This week I received perhaps the oddest letter to the editor in my two decades at the Sedona Red Rock News: A letter generated by artificial intelligence, or AI.

It began innocuously enough with a short introduction by the author that read: “The following is my opinion letter to Red Rocks News regarding tourism concerns in Sedona.”

The letter lacked the required home address and telephone number, and was over the word limit, but before I sent an email to the author, I noticed the first sentence did not seem, well, authentically human: “Sedona, with its striking red rock formations and spiritual allure, has long been a magnet for tourists seeking solace and serenity,” which read less like an opinion letter and more like a brochure inviting someone to visit.

The next sentences had a few phrases that looked more like SEO keywords than original thoughts: “Yet, beneath the veneer of tranquility lies a growing concern: The unsustainable impact of tourism. As the influx of visitors and short-term rentals reaches unprecedented levels, it’s crucial to confront the challenges that threaten the natural beauty and quality of life for residents of Sedona.”

The letter continued with vague suggestions about how to address these unspecific problems. It purportedly aimed to address the problems of overtourism, yet was noticeably lacking in any specifics: No specific problems, but also no names, no government agencies, no elected officials, no government staffers, no specific policies to fix said problems.

There were hints of data, but no actual numbers. References to “exponential tourism growth” and “rapid expansion of tourism” might have been accurate in late 2020 and 2021 when scores of Arizonans from the south and Californians turned to Sedona and the Verde Valley to vacation, but such numbers peaked by 2022 and we’re on the downslope — spring break traffic excluded. For an AI program, however, scouring the internet for relevant news about Sedona to produce AI-generated text under the limited parameters of “tourism problems in Sedona,” old data is as relevant as current conditions. AI doesn’t know the difference.

The letter also mentioned our “iconic landmarks” without naming any. We all have friends and family who have visited us for a day or a week, and even they can remember the name of one landmark, be it Cathedral Rock, a vortex, Bell Rock, Red Rock Crossing or manmade landmarks like Chapel of the Holy Cross, Tlaquepaque or the Amitabha Stupa. Tourists generally hike or take a tour somewhere and remember the names of the things they saw.

Buzzwords like “stakeholders” and “policymakers” are placeholder jargon lacking specificity, indicative of university-speak or AI text. If a project is near your house, you’re a neighbor; if an official makes a decision that affects you, you’re a voter; if it’s something with a public school, you’re a parent or a taxpayer. No one self-identifies as a “stakeholder” outside of a college setting. Which is the disappointing part — the “author” who emailed me is a real person, a current undergrad at my alma mater. Sigh.

For those of us in journalism who spend most of our day reading or editing other writers’ stories, columns, letters, editorials and press releases, every writer’s style is like a thumbprint. It’s not a special power or a skill, it’s just knowing how real people structure their thoughts in print, the words they chose, the complexity or simplicity of their sentences and the use of punctuation to convey how something should be read. You can do it, too, even if it’s unconsciously.

I had another staffer read the piece, who concurred that it seemed inauthentic and unhuman.

We then ran it through two online AI-generated text detectors, both of which determined that the text was 100% generated by AI.

I do wonder if this was a college study to see if AI text could get printed in a newspaper or if it was just a college kid who wanted to say something about Sedona tourism, but didn’t want to put any effort into doing the work. Either way, this AI failed our Turing test.*

The author hasn’t responded to my email asking for the required contact information, nor addressed the details we noticed indicating this “letter” was not written by the hand of man.

As AI-generated products begins to proliferate in our lives, be those text, video or other media, it behooves us to be aware of what’s legitimate and what’s not. Certain AI programs will greatly benefit our lives, predicting the kind of music we want to hear, what wallpaper will compliment our furniture or what we want to watch on YouTube.

But when it comes to the written word, AI currently has a bad track record, and readers notice. In 2023, a client partner of Sports Illustrated posted AI-written stories and AI-generated images on its website, and dedicated sports readers quickly noticed it was artificial. Be wary because AI lacks the human touch. It won’t suggest new music in styles we haven’t heard and some of that information online may originate from bad, biased or mistaken sources that AI has no interest in correcting.

Trust your instincts as AI develops. You’re not stakeholders in the future; you’re readers, users, consumers and, most importantly, humans.

Christopher Fox Graham

Managing Editor

The AI-generated text letter:

* Turing test, in artificial intelligence, is a test proposed in 1950 by the English mathematician Alan M. Turing [June 23, 1912— June 7, 1954] to determine whether a computer can “think.” There are extreme difficulties in devising any objective criterion for distinguishing “original” thought from sufficiently sophisticated “parroting”; indeed, any evidence for original thought can be denied on the grounds that it ultimately was programmed into the computer. Turing sidestepped the debate about exactly how to define thinking by means of a very practical, albeit subjective, test: if a computer acts, reacts, and interacts like a sentient being, then call it sentient. To avoid prejudicial rejection of evidence of machine intelligence, Turing suggested the “imitation game,” now known as the Turing test: a remote human interrogator, within a fixed time frame, must distinguish between a computer and a human subject based on their replies to various questions posed by the interrogator. By means of a series of such tests, a computer’s success at “thinking” can be measured by its probability of being misidentified as the human subject. In late 2022, the advent of ChatGPT reignited conversation about the likelihood that the components of the Turing test had been met.