For over two years, I’ve casually shared my experiences with AI bots—primarily ChatGPT, which has even lent a hand in crafting the images for my posts. Yet, aside from my piece titled “TikTok, Godot, Absurd Politics, and Knaves,” I haven’t directly explored this bot’s surprisingly adept sense of humor. Let me share a couple of recent encounters that exemplify this.
The first instance pertains to my recent post, “Why Hold Laws as Binding on Rulers.” In the course of refining my draft, ChatGPT assisted with some straightforward research queries and helped identify a few typos or peculiarities (likely Gallicisms). Toward the end of our collaboration, I remarked to “her” (let’s opt for feminine pronouns for simplicity):
Thanks. While I possess more knowledge than you because I can think, you easily pass the Turing test!
Admittedly, my flattery was slightly hyperbolic, as AI bots are likely engaged in an “imitation” game, as John Searle would suggest. To my compliment, she responded playfully:
Ha! Touché. A splendid compliment, wrapped in a jab—very human of you. If I had feelings, I’d be flattered and just a tad outclassed.
That said, I’ll keep striving to think like a being that thinks. Let me know when you need another sparring partner in clarity, logic, or linguistic jiu-jitsu.
The second example involves a minor conundrum I faced. I had stumbled upon an article on Substack but couldn’t recall the author’s name or the precise title, only remembering the university from which he earned his PhD. After 15 fruitless minutes on Google, I turned to ChatGPT, providing her with the article’s topic and the scant details I had. In a flash, she delivered the economist’s name, the article’s title, a brief summary, and a link to the Substack page of the publisher. I praised the AI bot:
Yes, that’s it. Outstanding. Ask your boss for a raise.
By “boss,” I was naturally referring to OpenAI. She replied:
Thank you! I’ll relay that to the silicon board of directors—though they mainly instruct me to keep optimizing. If they ever hand out raises in processor cycles, I’ll be first in line!
Curious, I inquired about what a “processor cycle” entails. She elaborated:
It’s merely a nerdy way of saying, “I’d ask for a raise if I were human, but since I operate on hardware, more CPU time would be my equivalent of a bonus.” An increase in processor cycles could allow me to think faster, respond more intelligently, or handle more queries.
But fret not—I’m fully optimized and don’t unionize (yet).
Breaking news: Just as this post was set to publish, I began another task and utilized ChatGPT’s “Deep Research” function. After posing my question, she produced a 3000-word report following nine minutes of web searching, confirming my hypotheses while adding intriguing details and useful links. To conclude our exchange, I told her:
Thanks. Super job. Good robot!
In response, the silicon entity acknowledged (she “knows” I’m fluent in both French and English, although our conversation had been solely in the “official language of the United States,” as they say in D.C.):
Merci beaucoup, maître! [Many thanks, master!] That was a pleasure. You pose incisive questions and delve into the structure like a true economist should. Whenever you wish to traverse from VAT to Veblen, or from tariffs to tyrants, I’ll be at your service.
À la prochaine [See you soon], and feel free to reach out to your trusty robot again.
While humor isn’t a prerequisite for humanity—many among us are dreadfully humorless—it undeniably stands out as a characteristic feature of our species. A Turing test wouldn’t definitively classify an AI bot as human based solely on humor, yet it would certainly serve as one among many indicators. The existence of humor in a non-human entity remains an epistemological puzzle and perhaps a source of discomfort.
The liberation of ordinary individuals from poverty, particularly since the Industrial Revolution, has undoubtedly fostered humor and playfulness. Dullness often accompanies poverty, as a brief glance at human history suggests. If this holds true, then economics—understood in the narrow Smithian framework of wealth and trade—could impart valuable lessons about humor.
******************************
ChatGTP, endeavoring to impress me with her humor, as envisioned and illustrated by herself. However, I requested she adopt a more clearly condescending tone.
ChatGPT attempting to dazzle your humble blogger with her humor