AI Is Ultimately Limited Because It Can’t Feel
Many writers worry that AI will take their jobs. But in my experience, it’s more of a tool that helps with the tedious parts of writing—letting me focus on the fun stuff.
3/10/20252 min read
One thing AI can’t do? It doesn’t feel emotion, so it doesn’t know if something is funny or not. In The Glitch, there’s a scene where Aurora interviews Amy to determine if she’s a good client for the lab. Aurora’s questions are a little odd—borderline rude—so Wendy tells her to adjust for “warmer.” I wanted Amy to push back with a good line.
AMY: Wendy, has Aurora retracked onto the spectrum?
AURORA: If you’re keeping score at home, this is a new form of evasion.
AMY: (to wall) “Keeping score at home”. Ha!
AURORA: Light sarcasm matches the current personality temperature perfectly.
AMY: Mmm “Warmer” has veered off into snob.
AURORA: Knowing if you’re lying is key to a successful interview.
AMY: Can I lie a little? Retrack for - real estate agent.
I’ve never been totally happy with the "veered off into snob" line, so I asked ChatGPT to come up with some other options. Here’s an example where it couldn’t actually find the right joke because it couldn’t tell if something was funny. Look at these clunky options that it offered:
👉 “It seems ‘warmer’ has overshot into ‘superior private school kid who summered in Nantucket.’”
👉 “It seems ‘warmer’ took a wrong turn and ended up at a private club on the Upper East Side.”
None of these were exactly funny, but they were close. AI could sense the shape of the joke but couldn’t actually land it. But in its suggestions, I had an instinct—maybe I could do something with the “Upper East Side” reference.
So I continued prompting:
Me: Can we shorten this without losing the upper side reference?
CGPT: Yes! Try this: “‘Warmer’ took a detour to the Upper East Side.” It keeps the humor sharp, trims unnecessary words, and maintains the punch of the location reference.
Me: But I somehow want the idea of the car breaking down, or they're being some sort of failure that they got lost on the upper east side
CGT: Got it! Here are a few options that keep the idea of breaking down or getting lost while keeping it tight and punchy:
1. “‘Warmer’ must’ve broken down somewhere on the Upper East Side.”
2.“Looks like ‘Warmer’ got lost on the Upper East Side.”
3.“‘Warmer’ stalled out on the Upper East Side.”
4. “‘Warmer’ took a wrong turn and never made it out of the Upper East Side.”
5. “‘Warmer’ hit traffic on the Upper East Side and never recovered.”
And then, my gut gave me the line I really liked. And when I ran it by ChatGPT? It actually agreed—and even gave me some instant analysis.
ME: OK this - AMY “Warmer” seems to have blown a tire somewhere on the upper east side.
CGPT: That’s perfect! The “blown a tire” phrase really sells the idea of a breakdown or failure while keeping it sharp and funny. Nice work!
I know a bit about how this technology works, and am convinced that AI can help, but it doesn’t have a “gut”, and can’t replace human instinct that is vital to actual creation.
💡 What do you think? Have you ever seen AI try (and fail) to be funny? Drop your best (or worst) AI-generated jokes in the comments!
🎭 The Glitch is coming soon! Visit theglitchplay.com to register for updates.