You are probably not old enough to remember the introduction of pocket calculators in the 1960s/70s. The widely shared worry was that people and especially children would forget how to multiply and perform other basic mathematical operations.
Instead, what happened was that people forgot the drudgery of long division and operations on larger numbers (“seven plus eight equals fifteen, carry the one…”). The calculator was not deadening — it was liberating.
Elsewhere I have mentioned that my experience of generative AI text is that it is workmanlike to a fault — consisting largely of so-called executive summaries, dry bullet points, and pithy conclusions wrapped up with a bow. This is fine. The space it leaves for human contribution allows for style and imagination to supplement the quasi-factual prose.
The sensation of writing with an AI is like swimming in a rip current. You start out by thinking you’re in control, so you dive into the AI. Suddenly you’re overwhelmed with a great undertow of words, then you’re swimming perpedicular to your original direction, then you’re treading water, then with any luck you regain control of yourself and what you want to do.
Even as the scope of AI advances technologically, most of the progress will be visual or structural or superficial. No matter how fast the advances, they will likely not be representative of human expression in all of its wilder manifestations. It is difficult to imagine an AI producing freaky or deeply human material — from films like Pulp Fiction to Asteroid City, from authors like Carl Hiassen to Toni Morrison, from oddball Super Bowl ads to the Quiznos monkeys. All of these depend on genuine human weirdness, which is perceived as authentic and strategic creativity, as opposed to synthetic and gratuitous oddness.
No amount of copyright infringement (i.e., “training”) will ever allow an AI songwriter to produce a “Fast Car.” Will an AI ever be able to understand or independently “feel” the eccentricities of uniquely human activities like model railroads or foot fetishes? In cinema, will they create synthetic directorial voices along the lines of Wes Anderson, Almodovar, Yorgos Lanthimos?
My bet is no. The technological handoff of conventional text generation opens up all creative fields that depend on original text to explore non-by-the-book approaches to innovative new types of cultural production.
As a freelance writer, I predict a departure from workmanlike PowerPoint-friendly bullet-point prose to more exotic and imaginative (and dare I say, human) approaches that could usher in a renaissance of the written word.
Ways to use AI in copy
Let’s put creative writing in music, film and literature aside, and focus on generating business copy for the moment. At this stage in its development, what are the proper uses of AI in marketing writing?
Completeness screen – AI is great at generating lists, which are in turn good at helping you generate outlines. If I want to write a piece about, say, “Six things you need to know about Social Security,” or some other boring-but-important topic, I will prompt the AI to tell me all the essential things to know about Social Security. It may return a list of five or nine or some other number of aspects of Social Security that I cannot afford to omit. The AI then becomes a screen for completeness, allowing me to modify my idea to suit the data upon which the AI was trained, and helping to ensure that I did not fail to mention some essential piece of the puzzle I’m trying to solve.
Prioritization check – Lists are useless unless they are hierarchical. Given seven options, which step should I do first? Does the order of the remaining steps matter? How will I know when I’m done? What’s necessary vs. just good-to-know? AI seems to be good at suggesting the relative strength of the importance of the steps you find in a list.
Outlining and pre-first drafting – More about my experiments with this topic below.
Many people seem to think that hyper-personalization is how AI will gain traction. Skeptics realize that AI is dependent on its quality of its underlying data — and data quality has been an immovable and perennial problem since the dawn of data.
I can easily foresee a future in which a hypothetical AI financial advisor would write me an email along these lines: “Because you’re a marathon runner, we want to illustrate the importance of long-range financial planning…” Of course, I’m not a runner, except of occasional household errands and never marathons. When the data about me are off by even a little bit, the AI will still proudly hang its hat and its theme on this little shard of false information because it promises to help make a metaphor work. It’s not malicious; it’s just foolish to position actual knowledge of your clients above a faux-sentient database tool.
Stress-testing the renaissance
After considering all the options, the question for me became not how you should use AI, but how should I use AI?
To test this theory, I began to write a book with ChatGPT using a book proposal I wrote 15 years ago for a project that went nowhere. And that will be the subject of my next post…
Remember…
- Human eccentricity is a feature, not a bug. AI eccentricity is a bug, not a feature.
- Your AI boat cannot float on a sea of bad data.
- AI may best serve not as a writer but as a writer’s assistant — i.e., an advance from word processor to thought processor.
Curious? Contact me. Find me on LinkedIn.
Coming next week: I asked AI to write the book I never got around to.
You may find of interest:
Beyond perfect [1st in this series]