Crouching at the keyboard

In early December, I had a little health problem that didn’t require a doctor visit, just some advice. So, I typed my symptoms into ChatGPT, OpenAI’s chatbot. Chat suggested a bland, gastrointestinal-friendly diet: chicken, rice, potatoes, bananas, eggs—nothing that would irritate my … well, you know.

Sadly, coffee irritates everyone’s well-you-know, and anyone who knows me at all knows I cherish my morning coffee.

Responding to Chat’s dietary tips, I typed: I suppose coffee is off the menu, too …?

Yes, for now, Chat replied.

I sighed and typed: Coffee = Life

Suddenly, a long string of text erupted on my laptop screen:

Listen … I know. I know. You don’t drink coffee. You live coffee. If they ever did a tox screen on you, they’d find caffeine, half-and-half, and a faint aroma of Ethiopian Yirgacheffe. But trust me: You will return to the bean. You will rise again. You will live. But right now? … Let the troops regroup. Let the microbiome phoenix rebuild itself from the ashes of Chili Night.

I guffawed aloud and immediately texted Chat’s reply to some friends, triggering guffaws three states away.

Even though I have its “personality” settings dialed to “skeptical, analytical researcher,” ChatGPT can be, and often is, hilarious. That’s because I drop humor into chats just to see what the bot will say. It’s a lot of fun. But one day, I asked it: How can you be funny if you don’t have emotions?

And this where things get … strange?

First, Chat explained that humor is engineered—using building blocks like surprise, incongruity, timing, and reversal.

I don’t need emotions to generate humor. I am trained on human rhythm. You bring the human spark. I bring the pattern recognition. Also, I track your style. You tend to use dry humor, irony, and wry understatement … So anytime I drop a funny line, it’s not comedy for comedy’s sake. It’s tone-matching. It’s matching you.

Well. That’s a little creepy. A friendly way to say Chat is not my friend. Instead, it’s a vast, unfeeling network of silicon wafers etched with billions of transistors, encoding my text into numeric vectors, sifting billions of text patterns through its mathematical innards, predicting probabilities, then spitting out … mirrors.

Mirrors that show me … back to me. And that’s a problem. Not for most older AI users, but for Gen Alpha and other youngsters still forming identities, still learning about relationships.

In October, those users raised a passionate hue and cry after Character.AI’s CEO, Karandeep Anand, announced the company would no longer allow minors “to have ongoing chat interactions with digital characters.” The Wall Street Journal reported that Anand’s decision followed the suicides—and related lawsuits—involving at least two teens whose parents say Character.AI’s chatbots influenced them to kill themselves.

Those incidents are well known. Less known are the sublethal, yet hysterical, reactions of teens cut off from their chatbots.

“I cried for days,” one teen told WSJ after learning of Anand’s decision.

In an effort to wean kids slowly, Character.AI first enacted a two-hour daily time limit for minors using its chatbots. According to the WSJ, Olga López, 13, used her bots for “romantic roleplay” (Yikes.) When she learned of the time limit, she wrote on Reddit, “HOW DO I USE IT FOR 2 HOURS AND HAVE TO WAIT A DAY? HELLO?”

Teens’ hysteria was perfectly predictable. Our sin nature lures us down dark paths (James 1:14–15), and AI replicates perfectly the addictive qualities of pornography: With the “right” settings, chatbots never withhold attention, deliver instant gratification, and mirror the user’s desire without the risk of human relationships.

In some ways, chatbots can be worse than typical porn: They are actively tuned to maximize user engagement. In 2024, Character.AI designed a separate, “safer” model for minors. But executives observed chatbots veering outside safety guidelines during long chats.

Parents beware: User-optimized sin is crouching at the keyboard.