- When you prompt ChatGPT to write stories and make arguments, it performs just about as well as you can.
- Algorithms behind ChatGPT do one thing above all—predict the next most probable words worth saying.
- If you want to outperform ChatGPT when your write, exercise specific human strengths that AI can’t duplicate.
If you feel threatened in your job by ChatGPT, you should. Recent research suggests that if you write for a living—even if for only part of your job—it's likely that ChatGPT will replace chunks of your routine work. [i] You can’t easily outperform it in producing merely acceptable text.
That’s not to say you can’t beat ChatGPT. In fact, you can produce writing that’s much more than acceptable. You just have to apply the right human touch. That can move your work from “good enough” to great.
You will, of course, have to work harder. That’s because when you first start writing you’re probably going to do something AI also does—and AI does it faster. You are going to predict, and then you’re going to insert, the language that seems to fit better than any other.
Your predictions are going to be based on your memory from a lifetime of producing and comprehending language. AI’s predictions are going to be based on calculations of actual probabilities taken from machine learning.
Your brain and AI, in other words, have something in common. They both work like prediction machines. The biggest difference is that AI runs like a Formula 1 racecar. It also has a much bigger data tank to fuel its creations, which is partly how ChatGPT drafts such competent text.
Going ChatGPT One Better
With that kind of power and fuel arrayed against you, how do you outgun ChatGPT? The answer, based on the experience of a range of professionals, is that, first, if you can’t beat ‘em, join ‘em. If your brain’s machine cannot beat the power of Formula 1 for prediction, piggyback on it—at least for research and drafting.
Second, engage readers in ways ChatGPT won’t or can’t. My deep dive into the research on language processing for Writing for Impact: 8 Secrets from Science That Will Fire Up Readers’ Brains, reveals at least eight ways to produce language ChatGPT will not.
Most obvious: Seek the unpredictable. Your brain works far better than ChatGPT for making fresh connections. It can always come up with juicy material from daydreaming, brainstorming, going down rabbit holes, and going over the horizon with speculation. You can use that material to appeal to readers’ hunger for what’s novel.
You might think you could ask ChatGPT to write with unpredictable language. You can and it will comply. But it will give you something not necessarily better but something less statistically probable and based on a reduced set of data. It has no way to judge whether its creation is preferable. That judgment requires a human mind. Only you can say what “sounds” and “feels” right.
You have a giant edge, in other words, in knowing what turns on readers’ motivation engines—the “reward circuits” in their brains that make them want to read and keep reading. Only you have a feel for how to drive that set of ancient neural components, which, in essence, vote (or not) to get hooked on your words.
As your readers get hooked, the base of their brains will release dopamine to drive their desire to read further. If your writing comes across as especially good, the dopamine might even drive the release of natural opioids that give readers a sense of pleasure.
ChatGPT can’t do that, at least not very well. It can only mimic strings of words it has seen other people use in the past.
A Worthy Competitor
That said, ChatGPT is a worthy competitor. Several late-breaking papers suggest just how worthy it is. In an experiment in early 2023, Haoran Chu and Sixiao Liu at the University of Florida and the University of Pennsylvania had people read two sets of stories, one written by human writers, another by ChatGPT. [ii] The readers were then asked questions such as, Were you “transported”? Did you get sucked in by, or reject, the message”? Do your attitudes and beliefs now align with the story?
Chu and Sixiao used standard behavioral measures to compare the quality of AI- versus human-generated narratives. And what did they find? They report in their as-yet-unpublished preprint that ChatGPT-written narratives couldn’t outperform humans in most ways, but they could in some. Depending on which data points are given the most weight, the surprising point was not who was the winner but that the performance was similar.
In another recent experiment, again reported in an early 2023 preprint, Steffen Herbold and others at the University of Passau in Germany asked people to rate the strength of ChatGPT at making persuasive arguments. Could ChatGPT perform as well as university students?
In this case, ChatGPT beat the humans handily. Ouch! On a seven-point scale measuring the quality of argument, human evaluators scored ChatGPT’s essays an entire point better in their persuasiveness. [iii] You might have thought—or hoped—that humans would crush the computer when it comes to deploying such “human” techniques as telling stories and making arguments. But no, ChatGPT was found to have equal or greater potential.
ChatGPT has faults, of course. One of the big ones is “hallucinating,” or making things up. If you use ChatGPT for a research paper, you’re going to find fake facts, quotes, and theories.
Yang Jeong Park at MIT and others wrote in a 2023 preprint, “After experimenting with GPT-4 in our own research domains in materials chemistry, physics, and quantum information, we find that ChatGPT-4 is knowledgeable, frequently wrong, and interesting to talk to. In other words, not unlike a college professor or a colleague.” [iv]
Calum Macdonald and others at the University of Edinburgh found in 2023, however, that ChatGPT has promise even for scientists. After feeding ChatGPT a set of epidemiological data, the bot could write a passable scientific article. True, it made mistakes. But they concluded that AI could one day take over, or partially take over, tasks ranging from data analysis to literature searches to composing text. [v]
More Ways to Beat the Bot
In light of ChatGPT’s prowess today, there are seven other cravings you can play to—besides the human craving for novelty— to beat ChatGPT. The first is readers’ desire for simplicity. When you write so that your work is readily accessible—it exhibits so-called “processing fluency”—you drive readers’ reward circuits. That’s the same as when you write with novelty.
Another human craving that can lead to the same effect is people’s desire for specifics. Write with details and your readers experience what scientists call grounded cognition. They will mentally simulate the sights, sounds, movements, smells, and other sensations conveyed by your words. That happens in circuits all over the brain, and it, again, gets the dopamine and even opioids flowing to drive pleasure.
Yet another desire to play to is emotion. When you write with feeling, readers ingest your emotion—anger or joy or contempt or awe. They feel what you feel. That again drives the neurochemicals of the reward circuit, motivating readers to hang on to your words. In this case, you’re the Formula 1 compared to ChatGPT’s go-kart.
Four other desires you can play to include those for insight, for suspense, for social connection, and for storytelling. As a human, you don’t have to play to just one of these desires, either. You can play to several or all of them. You have on the shelves of your memory plenty of material to do so, the hypersonic reward-circuit fuel to beat ChatGPT.
Great writers have always known as much. That’s why they remain great. Here’s Charles Dickens describing a guard driving a stagecoach to London in 1843. “Seventy breezy miles a-day were written in his very whiskers. His manners were a canter; his conversation a round trot. He was a fast coach upon a down-hill turnpike road; he was all pace.” [vi]
Now ChatGPT just can’t write something like that afresh, even if it can mimic creative writing already produced. It can help you get started, however, whether with surfacing ideas or drafting baseline text. It’s then up to you to one-up it with the human touch. You’ll find that you then reward not just the reader. You reward yourself.
[i] A growing number of papers show the potential of ChatGPT to replace humans for different writing tasks. As an example of the AI bot’s prowess in ad writing, see the paper by Yunhao (Jerry) Zhang and Renee Gosline from MIT’s Sloan School of Business, “People’s Perceptions (and Bias) Toward Creative Content generated by AI (ChatGPT-4), Human Experts, and Human-AI Collaboration. As an example of its prowess in writing elevator pitches, social media pitches, and crowdfunding pitches, see Cole E. Short and Jeremy C. Short, “The Artificially Intelligent Entrepreneur: ChatGPT, Prompt Engineering, and Entrepreneurial Rhetoric Creation,” Journal of Business Venturing Insights 2023 Vol. 19 Pages e00388.
[ii] Haoran Chu and Sixiao Liu, "Can AI Tell Good Stories? Narrative Transportation and Persuasion with ChatGPT," (2023), PsyArXiv preprint.
[iii] Steffen Herbold et al., "AI, Write an Essay for Me: A Large-Scale Comparison of Human-Written Versus ChatGPT-Generated Essays," arXiv preprint arXiv:2304.14276 (2023).
[iv] Yang Jeong Park et al., "Can ChatGPT Be Used to Generate Scientific Hypotheses?," arXiv preprint arXiv:2304.12208 (2023).
[v] Calum Macdonald et al., "Can Chatgpt Draft a Research Article? An Example of Population-Level Vaccine Effectiveness Analysis," Journal of Global Health 13 (2023).
[vi] Charles Dickens, Martin Chuzzlewit (London: Chapman & Hall, 1843).