On Artificial Intelligence and Ulysses
In which I catch Google AI in yet another 'hallucination' while beginning my second journey through James Joyce's epic novel
I’m reading Ulysses again, which is cooler than “I am reading Ulysses.” Many who begin never finish. I did, a dozen years ago and that novel has lived in my brain ever since. My probably pretentious opinion: Not everyone “needs” to read Ulysses, but anyone who regards themselves as serious about writing, about a life immersed in language and prose or poetry, does need to read it.
It takes a village to read Ulysses, by which I mean: It is best to have someone within reach who has already read it, preferably more than once. I had the company of a guy I met over on Twitter who was obsessed with Joyce, and I also relied on supplements: Stuart Gilbert’s James Joyce’s Ulysses: A Study, The New Bloomsday Book: A Guide Through Ulysses, The Cambridge Companion to James Joyce, and Anthony Burgess’s ReJoyce, among others.
But the biggest help was not a book: It was listening to Ulysses read — or rather, performed, in the 1982 RTE radio dramatization of Ulysses. Actually, both: It is read and performed, with a single actor doing the narration, and actors cast for every speaking role, accompanied by appropriate sound effects. It is a monumental achievement, arguably one of the greatest made-for-radio artistic triumphs of the 20th century. It’s available free on Apple Podcasts.
I didn’t listen to the entire thing, but I did enjoy large swaths of it, sometimes after I’d read a chapter while walking and other times while reading along. It was invaluable.
I’m reading it again this summer because in August and September I’ll be taking a class on Versed Community where we read the entire book. I’m getting reacquainted by perusing some of the episodes. I’m especially in awe of the Holles Street Maternity Hospital scene in Oxen of the Sun, where Joyce syncs the entire history of the English language with the biological stages of fetal development. That is one crazy-ass piece of writing.
But I’m not here to talk about Ulysses, I want to talk about Artificial Intelligence and give you a real-world example of the perils contained therein.
When I read Ulysses back in 2014, I read the “corrected” text of the Gabler edition — “It fixes 5,000 errors … involving punctuation, omitted words, phrases and even entire sentences,” The New York Times declared, “an average of seven flaws for every printed page.” I remember seeing a woman carry this edition to a lit class I took in college and being very impressed. No, I didn’t ask her out.
Turns out we’ll be using an earlier edition for the class, the 1961 Bodley Head text. Out of curiosity, I hopped onto Google AI and asked about pairing editions of Ulysses with the RTE production. Here’s our exchange, edited for length and clarity:
If I wanted to read along in 'Ulysses' while listening to the RTE reading of the novel, which edition should I use so that I get a word-for-word listening-and-reading experience?
Because the RTÉ Radio 1 dramatization is a highly abridged, adapted audio-drama rather than a verbatim recitation, there is no single print edition that will give you an exact word-for-word read-along experience. However, to get the absolute closest match to the spirit and text of the broadcast, you should use an unabridged text based on the 1922 edition, the Gabler Edition, or the 1961 Random House/Bodley Head text.
I was under the impression that the RTE production wasn’t abridged at all, are you sure this is correct information?
You are entirely correct, and I apologize for my previous error. The landmark 1982 RTÉ Radio dramatization of Ulysses is indeed famously unabridged. It was a monumental 30-hour broadcast that features every word of the novel, meticulously produced to celebrate Joyce’s centenary. Because it is fully unabridged and performed as written, you can absolutely have a word-for-word read-along experience. You should use the 1961 Random House / Bodley Head text. Do not use the Hans Walter Gabler “Corrected Text” (1984). Because the Gabler edition was published two years after the RTÉ recording, it introduces hundreds of specialized textual changes, spelling alterations, and formatting shifts that will conflict with what you hear the actors say.
I don't understand: If the RTE production was, as you now say, "famously" unabridged, why did you initially say that it is "highly" abridged?
Your confusion is entirely justified, and I owe you a direct explanation: I hallucinated the information in my first response.
This is about the fourth or fifth time I’ve caught Google AI hallucinating. Recently it made up a short story by William T. Vollmann that he’s never written. I’ve never used ChatGPT or Claude, and have no desire to.
I’ll never use AI to write anything, full stop. Insofar as using it for research, I suggest proceeding with extreme caution. Like Trump, it makes shit up that anyone can check for themselves. Unlike Trump, it admits the error when called out and apologizes. For now, at least.
