A human wrote this. Probably.

If you want to know where people are paying the most attention to generative AI’s effect on writing, look at where they're searching.

Lafayette, Indiana, had the highest relative search interest for “AI detector” over the past year, according to recent Google Trends data. Gainesville, Florida, came next, followed by Laredo, Texas; Mankato, Minnesota; and Harrisonburg, Virginia.

There’s a noticeable higher-education pattern among these relatively small, lesser-known cities: Lafayette is home to Purdue, Gainesville to the University of Florida, Mankato to Minnesota State University, and Harrisonburg to James Madison University. Charlottesville, Virginia, sixth on the list, is home to the University of Virginia.

Higher ed is certainly being upended by the tech. Darryl Glade, Tulane professor and cofounder of OralExam.AI, built a tool to help deal with this. Rather than trying to “catch” students using AI, it aims to evaluate whether the knowledge actually stuck with them. AI detectors will never know who actually wrote the paper, Glade told me, and they definitely can't determine if the author understands what they “wrote.” 

(Not to mention the high number of false positives, when detectors flag something a student actually did write, that have led to controversy and tension on campuses around the US.)

Beyond the classroom, “understanding” starts to take on a different meaning. If an AI-generated ad still sells the product, or a LinkedIn post still gets engagement, the end goal is met, right? People even seemed to prefer AI-generated storytelling in a recent study by researchers at Villanova — but they still gave higher scores if they were told something was human written. The content didn’t matter as much as the provenance.

So there’s a push to build better detection methods. First-gen tools mostly looked for patterns in the writing itself, which is where much of the recent em dash hate has stemmed from. 

The newer ones are trying to do more, but none are foolproof. Results can change depending on the detector algorithm, the kind of writing being tested and even how much a piece has been edited. More studies are coming out to guide us in the right direction, but even those generating buzz like Pangram may be less accurate than advertised.

Now, the AI companies themselves are working on ways to mark AI-generated text at the source, partially prompted by new rules requiring transparency. Those systems could eventually make it easier to answer what used to be a fairly basic question: Did a person write this? 

None of the detection methods give the nuance on whether the answer of “AI vs human” matters, or will just get met with an eye roll. 

Like the difference between paying for legal expertise and getting slop, versus seeing generated marketing copy and thinking it’s corny.

The regional search data might be able to tell us more about how to plan an approach. 

Big tech companies and top research centers may not be where these norms actually take shape. Instead, it could happen in the places where universities and colleges have a lot of sway over the local economy — as major sources of employment, talent and spending. Plus the indirect economic impact that comes along with it. That’s especially true in “college towns,” where the schools are often anchor institutions.

The places where searches are up is where I'd be watching to see where the technology — and the culture around what work we're willing to accept as genuinely human — goes next.

↩️ Stump speech sidetrack: Meta’s AI-generated summaries on social media posts could be sending you on a major detour. A Massachusetts House candidate looked into his own results and found it pointed readers to a voter guide 1,300 miles south. [Cambridge-Somerville Independent]

🏈 Beat the odds: Prepping your sports-betting strategy ahead of football season? Maybe AI can help. A publication in Mississippi recently wrote a guide to who will win college games this year, based on ChatGPT’s POV. [SuperTalk Mississippi Media]

💊 One man's trash, another man's treasure: It’s rare to see the consequences play out in real time, but the same week Utah launched a pilot that lets an AI chatbot handle prescription refills online, Vermont pharmacy chain Kinney Drugs pulled its AI tool after customer complaints. Two very different signals for AI in the pharmacy — and a reminder that when medication is involved, convenience has its limits. [12News/VTDigger]

POLL: Where do you draw the line on AI use — and when are you willing to let it go? One-click vote, or see results. 

➡️ Click to make Technical.ly a preferred source on Google
➡️ Join the Technical.ly community Slack