Can a Chatbot Help You Write Humor?

Some thoughts on writing humor in the era of AI

Photo by Steve Johnson on Unsplash

Hello, and welcome to my new subscribers! I think a lot of you found me via this interview in the excellent substack Evil Witches, and I’m happy to have you here! Today I’ve got a bonus piece of content about humor writing and my next newsletter will be my monthly roundup of funny stuff.

As you have probably heard, new AI technologies like ChatGPT have been much-discussed lately in the writing world. It’s something I’ve been reading and thinking about both for my own writing and for my college teaching job. I’m currently teaching a new special topics course focused on writing for the web and at the rate that things are changing, I suspect I will need to make a lot of changes to the content of the course the next time I teach it. Or maybe AI will just do it for me? I guess we will see.

Testing out jokes in ChatGPT

I decided to play around with ChatGPT to see what kind of humor writing it can put out. First, I just started with a basic ask for a parenting joke:

I would call this a dad joke and although popular, many of these pun-based jokes have been done before. So I tried asking for a mom joke:

Okay, so this is just a dad joke with the word “mom” in it, but I guess at least it didn’t throw in a wine mom cliche? Unless the drinks on the roof were wine and then I guess it did not avoid that.

I decided to try to adjust a bit to ask for dry humor since that is more my style:

This one is leaving dad joke territory but is relying on a very well-known joke structure (chicken crossing road) and something that feels like well-worn parenting advice territory (looking both ways).

But there are plenty of Tweets and memes that rely on tried and true joke structures that come from human brains too. Sometimes these are knowingly being mimicked (and sometimes outright copied) but other times a joke occurs to a person as their own idea but then others have already done it. As was discussed in the Conan plagiarism case, in the era of Twitter, a lot of people are going to arrive at the same joke at the same time.

I could see you getting to a point where AI helps you tell okay (although probably not wildly original) one-liner jokes for things like social media, but I think the best jokes are fresher.

I’ve also wondered how well AI could write a satire piece (and I’m not the first person to wonder this).

I tried having it write something in the style of this piece I’d already published on McSweeney’s called Holiday Air Travel Haiku:

Yeah, this one is not really working for me. Definitely doesn’t feel like my voice and the last line feels kind of dated and ugh-worthy itself.

I tried another search adding “McSweeney’s” to it to see how the style would be:

ChatGPT really likes the word “Yule” I guess? The voice and jokes aren’t working for me here either.

While it seems tough to have it write a wholesale satirical piece I could still see writers using it in sort of hybrid ways (and some probably already are).

What other humor writers have tried with generative AI

I’ve read a few other pieces about people using AI to write humor with varying results.

In this recent Vulture piece, the writer had ChatGPT write jokes in the style of six different comedians and then asked each comedian what they thought of those jokes. There were maybe some usable lines in some of the responses but some of the setups and language felt tired and the jokes didn’t really mimic the comedians’ style.

From what I’ve read, this seems true of a lot of what ChatGPT can do — churn out something that has been done before. But what it is not (currently) good at is original, creative ideas and since much of humor writing is about trying to write something that hasn’t been done a million times before already, it doesn’t seem like it’s going to be a good tool for just spitting out a good satirical piece with one simple prompt.

That said, some writers are experimenting with using tools like this for parts of the writing process —like brainstorming topics or headlines—which I think will continue to be a thing.

At the end of the Vulture piece, one of the comedians says they believe ChatGPT can get good enough to write something with a specific style like New Yorker cartoons. And in this Medium piece, a writer tried to get it to write New Yorker cartoons and in the end, he thought some of the captions were “not too bad.”

But it took some finessing to enter very exact directions to arrive at a cartoon punchline that the writer was happy with and the writer added his own drawings to it, which I would argue gives the cartoons an original spin and makes up a portion of the creative work being done.

What counts as original?

It seems like the bigger question is going to be “How much of something written by AI is too much to count as original?” Some people have said that using it to brainstorm is probably not a problem but using it to write actual parts of a piece without acknowledgment potentially is.

From what I’ve heard, there isn’t really a reliable way to detect something written by AI, so checking for it in a way you might check for plagiarism is trickier.

ChatGPT does make you sign up for an account and I’m assuming it stores all the searches you run, so the conspiracy theorist in me wonders if at some point they will use that info to point out all the ChatGPT content that is surely already being written.

Of course, I am already using some AI tools to write—I use tools like Grammarly to check grammar and Rhymezone to help come up with words for pieces at times, but crossing over into ChatGPT to help write humor doesn’t feel particularly useful to me right now.

The question of what counts as original is one I’ve run into with my college students in discussing and managing plagiarism too. One piece we sometimes talk about in class is this Malcolm Gladwell one where a writer took parts of existing works for a play but essentially made something new. Since she transformed it, how problematic is it? It usually makes for a good discussion because it’s complicated to answer.

As a writer, I’ve also come to realize that a lot of people do come up with the same idea at the same time. I’ve had basic ideas that have never left my brain or the notes app of my phone that I’ve seen another writer publish. Other humor writers are also often inspired by other pieces — trying out forms or techniques they see others using. While just copying exact wording from another writer is obviously problematic, other types of inspiration are different.

Several years ago I was discussing what is and isn’t plagiarism with one of my comp classes. We were talking about the “mosaic” form of plagiarism where someone copies a chunk of text but then changes some words here and there. At one point in the discussion, one of my students said, “It seems like it’s probably easier to just write your own paper. Doing something like this would be a lot of work.”

And I could see the same being said about trying to use ChatGPT to write something like a satirical humor piece. A writer could probably get it to help, but it would take a lot of refining and editing and at the end of the day maybe it’s still faster to just use your own brain?

---

March 21, 2023 Addendum: Of course, after I had written and scheduled this post a new version of this AI (GPT-4) was released. But from what I’ve read, it’s still mostly telling dad jokes.

It sounds like it is good at explaining memes, so I guess that might be helpful if you want one explained?

The technology is changing so fast that it may obviously get better at humor, but it doesn’t seem like this current update is vastly better from what I’ve read so far.