Tactility and Care

Why Do I Hack Comedy?

Lan Zhang

The word “hack” has many meanings. In the computer programming realm, “hack” is known as an act of gaining or attempting unauthorized access to a network or computers[1]. I found out its double entendre halfway through my thesis research and experiments into Computational Comedy, where I was attempting to produce comic text from borrowed materials with procedural methods. While scraping comedy transcripts online and jumping down the rabbit hole of hyperlinks, I found the Wikipedia page for “Hack (comedy).” This term is long-established in the comedy industry, referring to materials obviously copied from original comedians. The coincidence felt uncanny but gratifying as if I won an internet treasure hunt. It was apparent that whatever computational comedy I was pursuing, the process can be considered unorthodox and potentially upset some people, in both the programming and the comedy realms. “That’s perfect,” I thought. 


Rewinding the story to the summer of 2019, I was riding in the car with my partner. We were listening to a podcast as we often did, and we started laughing about numerous things mentioned by the host. At one point, I began to notice that our synchronized laughter was intuitive and harmonized. I couldn’t have mimicked my partner as we were looking ahead. My reaction was solely in response to the English speech coming out of the radio. It had words, words being composed and pronounced rhythmically, phrases, short and long sentences, broken-up sentences, gibberish, pauses, accented expressions, repetitions of sound bites, sighs, gasps, accumulation of emotions, made-up scenarios, references, narratives that reflect the present and the past, and everything else in American English. I suddenly had a flashback of me struggling to get a grip on using English freely and comfortably years ago. Unlike the past me, I, in that car ride, unlocked a secret. How did my brain wire to this fascinating space, laughing at the surrounding humor, no longer needing a prior rehearsal? Yet, there are still so many aspects of American Culture I fear being in conversations about. Yes, I laugh to more things now, but overall, American humor and comedy are still in my uncomfortable zone of top tier intangibility. I believe many people like me possibly feel the same way. Peter McGraw, in his book “The Humor Code,” said:

“Of course, you don’t have to be a minority to be a great comic. But either way, it seems helpful to cultivate what W.E.B.D Du Boise called your ‘double-consciousness,’ you “two-ness”. Yes, in the United States this phenomenon has been a bad thing, something that’s kept people fractured and suspicious and struggling with self-identity. But on the bright side, it also makes for good comedy.”[2]

I wanted to bring these questions and tangled thoughts embedded in my identities in my creative practice through the lens of computation and conversations with myself. I was told that humor is seen in most of my creative work. I used humor subconsciously as if it was my voice of delivering difficult messages and my diplomatic means of connecting with my audiences. However, If I wanted to use programming as my method to creatively conceptualize this comedy riddle that I felt pretty clueless about, I needed to know what I was breaking into and what was “hackable.” People have written books and theories about breaking down the humor code. Yet, dissecting humor is still one of the most complicated machine riddles. Many factors go into the complexity such as nuances, cultural relevance, language gaps, joke deliverers’ charisma, their personalities, and so much more. These are already difficult for humans to comprehend, not to mention machines. However, we are also at the most promising stage where powerful learning models and computational methods are available for taking up more significant challenges. At least that was how I optimistically believed throughout the first months of my thesis.

I invested a lot of my energy in trying out cutting-edge text-based machine learning. I also conducted a lot of small-scale experiments that I took a stab in such as two-liners, jokes, long-form analysis, improv games, improv game rules, generative writing, and so on. By late November 2019, I ended up with a chat application that uses the most advanced language model GPT-2, that I fine-tuned with hundreds of scraped stand-up comedians’ transcripts. Relying on the existing model made many steps of the process out of my control and analysis. The front-end interface took forever, after users entered their chat messages as the starting point for the back-end machine learning model to calculate and generate. I landed somewhere excitingly interesting, but not quite personal and low-level as I wanted.

I did find that, though the machine has yet to master the art of humor as seen in human comedians, the act of machines or algorithms attempting to become funny with even non-sense, ridiculous language, is entertaining and thought-provoking to people. Over the winter break, I started to reevaluate methods of reconfiguring comic text with a library of scraped stand-up comedians’ transcripts on my hand. Daily, I tried to make a small sketch/algorithmic function that manipulates these texts in some ways. Gradually, I started to filter out specific comedians and worked with their materials individually.

Some have asked me if I was aware of my source not being objectively chosen. It is undoubtedly true that the feeding source of my application was biased. However, my intention and trajectory need my decisions to be personal, and the strongly I feel connected to the project, the more relatable it is to me. Therefore, not only did I start to work and create content with individual comedians, but I also started to program my version of the n-gram/Markov Chains algorithm, (a statistical model that represents probabilities of events in sequential order), from scratch. This method was not as “magical” as machine learning because every output is precisely calculated and reflective of the source. I started to modify the model by merging different transcripts and creating mini conversation portals between me and these computational comedian decoys. I began to have so much fun playing and observing these virtual comedians deliver materials as if they were the real comedians. (Well, the content technically was from the real comedian).

Slowly, I realized that you’d see most female comedians’ output content to be about birthing children, pregnancy, and frustration or pride being women. Most comedians of color poke fun at their stereotypes. Men talk about women versus women joke about themselves. These generated outputs composed a larger narrative. They revealed the themes and identities of the American Comedy landscape to the barebone. Unfortunately, I didn’t become a comedian from building these modules, but I did unravel some exciting insights from watching these virtual comedians “deliver bits” side-by-side. With the help of precisely set-up algorithms, maybe I did unlock and hack American comedy from an alternative perspective.

My final deliverable for Hack (Comedy) is a web interface that allows visitors to write with these virtual comedians of my choice, individually or side-by-side. I hope my visitors have fun and gain insights while exploring this unexpected way of programming to reach the American humor pedestal. The process, however, reinforces the failure of my ideal pursuit of cultural competency and the comic absurdity of the pursuit itself. You can try out the application at www.hackcomedy.net.



1. Merriam-Webster.com. "hack." 20th April 2020.
2. McGrawPeter. The Humor Code: A Global Search for What Makes Things Funny. New York: Simon & Schuster, 2015.
1. Merriam-Webster.com. "hack." 20th April 2020.
2. McGrawPeter. The Humor Code: A Global Search for What Makes Things Funny. New York: Simon & Schuster, 2015.
Back to Top
Lan Zhang is a developer and designer based in Brooklyn, NY. She is an MFA student at the Design and Technology program at Parsons. She is also a Recurse Center Winter2'20 alum.