As NTCA members cruise into the 30-Day AI Challenge, let me share some war stories about human supervision and brain activity. But, sensitive to slander, I will not identify the chatbot that nearly got me into trouble. Suffice to say it was an object lesson in AI's golden rule: "Human supervision required."
The task seemed deceptively simple. I uploaded a dozen legal filings and asked the bot to identify where specific legal issues were discussed. The bot performed like a champ, or so I thought. In less than two minutes, it returned a topical memo featuring a summary of arguments and citations to the filings in which the parties had advanced their positions.
Except that when I looked at the original filings, no such citations could be found. The bot created them. In AI parlance, it hallucinated.
I wouldn't be the first lawyer to get punked by AI. Lawyers in Utah, Maryland, Massachusetts and elsewhere have suffered sanctions for citing non-existent cases, including one who (sigh) used "unvetted AI" to defended his use of . . . unvetted AI.
Did my experience dissuade me? No. It simply confirmed the practice of cite-checking. I refined the prompts. I double-checked everything. And ultimately, AI, with my human supervision, carved about 50% off the anticipated time for the task.
And before I promote the 30-Day AI Challenge, consider this: Researchers at MIT conducted a study that pitted AI users against "brain only" and other groups. They found (how do we say this politely) "significantly different neural connectivity patterns" among AI users in an essay writing exercise. More specifically, "[b]rain connectivity systematically scaled down with the amount of external support." More bluntly, the researchers identified "cognitive cost."
But digging deeper, the MIT research supports the proposition of using AI creatively to support and enhance human endeavors, rather than replace them. Which brings me to the 30-Day AI Challenge (and the Bulwer-Lytton Fiction Contest).
From 1983-2025, the Bulwer-Lytton contest awarded prizes for the worst opening sentences to terrible novels. Creator Scott Rice (an English professor at San Jose State University) called it "getting to start a fight and not having to finish it." The winners were spectacularly awful, but here's the thing: Only someone who truly understands the English language could intentionally make it misfire badly (or well) enough to win the contest.
That's an analogy I've been using for AI. It is a great tool. I have fed AI engines 32-field survey reports from more than 150 respondents including multi-part open-ended response segments and the systems crunched the results and provided analyses based on any parameters I requested - breaking results down by age; ranking preferences from open-ended answers; providing sentiment analysis across demographics. But the knowledge base is only as good as the data in, and the outcomes are only as good as the prompts.
Enter NTCA's 30-Day AI Challenge: four categories, two divisions (Newbies and Denizens). It's about test-driving tools in a competitive but safe space, having fun and maybe winning robot trophies. As a popular gym advertises, it's a judgment-free zone.
As starter idea, design a marketing survey testing price sensitivity and bundling preferences across generations - Gen Z, Millennials, Gen X, Boomers. What bundling do they prefer? Who's streaming? Who's working from home? Multiple choice or rating scales? Can AI help avoid tedious open-ended questions while still capturing useful data? How can the data be analyzed to recommend actionable plans?
A skilled survey designer could handle all this manually. But I'd wager that AI can help create a better survey faster and capture more valuable insights.
And circling back to where I started: the higher your skills, the better your AI results. Just like the Bulwer-Lytton winners.
So, take the Challenge. Test it. Break it (that's another tip – despite the fact that AI is emotionless, it likes to please – challenge it). Learn from it. Just remember to check the citations.