Although I hate most uses of generative AI, I sometimes find myself curious about it. I think that’s just because it’s incredibly powerful and whilst most people will use it out of laziness, greed, stupidity, hate, and selfishness, there are definitely a great many potential positives. Its application in medicine is one example.

Then there are the stupid things I’ve done by testing it on the Beat Generation. Not long after ChatGPT was first released, I asked it a range of questions and assigned it various tasks related to the Beats. It was not hugely impressive and I was not surprised. To be honest, even real humans who spend hours doing research still get the basics wrong. (Not included in the previous link is the fact that it gave me totally made-up dates for Neal Cassady’s birth and William S. Burroughs’ death!)

Later, I asked it to analyse and then rewrite Ginsberg’s poem, “Howl.” The results were sort of interesting but of course it was not an improvement. It just served to show the limitations of AI.

When you learn a little about AI, and when you are used to studying the Beats, you see why it gets so much wrong. After all, there are a huge number of misconceptions even about famous events. Even the experts get stuff wrong, never mind a computer program that is at best just pulling information from a few highly ranked websites.

Today, I gave it an easier task. In fact, this is the sort of thing that AI is made for: basic statistical analysis. Let’s leave art and history to the humans and let computers handle the boring stuff.

I asked ChatGPT to find the 20 most common adjectives used to describe Allen Ginsberg, then repeated that assignment for Jack Kerouac and William S. Burroughs. I figured this would be a nice, easy task because it has the capacity to search websites and documents quickly, pulling adjectives and then ordering them according to frequency of use. A human could only do a small sample over many hours or days, but ChatGPT managed it in seconds, with another minute added to compile it as an Excel document. Here are the results:

Allen GinsbergJack KerouacWilliam S. Burroughs
VisionaryRestlessSubversive
RadicalNomadicDark
OutspokenFree-spiritedProvocative
IntrospectiveSpontaneousCynical
ProvocativeMelancholicExperimental
CounterculturalReflectiveAlienated
Avant-gardeRomanticSatirical
PassionateIntrospectiveDisturbing
ControversialRawNihilistic
BohemianBoldAnti-authoritarian
Thought-provokingExperimentalSurreal
CharismaticLyricalDetached
HonestSensitiveParanoid
InsightfulObservantMacabre
VulnerableIdealisticUnorthodox
PoliticalUnconventionalDystopian
RevolutionaryAuthenticFragmented
Anti-establishmentPassionateControversial
ExperimentalExistentialPessimistic
CompassionateAdventurousInfluential

As expected, the results are pretty reasonable. The task was well within the capability of ChatGPT.

Considering how successful this experiment was, I repeated it with Gemini, Google’s AI bot. It was actually much less impressive. It basically picked the same few adjectives, suggesting that it had chosen a certain source that grouped all three writers together rather than searching across the internet for common use. It even struggled to make a table after multiple instructions!

Beat PoetAdjectives
Allen GinsbergVisionary, Radical, Revolutionary, Controversial, Iconoclastic, Provocative, Experimental, Innovative, Groundbreaking, Influential, Charismatic, Dynamic, Passionate, Intense, Raw, Honest, Vulnerable, Sensitive, Spiritual, Mystical
Jack KerouacVisionary, Radical, Revolutionary, Controversial, Iconoclastic, Provocative, Experimental, Innovative, Groundbreaking, Influential, Charismatic, Dynamic, Passionate, Intense, Raw, Honest, Vulnerable, Sensitive, Spiritual, Mystical
William BurroughsVisionary, Radical, Revolutionary, Controversial, Iconoclastic, Provocative, Experimental, Innovative, Groundbreaking, Influential, Dark, Disturbing, Narcissistic, Nihilistic, Cynical, Misanthropic, Paranoid, Addicted, Alienated, Existential

Notice how all the adjectives are the same until “influential,” after which they differ for Burroughs, whilst the lists for Kerouac and Ginsberg as the same. How lazy!

Another quite amusing fail from Gemini occurred when I asked it to make a word cloud image of the above data. It produced the following:

An AI-generated word cloud image.

I can’t even make out a single word. It looks almost like Russian… I asked it to repeat the task several times and eventually received an apology and an explanation that it is not good with word clouds. Ok, but there are dozens of word cloud programs online that are comparatively simple. I used one to make the header image for this article.


What is the point of all this? Well, I still maintain an ambivalence towards the technology but admit that it has certain purposes. However, you do have to understand its capabilities and give it tasks that are realistic. In the past, I jokingly told it to improve “Howl,” knowing that it would fail. I asked it to write about the Beats, expecting uninspired, formulaic, high-school-level answers, and that’s what it did. When it comes to images, it is just awful.

Someone once called ChatGPT “the perpetual B+ machine” and that’s about right. Beat scholars don’t have to worry about AI replacing them any time soon. However, they may consider incorporating it into their work. Think about how the above task could be altered for statistical analyses of Beat texts, for example… It seems to me that training AI to do boring, difficult, time-consuming tasks could provide people with the opportunity to do more meaningful and creative work, and this is true for those of us in the Beat Studies world.