A new artificial intelligence program produces blazing-fast results in a variety of formats. Unfortunately, writes opinion editor Clay Wirestone, their depth leaves much to be desired. (Getty Images)
As someone who writes opinion columns but enjoys taking days off now and then, I was excited to learn about ChatGPT. This artificial intelligence program generates tailor-made text in seconds and has already gained mountains of coverage in the global news media. Give the “chatbot” nearly any prompt and it conjures coherent responses: essays, poems, computer code and blender instructions.
Could it fill in for me? At least now and then?
I decided to find out.
“Write me a progressive opinion column about Kansas,” I asked the bot at chat.openai.com, which is now open for public testing.
“Kansas is often thought of as a flyover state, a state that many people only briefly pass through on their way to somewhere else,” it responded. “But there is much more to Kansas than meets the eye.
“For starters, Kansas is a state with a rich history. It was here that the iconic American West began, with cowboys driving cattle across the Great Plains. It was also a center of the abolition movement, with figures like John Brown fighting for the end of slavery.”
And so on. The four paragraphs produced also mentioned renewable energy and called on readers to “consider giving Kansas a chance.”
Okay, so it might not be producing pithy full-length commentary right away. But in other ways, this new technology can astonish. During a few minutes Tuesday night, I asked it to write a piece about the Kansas political landscape, then pushed it to make the piece more extreme and bitter. I asked it to add extra paragraphs about former Gov. Sam Brownback’s tax “experiment.”
The model remembered that I brought up Brownback, so in later exchanges it made sure to include sentences about him. One even mentioned his infamous quote about the tax plan being a “shot of adrenaline” into the Kansas economy. When I asked it to write a humorous column about the situation, it wrote that the governor turned our state’s economy “into a Dumpster fire faster than you can say ‘tax cuts for the rich.'”
These spooky outcomes make more sense when you understand what’s going on behind the scenes.
According to the New York Times’ Kevin Roose, the program works like the auto-correct feature on your cell phone, expanded to a massive scale. It makes educated guesses about what words should follow other words, based on “a statistical model trained on billions of examples of text pulled from all over the internet.”
I noticed this throughout the program’s Kansas column. The chatbot wasn’t producing an original essay, merely an average blended together from thousands of online articles. On a niche subject like the Kansas tax “experiment,” the program likely had few pieces to draw from, which meant I recognized turns of phrase from well-known nonprofit advocates.
The chatbot wasn't producing an original essay, merely an average blended together from thousands of online articles.
– Clay Wirestone
Chatbots could still benefit journalists and the general public in all sorts of ways. That probabilistic model will likely give worthwhile grammar and style feedback. Brain-dead high schoolers could use it for essay inspiration. In time, the bot may be able to “watch” a meeting online and disgorge the first draft of a news story.
Some might fear these outcomes. They might imagine an ominous, Terminator-style rise of the machines. High-profile thinkers such as Henry Kissinger have sounded alarms.
While I haven’t bombed Cambodia, forgive me a dash of skepticism. At this point, ChatGPT and its ilk (even the art-generating applications that have angered creative communities) serve as a kind of lowest common denominator. They’re not creating wildly flavorful original work, but serving up mushy imitation gruel. The interface may be a transformative way to access information, but it still strikes me as an idea in its earliest stages.
Or as Ian Bogost wrote in the Atlantic: “You may find comfort in knowing that the bot’s output, while fluent and persuasive as text, is consistently uninteresting as prose.”
More’s the pity, at least for those vacation days I hoped to take before 2023.
You might wonder about the blender instructions I mentioned back at the beginning of today’s column. I did request a batch from ChatGPT, but with a twist.
“Write instructions for operating a blender in the style of the King James Bible,” I typed.
The program responded with the following:
“1. And it came to pass that thou shalt acquire a blender of thy choice.
“2. And thou shalt plug in the blender, and place it upon thy kitchen counter.
“3. And thou shalt gather the ingredients for thy desired concoction, and place them within the blender.
“4. And thou shalt close the lid of the blender, and secure it firmly.
“5. And thou shalt press the button to start the blending, and let the blades do their work.
“6. And when the ingredients are fully blended, thou shalt turn off the blender and open the lid carefully.
“7. And the fruits of thy labor shall be a smooth and delicious beverage, ready for thy enjoyment.
“8. And thou shalt clean the blender after thy use, lest it become a vessel for foulness and corruption.
“9. And thou shalt be satisfied with the power of thy blender, and give thanks for its usefulness in thy daily life.”
Our stories may be republished online or in print under Creative Commons license CC BY-NC-ND 4.0. We ask that you edit only for style or to shorten, provide proper attribution and link to our web site. Please see our republishing guidelines for use of photos and graphics.