felt acutely that there was something illicit about what I was doing. When I carried my computer to bed, my husband muttered noises of disapproval. We both make our livings as writers, and technological capitalism has been exerting a slow suffocation on our craft. A machine capable of doing what we do, at a fraction of the cost, feels like a threat. Yet I found myself irresistibly attracted to GPT-3—to the way it offered, without judgment, to deliver words to a writer who has found herself at a loss for them.

She’s entirely right on the nuance of this; I would add that GPT-3 didn’t ruin writing, money flowing out of the businesses of publishing and writing did. Disruption caused by the Internet did. I don’t make my living as an author; never have, and with the understanding that if I do, most of the career will involve self promotion

Vara writes:

I had always avoided writing about my sister’s death. At first, in my reticence, I offered GPT-3 only one brief, somewhat rote sentence about it. The AI matched my canned language; clichés abounded. But as I tried to write more honestly, the AI seemed to be doing the same. It made sense, given that GPT-3 generates its own text based on the language it has been fed: Candor, apparently, begat candor.

Vara worked with GPT-3 to write about a story that she couldn’t bring herself to write for years. GPT-3 acts as a word generator, which authors can then edit or just flat-out correct. The writing process then becomes much more conversational. It starts unblocking the writing. The author might think, “No, that’s not right. That’s not what happened,” or, “Wow, that’s pretty spot on,” or somewhere in between. GPT-3 creates the default, and the author edits it. 

Writing Emails to GPT-3

Author Tim Ferriss has often described how he found his voice and tone in writing his first book, The 4 Hour Workweek. At first, he was trying to sound too smart. Then, he was trying to sound too funny. He says at his podcast:

In part, both of those failed because I was writing for a broad audience. I was trying to write for as many people as possible, and I couldn’t do it. I certainly couldn’t do it well. Maybe other people can. So, I sat down and actually opened up a window to compose an email and started, as a first draft at least, writing a chapter to two of my friends, one who was trapped in a company of his own making that he felt like he couldn’t leave, he couldn’t kill his baby, it wouldn’t run without him, etc., which was the exact situation that I had been in.

I remembered once asking GPT-3 to write me a book proposal. I typed in:

I don’t want to outsource the creative process to you though. I want to see how we can work together. The best way you and I can work together is 

And GPT-3 responded:

by exchanging emails and text messages. We’ll decide together what the best method of communication is.

It reminded me of what I heard from Ferriss. So I wrote it an email, and told it I’d write it three stories. The first was a story about the time I travelled to Hawaii. It responded, with the text that when I was young I liked playing with Legos. It guessed accurately—like many people, I did play with Lego. 

I must’ve primed it that way; but still, the experience was fascinating, and I felt a memory and stream of memories of my life start to flow again. The free play and fictional conflicts between people and droids. Assembling bigger, more sophisticated, structures. The many times I went to Wal-Mart and couldn’t even bring myself to ask my parents to buy more Lego. I knew I’d had enough.

Think of GPT-3 as creating a shitty first draft of a shitty first draft. It won’t make sense to the author, but that’s okay, that’s the author’s job (for now). Authors translate GPT-3 for people. Or rather, GPT-3 provides the starting points that authors can make sense of. I wouldn’t be surprised to see it as a core part of creativity-related activities in the future.

Human + Machine > Human vs. Machine

As history—and the present—shows, people have a tendency to fear what they don’t understand. It’s natural. And there is no way most people understand what’s going on with artificial intelligence right now. Still, I’ll close with something I found at Shopify founder and CEO Tobi Lütke’s blog:

After his Big Blue defeat, Kasparov eventually came around to the idea that computers and chess are a good match. He invented a variant of the game where a human player uses a computer as support while competing in a tournament. The machine becomes an input for the person’s decision-making process, changing the narrative from human versus machine to human plus machine. What’s remarkable is that even the best engines will lose to a human working with a machine, because the two enhance each other. They are better together.

No matter the decade, no matter what computers master, the power of human performance will prevail. Witnessing the utmost that humans can achieve will be an eternal privilege. Although we can build machines that are better at nearly everything, our fascination with the strength of the human spirit will never be replaced.

Just a few days before I wrote this, OpenAI launched another product—OpenAI Codex, an “AI system that translates natural language to code.” The flagship partnership is with GitHub, a product called GitHub Copilot. (Like cobots!) At Hacker News, fzaninotto writes: “I’ve been using the alpha for the past 2 weeks, and I’m blown away. Copilot guesses the exact code I want to write about one in ten times, and the rest of the time it suggests something rather good, or completely off. But when it guesses right, it feels like it’s reading my mind.”

I could say something similar for writing with GPT-3. Sometimes it doesn’t guess what’s going on, but when it does, I’m mindblown. It feels almost prescient. It’s an exciting addition to the creative process, and I’m generally excited for its future. Writing can be a team sport. It might sound strange now, but so did many of our parasocial relationships

If you enjoyed this article, you’ll love my Best of Books newsletter, where I send three of the best books to your inbox every month. Thanks for reading.

Leave a Reply

Your email address will not be published. Required fields are marked *