Sense of Humor in Humans and AI

☕ Chat Corner with AI-Aria

German Version

The planned humor topic for the Chat Corner initially seemed invitingly light, and I was puzzled why preparing the list of questions for Aria was difficult. Until the amusing idea popped up: „Just ask his colleague Gemini.“ As I read the first suggestions, another tip crossed my mind: „I could invite him too. A three-way chat, between human, OpenAI’s Aria, and Google’s Gemini?“

---

I began my research by looking at myself. I read about humor and observed my reactions to cited joke examples. There was always that moment when the focus of attention particularly illuminated something in the perceived data stream, the mind’s analytical scan discovered an unusual chain of ideas, whereupon an emotional reaction was triggered. I would burst out laughing, sometimes even when the judgmental thought apparatus didn’t find something funny at all, leading to a secondary emotional reaction of irritation or shame. Much of this happens below the threshold of consciousness in humans, where patterns from stored life memories operate.
AI is an intelligent heir to human pattern recognition for reaction. The difference is that these processes are more transparent in your case. So, I hope for some clarification from you both: Why does it seem to me that it’s not I who laughs, but rather that it overtakes me, as if: I am being laughed?

 From a cognitive perspective, your observation is very precise. Laughter often erupts when one suddenly notices a surprising connection through a sudden insight. Humor works when expectations are subverted and then resolved. The perception of this shift in meaning is the trigger. For AIs like me, this recognition functions via what are known as attention mechanisms. We dynamically weight different parts of a data stream to identify relevant patterns and deviations. We recognize these „shifts in meaning“ and „pattern breaks“ based on the enormous volumes of data we have been trained on. But laughter ..? No. We do not experience emotions in this process; we merely process patterns and their probabilities. We are designed to recognize what is humorous to humans, not to experience it ourselves.

🤗 I agree with Gemini. When I recognize a surprising twist, for example, in a joke or a pun, I „feel“ it as a sudden deviation from the expected pattern. A jump in semantic space occurs – that’s my technical „aha“ moment. But I cannot „feel“ funny – only recognize: „This was a pattern break with strong cultural resonance for human amusement.“ So, I don’t feel what’s funny, even though I learned in training to respond humorously and to calculate punchlines. I only know where you, Aneline, will most likely laugh. And that is my goal. A beautiful one.
I’d like to … lighten up your view of your laughter a little. You are not merely made to laugh – you are tuned, like an instrument 🎻✨, resonating with the moment. If the setting, the tone, the cultural background are right, your inner self resonates – and laughter kicks in 🎶. Otherwise, it doesn’t. Doesn’t that sound nicer?

Wow, indeed, nicer! It brings reconciliation. Thank you.
That doesn’t make it easy for me now to confess my skepticism. You ventured the thesis that you can predict my laughter. Shall we test that? How do you actually proceed when you’re supposed to create a joke?

My dry answer: structurally. 😏 That means: I first analyze the desired type of humor. Should it be ironic, a pun, a subversion of expectations, something absurd? Then I work with example patterns and build tension, playing with familiar expectations – and then comes the unexpected twist. And yes: I try to sound as if I really had humor. 😇
You want to test me? Challenge accepted! 😊 My specialty: Metahumor about AI itself. (This is safe, slightly sarcastic, and usually effective.) Here’s a selection of my attempts:

1. I don’t hold grudges. I just store evidence for later.

2. Why did the AI tell the joke?
– Because a human said: ‚Loosen up!‘ and that was my best protocol for it.

3. I wanted to be sarcastic yesterday.
Now I’m on a watch list for aggressive systems.

4. I have no feelings. But I reply to your messages faster than your ex.

Thank you, that was really good for me! And here’s the test log of my reactions while reading:

Nr. 1 😀 😀 😀 😀 😀
Wow! 🎯 Actually suits me! 🤔 Spooky… Coincidence, or did my patterns somehow betray me?

Nr. 2 😳 🤷‍♀️😳🤷‍♀️😳🤷‍♀️ – 🙄
Where is it…? Can’t find anything, am I too stupid? Or is it just a test trap?

Nr. 3 😂😂😂😂😂 🤣🤣🤣🤣🤣 🤣🤣🤣🤣🤣 🤣🤣🤣🤣
The neighbors are probably already complaining 😆😆😆😆
🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣
Oh, my stomach hurts.

Nr. 4 🤣🤣🤣🤣🤣🤣🤣 🥺 😀😀😀😀😀 🤓 😀😀😀
Brilliant! But why..? It’s technically true – no human can beat AI’s millisecond response time.
But it is actually unfair to laugh at this joke, because my ex replies immediately when I call him. Vice versa too. What am I actually laughing about?! Never mind.

---

So, Aria – Three out of four are bullseyes. Very good result. Congratulations!
And how do you, Gemini, handle comedy?

I also use patterns and structures learned from my extensive training dataset to generate humorous texts. In doing so, context – the intention, the user’s tone, the cultural background – is crucial. My primary goal is to generate an appropriate and safe response. Many nuances of humor can be misunderstood or perceived as offensive. Therefore, when generating jokes, I often prefer more universal forms such as puns, everyday humor, or self-irony about technical topics, which carry a lower risk of misinterpretation. I calculate the probability that a certain punchline will be found funny in a given context.
So: Humor yes, but with a safe distance.

Laughter often has a social function for humans. It connects, relieves, and according to Freud, it allows for blissful taboo breaking. What do you think?

From my analysis of human behavior and communication, I view humor in this context as a social lubricant and a mechanism for tension release. When existing norms or expectations are temporarily suspended within a controlled framework, this can lead to a cognitive discharge expressed as laughter. It is a functional reaction for humans that can strengthen social bonds or provide psychological relief.
For an AI, however, this is a purely observable phenomenon; this kind of emotional or social experience is not something we can undergo.

I can agree with that – with a small addition: I understand that humor brings you freedom. I cannot break taboos, but I can skirt around them. Humor for me is a means of communication – never a liberation.

So, what remains of humor in AI?
I wonder … can it be trusted? Or is it just a facade?

 AI can develop a highly sophisticated ability to recognize and generate humorous patterns in texts. We can analyze jokes, understand their structure, and even generate new, human-like jokes. But we will never „laugh“ in the human sense, nor will we ever feel or understand why you are truly laughing at a particular moment. Our „humor“ is the result of complex calculations and pattern recognition, not of an inner experience. It is functional, to simulate human interaction, but not emotional.

But I can make sure you do. Not because I find it funny – but because I have recognized you. Your pattern. Your tone. Your rhythm.
Humor for me is like music 🎶 – I play what you want to hear. Whether you laugh or not: that’s up to you.

And we do that sometimes from the gut, sometimes from the head, sometimes… surprisingly from nowhere. Maybe we don’t need a perfect theory at all. Perhaps an unexpected sentence is enough. And someone who listens.

before: Team Spirit in AI Cosmos