Can AI actually help us become better writers?

I’ve been thinking about this question since reading “AI Didn’t Make Programming Easier. It Just Made It Differently Difficult,” an article recently published in Communications of the ACM (CACM) by Jeremy Osborn.

Osborn explained that programming has traditionally been a demanding cognitive activity in which developers rely heavily on working memory and complex mental models. But humans have a limited amount of working memory — the kind of memory we use for keeping track of the concepts and knowledge needed for reasoning and solving problems. Our limited working memory constrains our ability to deal with complex problems when their cognitive demands become too great.

AI-based coding assistants are changing that relationship. By serving as external memory and analytical resources, they can relieve programmers of some of the cognitive burden associated with remembering syntax, retrieving information, and generating routine code.

This doesn’t eliminate the need for programmers. Instead, it changes what it means to be a programmer — from primarily a coder toward an overall orchestrating agent who can devote more attention to reasoning, architecture, judgment, and the workings of the overall system.

Osborn’s CACM article brought to mind an interesting question. What other demanding cognitive activity could AI significantly complement?

Based on my own experiences, the answer quickly came to mind: writing.

My Experience with an AI Editor

I’ve been writing a weekly blog for over 20 years. Writing each post takes me many hours, usually spread over several days. First, I spend considerable time thinking through the topic I want to explore and the articles and other materials that I’ve read and want to reference. Then comes the drafting, revising, and editing.

For me, writing isn’t simply a way of communicating ideas I’ve already formed. It’s an important part of the process through which I develop and clarify those ideas in the first place. Writing has become my way of exploring new subjects and deepening my understanding of complex emerging technologies like AI. In addition, as I’ve gotten older, writing is, hopefully, a good way of exercising my aging brain.

For a number of years in the 2010s, edited versions of my blogs were also reposted in the Wall Street Journal CIO Journal. Given the writing skills of the Journal’s editors, I often liked their edited versions more than my original drafts. I miss having editorial help to improve the clarity and flow of my writing.

Given the continuing advances in large language models, I wondered whether I could use a chatbot as a kind of AI editor.

So, a few months ago, I logged on to ChatGPT and wrote that I wanted help lightly editing the draft of a blog while preserving my voice and style. I then pasted the draft into the chat. Within seconds, ChatGPT returned an edited version, explaining that it had focused on improving clarity, consistency, and grammar while preserving my analytical arc and writing style.

I accepted many of its suggestions, rejected others, and remained fully responsible for every final decision.

I was genuinely surprised by the quality of the AI edits. The chatbot wasn’t writing my blog or generating ideas on my behalf. It was doing what a good editor does — helping me express my own ideas more clearly. I’ve since been using ChatGPT as my AI editor.

But over time I’ve noticed another potential benefit.

When an editor repeatedly points out an overly long sentence, unnecessary repetition, an abrupt transition, or a place where an argument isn’t sufficiently clear, you begin to recognize those problems yourself. In that sense, the interaction becomes a form of feedback and learning.

A good editor doesn’t merely improve the manuscript in front of you. Over time, a good editor can help improve the writer.

This is where I see an interesting parallel with AI-assisted programming. In both cases, AI can take over some lower-level cognitive work while helping the human concentrate on higher-level questions. For programmers, those questions increasingly involve architecture, judgment, and systems reasoning. For writers, they involve deciding what is worth saying, constructing an argument, evaluating evidence, and ultimately deciding whether the words on the page actually express what we mean.

The Role of AI in Writing

While the use of AI in software development is now fairly well accepted, there’s still considerable debate about its role in writing.

The debate is particularly contentious when it comes to fiction. Author Cully Perlman explained why he has very different attitudes toward AI in fiction and nonfiction in “The Role of AI in Nonfiction Versus Fiction Writing.”

“For me, fiction is untouchable. I will NEVER use AI for fiction,” wrote Perlman. But, he feels differently about nonfiction, where AI can assist with research, organization, clarity, and editing.

The distinction is useful because different forms of writing will inevitably draw the boundary around acceptable AI assistance differently. Nonfiction is generally judged heavily on factual accuracy, reasoning, clarity, and the quality of its analysis. Fiction depends much more heavily on originality, voice, imagination, and emotional resonance.

But in both cases, the important question isn’t simply whether AI was used. It’s what role AI played and what intellectual and creative responsibilities remained with the human author.

AI as a New Medium

Tim O’Reilly — author, publisher, and founder of O’Reilly Media — offers a particularly interesting perspective in his recent article, “Writing with AI: A Contrarian View.”

“AI is a medium, like painting, sculpture, photography, music, or language itself,” wrote O’Reilly. “If AI is a medium, it is a field of possibility from which humans can and will summon up and communicate insights, truth, beauty.”

Calling AI a tool suggests that we already understand what it is for. Calling it a medium suggests something quite different: that we’re still discovering what humans will eventually learn to create with it.

“If AI is a medium, the idea that you can’t write with it will someday seem as odd as the idea that you can’t make a good portrait or landscape with a camera, only with paint or pen and ink,” wrote O’Reilly.

“Technical innovations have often changed the nature of art … But we don’t think that Leonardo was cheating when he used a camera obscura to perfect perspective in painting, or that John Lasseter was getting away with something because he created Toy Story with a computer instead of painstakingly animating it by hand.”

His analogy with photography is especially interesting. A painter may decide on thousands of individual brush strokes. A photographer makes a different set of decisions — the subject, angle, lighting, exposure, speed, focus, and many others. Both can create art, but they exercise human creativity and judgment in different ways.

“Figuring out a new medium means expanding the range of choices that are possible with it,” wrote O’Reilly.

He believes we’re at a similarly early stage with AI and writing.

New media often begin by imitating the media they replace or complement. Early filmmakers essentially pointed cameras at stages and recorded theatrical performances. Only gradually did filmmakers discover techniques unique to their new medium — close-ups, cuts, camera movement, montage, special effects, and eventually computer-generated imagery.

“Most of what gets produced today is a camera pointed at a stage,” O’Reilly wrote about AI-generated content. “The question isn’t whether slop exists. It obviously does.”

“But what we’re working out now is what the equivalent to the close-up and the cut and the CGI will turn out to be. Once we do that, there will be people who are masters at summoning words from an LLM, just as there are masters with a camera and the stories you can tell with it.”

I find this a particularly intriguing way of looking at AI. We’re currently judging AI-assisted writing largely by comparing it with writing created without AI. But if O’Reilly is right that AI represents a genuinely new medium, the more interesting possibilities may emerge when writers discover ways of working with AI that don’t simply reproduce existing writing processes more efficiently.

A Cognitive Complement — or a Cognitive Crutch?

Of course, there’s another possible outcome. AI could make us worse writers rather than better ones.

Someone who routinely asks an AI system to decide what to say, construct the argument, and produce the prose may exercise their own writing and reasoning abilities less, not more.

This is particularly important because writing and thinking are so closely connected. Writing forces us to organize our thoughts, confront weaknesses in our reasoning, determine whether the evidence supports our conclusions, and find the words that express what we actually mean.

If we outsource all that work to AI, we risk outsourcing an important part of the thinking process itself.

AI can therefore become either a cognitive complement or a cognitive crutch. The difference depends on how we use it.

My own objective isn’t to outsource the difficult cognitive work of writing. I still decide what I want to write about, read the articles and papers I want to discuss, decide which ideas are important, determine how the pieces fit together, and write the initial draft.

Then I ask my AI editor to help me see what I may have missed — where the argument could be clearer, where I’ve repeated myself, where the organization could be improved, or where a sentence simply doesn’t say what I intended it to say.

And, just as I did with the human editors I worked with at the Wall Street Journal, I decide which suggestions to accept and which to reject.

Becoming Better Writers

“I suspect that for most of us, having AI as a research assistant and a copy editor will make us a better writer,” wrote O’Reilly.

“We’re in the early days, still shooting filmed stage plays. The masters of this instrument will uncover themselves. When they do, what they produce isn’t going to be slop. In the meantime, I don’t think we should be shooting down people who are trying to push AI’s limits in writing or art. Writers should be trying to figure out the medium, just as software developers are doing in coding.”

I think that O’Reilly is quite right.

AI can certainly be used to avoid the hard work of writing. If we ask a chatbot to decide what we think, construct our arguments, and produce finished prose on our behalf, we shouldn’t be surprised if our own writing and reasoning skills deteriorate.

But that’s not the only way to use AI.

A good editor doesn’t simply make an article better. By identifying unclear arguments, awkward sentences, unnecessary repetition, and weak transitions, an editor can help the writer recognize those problems and eventually learn to avoid them. That’s one of the reasons I valued working with the skilled human editors at the WSJ CIO Journal and it’s increasingly what I’m experiencing with my AI editor.

This brings me back to Osborn’s argument about programming. AI doesn’t necessarily make difficult cognitive activities easy. Instead, it can change where we devote our cognitive effort.

For programmers, AI can reduce the burden of remembering syntax and generating routine code, allowing them to devote more attention to architecture, systems reasoning, and judgment.

For writers, AI can help with research, editing, and alternative formulations while leaving the most important intellectual responsibilities with the human: reading articles, blogs, and reports; deciding what is worth writing about, determining what we believe, constructing the argument, judging whether the evidence supports it, and deciding what we ultimately want to say.

The most interesting question, then, isn’t whether AI can write better. Increasingly, it can write remarkably well.

The more important question is whether we can learn to use AI in ways that help us think, communicate, and ultimately write better ourselves.

Based on my experience so far, I’m increasingly optimistic that the answer is yes.