How humour is a multitool for writing and communication, and why AI struggles to use it.
There’s a quote attributed to George Bernard Shaw, Oscar Wilde, and Charles Ludlam that goes: “If you want to tell people the truth, you’d better make them laugh or they’ll kill you.” It highlights an often overlooked hazard of human connection: When you reach out, sometimes you get more than you bargained for. The opening track of El-P’s 2007 album, ‘I’ll Sleep When You’re Dead’, acknowledges the potential for this unexpected recoil and the difficulties of extricating yourself from an uncomfortable situation that you have landed yourself in.
AI isn’t funny. Occasionally, it makes us laugh, but these responses are never with the AIs, always at the AIs.
While Anthropic’s latest test paints a picture less Asa Candler and more Jacques Clouseau, Barclays makes dour portents about the advertising industry’s adaptation to the AI age, and yet another random venture capitalist in tech pops up on a podcast nobody cares about to preach that we’re on the brink of AI-driven market collapse… not they have any financial incentive to propagate that message, of course. They’re just passing on the good word. All the while, left to their own devices, the promised next frontier in agentic AI can’t make it through a children’s game without significant human interference as demonstrated by Gemini…
This leads me back to writing, and another human oddity that is so irrational and esoteric that it could be the result of trillions of cells tied together by electrons understanding itself through a series of electro-chemical signals being abstractly interpreted by another series of electro-chemical signals: Humour.
In fiction, humour can highlight an element of tension or a key theme in the narrative. Consider how Joseph Heller uses humour in Catch 22, and how Blackadder’s final episode utilises humour to hysterically outline the paradoxical logic of war. The juxtaposition of the severity of mass death against the mundanity of living in the absurdist bureaucracy of the industrialised military system. Humour in these cases serves to highlight and magnify the sense of farcical tragicomedy whilst making it bearable.
In nonfiction and professional life, using humour in writing can strategically mediate tone, audience reception and outcomes, across a variety of circumstances and intentions.
Humour can function as an expectation setter: Someone who doesn’t want to give their audience or conversation partner the impression that they desire a deep discussion might set the tone with a joke. Humour can function as a coping mechanism: The famously bleak gallows humour of soldiers serves as a means to mediate the emotional toll of their profession, enabling them to continue to function. It is, of course, not restricted to the military – everyday people do the same thing. If you’ve ever had a friend who is just a little too funny too often, sometimes it pays to keep an eye on them.
Humour, then, can be understood as a powerful and versatile means of connection. And connection, ironically, is the profession of that most romanticised recluse: The writer.
Humour is one of the most effective, and one of the hardest to use, tools at a writer’s disposal. I’ve talked about why AI sucks at writing through the lens of symbolism and subtext, and intuition. Now we turn to humour, a powerful but situational tool in the writer’s repertoire, and why AI will continue to struggle with it. While the mass of talking heads, social media pundits, and financially incentivised folk continue to assure us that AI is coming for all artists, the simple truth is that these people don’t know what they’re talking about. Probably because they don’t create themselves and they’re too lazy to put in the effort needed to get good.
Let’s have a gander, shall we?
How does humour work?
Alright, first things first: I’m not going to write a comprehensive thesis here. This is an incredibly complicated subject with a multitude of nuances, variations, and theories. There are probably more books, courses, and media devoted to it than you could fill a large library with. I haven’t read, watched or listened to any of them, so if you know a good one, by all means, fill us all in. But you didn’t click this post to learn how to be funny and I ain’t a comedian. If you want to learn how to be funny, have an affair with Mitch Hedberg. With that said, let’s start with a few disconnected notes and see where we end up…
‘Benign violation theory‘, child of Pete McGraw, suggests that amusement arises when something is simultaneously benign and violates some kind of norm in an abstract fashion. Thank you for coming to his TED Talk…
YouTuber Casually Explained suggests subverting an expectation. For instance, someone I knew was complaining about driving through London during the school run. She ended a short rant with “bloody people and their procreation!” I responded, “Christ, on the motorway!?”
Another funny thing – you didn’t laugh at that, did you? Thought not. A bunch of people in the room did, but if they did and you didn’t, then either they were humouring me, or there’s something interesting going on here. I’m not trying to excuse a dad joke, this is centrally important to using humour in writing. You know that ‘you had to be there’ thing that people out when a joke is about as funny as a flayed octopus? It impacts the way in which humour functions fundamentally differently in writing as opposed to in meat space. Written humour cannot take advantage of the multitude of subtle social cues that inform our ability to be funny in person. That lack of audience response doesn’t make it easier, it makes it much harder.
And we have, of course, all heard that the secret to comedy is timing. Arnold Brown noted, “You the audience come here tonight… and I arrive one week later… that’s bad timing.” One of the reasons you didn’t care about the aforementioned motorway sex joke is because there’s no sense of timing to it.
And then there’s the ‘brevity is the soul of wit’, advice from Hamlet. Robert Lund may have responded with “If so, then why’s this bloody play so long?”, but Shakespeare was a funny writer. The reason kids hate learning about him, aside from the chore of translating Elizabethan English to modern parlance, is because their English teachers aren’t allowed to point out the fact that every other line contains a dick joke. And they do. If the English teachers were to point this out, more kids would enjoy Shakespeare, and more mummies and daddies would cack themselves dizzy in a fit of self-righteous indignation, before sending the respective schools of their delightful little troglodytes, a deluge of strongly worded letters with thinly veiled threats about funding.
But think about it: Shakespeare wasn’t writing for educated idiots like myself, who sit here bashing out thousands of words about the intersection of technology and art. He was talking to the guy fresh in from ten pints in the pub after a week of doing awful manual labour in awful conditions, trying to get it up for prostitute at the side of the pit. That is not hyperbole. The aspirational sheen of academic civility attached to the theatre, and by extension Shakespeare, in the modern era is lightyears distant from the world in which English Literature’s favourite bard was scratching in. Now that’s hilarious.
Why is humour important for writing?
One benefit of humour is its ability to accelerate the process of connection forming. Where writing is concerned, you can use this when onboarding an audience. You don’t know me, you’re trying to feel me out, to figure out whether or not we are going to get along. If we can share a laugh, the chances are that we will. Just like that, you’re in. You might not be a convert, but you’re at least willing to listen.
At a previous job, I was often the introductory point of training for the newbies. Showing them around, giving them a high-level overview of the systems and workflows. This meant that they were bombarded with information in the first week. The very first email that received contained about five or six attachments, most of which wouldn’t be relevant until a week or two later. As you might imagine, that can be overwhelming to someone who has just walked through the door.
So in the introductory email, and for several other email templates I had saved (like the absolute minefield that is open access), I made sure that the tone I was using was light and included some humour. I could anticipate and recognise the probability that they were likely taken aback by all of this arcane crap I’d just dumped on their laps. The use of mild humour allowed me to pre-emptively respond to it by establishing the idea that they didn’t need to worry if I wasn’t taking it super seriously.
An LLM doesn’t have that foresight, nor that connection to the human experience of starting a new job, and the pressure to perform and demonstrate that you’re a valuable asset. Especially as a new entry into a precarious job market. It knows how to answer questions and provide information, which is what we want out of it. Despite the algorithms tailoring their output to suit the tone of the user, it has no real concept that it should be framing its answers in reference to the broader context.
If I’ve just read yet another spew of jargon-drowned word salad from some company that wants to sound smarter than it is, then I might ask ChatGPT to ‘translate this into English, please…’ When my questions are snarky, I’ve noticed that it tends to reply with a similarly flippant tone. That’s actually reasonably impressive. On the other hand, it’s often the case that tone matching isn’t the answer in many situations.
If someone is noticeably nervous, then you certainly don’t want to respond with nervousness. In some professional situations, if you’re in a position of responsibility, sometimes someone is being a bit emotional because they’re on edge or worried. In those situations, there is no benefit to mirroring them. Your role is to demonstrate and embody the characteristics that you want them to demonstrate in turn; to lead by example.
If you can make someone laugh, then you can probably get along with them. If you can make someone laugh in writing, you can get them to turn the page or to read another post, and also achieve whatever aim it is that you’re setting out to achieve with your writing, be that to inform or entertain or whatever.
Why so serious? Humour, severity, and criticism
You’ve probably done this once or twice without thinking about it: attempt to inject some levity into a serious situation in order to regulate tension. It may or may not have worked. Regardless of your experiences, it’s worth keeping it in mind when dealing with For instance, you can use humour to deflate the perception of aggression or harshness when providing feedback or criticism, by using it to bolster the idea that you’re still on the recipient’s side and you are attempting to address or clarify a problem rather than attack them.
It’s extremely important to note in these cases, that you need to be very careful with the tone and audience. Humour should be very limited and reasonably subtle, and relevant to the subject, injected at strategic intervals. The aim is to lighten the mood and support your position not distract from the point. But for the love of God, read the room first. If you’re in any way unsure as to whether making a joke will work in a serious context – just don’t. I don’t need you sending me all-caps death threats because you told your CEO a dead baby joke… A joke that doesn’t land or which is interpreted in the wrong way can be more harmful than simply being perceived as a bit hostile or dour. Misconceptions can be clarified and accounted for, with ease, at a later date. It’s far harder to repair reputational damage if you accidentally offend people with blundering attempts at making them laugh.
On the rare occasions that I’ve emailed higher-ups to provide feedback on things that weren’t working particularly well, I’ve employed strategic humour. Obviously, you need to keep them on your side, and a straightforward way of doing that is to highlight a sense of connection by empathising with their stressors and acknowledging that they’re just as swamped as you are. You don’t need to turn that into an excessively earnest bid for emotional vulnerability; in fact, in a professional setting that’s probably going to work against you. Putting in a little shared joke early on let my managers know that I hadn’t just barged into their overstuffed inbox to make an all-caps Facebook rant. It also helped that I was on good terms with them.
Naturally, if there are problems in professional settings that don’t need immediate attention, then you’re not going to get off with just dropping a joke on someone at random and then launching into a tirade about how your company needs to upgrade the RAM of the employee laptops or something. This is also more effective in small companies where people know each other and have a better idea of who they can be less formal with and in what situations. Small companies are also less likely to have full-on HR departments, so there is that…
Like I said: Read the damn room.
The same idea can be applied in more general ‘serious’ contexts to put other people at ease. If you’re going back and forth with some kind of customer service member, you can use humour to make them more willing to help you. Put yourself in their position: they get screamed at 10 hours a day by people who are barely intelligent enough to tie their own shoelaces. Most of the time they have no control over the problem being screamed about, they’re just there to absorb the initial barrage of frustration. The other half of the time, the customer has created the problem themselves, and will never admit it, and now the service rep. has to appease the ego of that idiot, while also extracting productive information out of them before directing them to a solution in a way that won’t make them fly into a fresh tantrum.
Years ago, I ordered some food to be delivered and the food didn’t come, I wondered what was going on, and I had this series of exchanges with a member of support to establish the situation. At some point, she asked whether I wanted her to contact such and such a team to see if they could do anything. Based on the information she’d given me, I knew she was just saying that because of the amount of time that had elapsed and so she felt compelled to. So I said, “Let’s be real, that’s just going to result in more work for you and a circular answer from them that won’t help either of us, so let’s save both of us the wasted time.”
How is that funny? Well, because humour doesn’t have to be laugh-out-loud jokes. In this instance it was a shared wink and a nod recognition of the futility of both of our positions, the fact that we both knew this and yet she was compelled to continue acting as if circumstances were contrary to what they were, ultimately creating an absurd situation.
This is what I mean about subtlety: you don’t have to tell a knock-knock joke. You can just lean on the absurdity of a situation that all parties will recognise. The good thing about that is that the world is becoming increasingly absurd at an increasing rate. So we’ve all now got an infinite source of humour at our disposal.
In the end, I got my problem resolved. Zero stress and they were more responsive to me because I’d used humour to bring them on side and recognise their situation, and I got a more prompt response and a better experience as a result. They got an hour’s respite with a vaguely amusing berk who didn’t threaten to burn their house down.
Humour is also a great tool for maintaining engagement with a generic or boring topic. My old job had periodic waves of people leaving in other departments, which meant that every so often we’d need to tell all these new people that we were working with, who we were and what we did. That meant I gave a bad PowerPoint presentation to a department that probably couldn’t practically benefit from the information I gave them and would almost certainly forget all of it under the tide of admin and emails and day-to-day responsibilities. This meant, in practice, that those meetings were well-intentioned, but from a practical perspective consisted of editors bombarding them with very dry information for which they had very limited use.
People who are bored with you are not your friends. You still have to work with them. Working with people is a hell of a lot easier when they like you. How do you make them like you? Keep them engaged. How do you keep them engaged? Make them laugh. Or at least smile. PowerPoint isn’t exactly the chosen vector of delivery for the comedy greats, so calibrate your expectations accordingly.
So, for example, on one slide, explaining feedback to authors, I had a gif of Clippy and a speech bubble: “Looks like you’re trying to DELETE YOUR ENTIRE ARTICLE – would you like help with that?” This was legitimately a thing that happened with startling regularity. Not exactly the height of comedy, but in a small conference room on a hot afternoon, it helped to keep them a little bit more focussed on a subject that, realistically, they didn’t have a great deal of practical investment in. That shred of emotional engagement with the bland subject matter got them on side. If they had a question for me later on, they might have had the vague impression that I’d made their mouths twitch upwards for a fraction of a second at some half-remembered point in the past. They add an extra smiley face to their Outlook message. Business chortles along…
We get it, you don’t like AI!
Wrong!
I’m just slightly tired of all focus going to the wrong places. As I’ve said previously, it only makes sense that most of us tend to think of LLMs are purely glorified chatbots and image generators – because most people aren’t involved in logistics, data analysis for real-time targeted promotions, or inventory management. Most people don’t care about what goes on during the process of drug development, they just want the end product. And to be fair to OpenAI, Claude, Gemini et al., the public-facing applications have still been quite useful. Instead of spending hours trying to gather relevant sources for research, grabbing a list of relevant articles, books, and data on any given subject is just a prompt away. The foundational step in the research process has never been easier.
There are uses for this technology, and we are going to see massive developments as we take advantage of its vast pattern recognition capabilities. Prior to Apache Hadoop, we struggled to deal with ‘big data’ sets that no human could hope to make sense of and no singular software could tackle. Now we seem to have built the tools to help us with that stuff, but we’re trying to convince everyone that getting it to draw ever more elaborate nudes of your mother is the best use case for it.
As ever, the best use cases for all of these dynamic algorithmic systems tend towards the logical. Because, as anybody who has ever tried to learn code has discovered: You do not realise how much intuitive thinking you do until you try to get a computer to do something very very simple. I wonder if it’s due to people’s general unfamiliarity with code that they mistake the way in which AI thinks, for a genuine analogue of the human brain.
I wonder if that’s the reason ‘vibe coding’ became such a big thing for a couple of weeks; because it cuts out the tedious bit where you have to actually understand that a computer will interpret what you’re telling it in very specific ways, and then reverse engineer that into something that won’t cause your computer to blue screen.
Why can’t AI be funny?
Before the pandemic, the London rush hour Tube was full of people crammed into every square inch of carriage space possible. Often enough in the crush of bodies, someone would be forced into an unexpectedly uncomfortable position that, in another circumstance, would provoke some kind of response. But in these instances, you’d often observe a sort of silent shifting in an abortive attempt to solve a problem that neither party thought would work, and then a silent exchange, a grimace and a shrug; both recognising that neither of them wanted to be in that position, but were effectively forced into it by economic and social demands beyond their control, and that their current predicament was neither novel nor confined to them.
As with a lot of communication, humour is conspicuously dependent on intuition. AI has no sense of the absurd or empathy or norms and the nebulous makeup thereof. It just has vast stores of data. It can recognise a contradiction, but it can’t empathise with the sense of bemused dissonance that contradictions evoke in people, nor can it remember or connect with the shared unspoken knowledge that such scenarios are increasingly common.
AI has no sense of delivery. Humour is hard enough when spoken aloud. But as evidenced by my anecdote about my friend complaining about the commute, what works in person often doesn’t translate to the page. Writers like Terry Pratchett and Joseph Heller, who can be consistently funny in a medium that lacks many of the immediate contextual clues that tickle the nodule of brain matter that makes the laughter happen, are way more impressive than we give them credit for.
Humans draw on their lived experience of the world around them in physical space to inform their humour. As Anthropic’s disastrous attempt at putting an LLM in charge of drinks vending illustrated, AI just doesn’t have that. There isn’t an algorithm to express all of the long-term facets of lived experience that drive the creation of humour, let alone an individual’s ability to assess whether their audience is likely to be receptive to their particular brand of humour, with environmental context, when judging whether it is appropriate to make a joke, and if so, what kind of joke to make.
Could AI be funny in the future? In the immediate term, I wouldn’t get your hopes up. We’re still trying to navigate short and long-term memory access and compression in these models, let alone whether they can mimic the multitude of subtle data points that go into making a joke that works. In the long term… I won’t rule it out, but I’m also going to assume that in a theoretical world where an AI can perceive its environment with sensors, and use that to inform the creation of a synthetic personality, that said personality is likely to be quite alien to human perception.
Humour, at its core, is about connection and connection relies on shared experience, if only in the abstract. The way in which a machine processes experience and a human processes experience are fundamentally different, and the output from that processing will therefore inevitably be distinct. That being the case, I would assume that, in the event that an AI were capable of organic humour, that said humour wouldn’t translate to humans. Conceivably, where communication is concerned, AI will always struggle to “connect” with humans.
The idea of the alien often stands in for something we do not recognise and cannot fully understand. Science fiction authors have always pictured AI as being human-comprehensible because AI usually stands in for human-based themes, and left aliens in galaxies far far away. In the theoretical future where anything approaching AGI is even possible, we might find that we have built the aliens here on Earth long before we discover them in outer space.
Words and Chaos 
Leave a comment