Issue #314: An Overblown AI Scandal
In Massachusetts, the weather is turning a bit, although it appears it will be much worse tomorrow. But I’ve been able to get outside. I haven’t felt like I have had much time to watch movies, but Badland Hunters (2024) is on the top of my list. I’m also a bit behind on TV. I’ve been watching Mr. & Mrs. Smith (2024). It has an impressive legion of guest stars and deserves the honor of being the first mumblecore spy drama. True Detective season 4, Tokyo Vice season 2, and Fargo season 5 are all competing for my attention but I haven’t started a single one. I only watched the first season of Noah Hawley’s Fargo, but it’s one of my favorite seasons of TV ever.
I also wrote what I thought was one of my better newsletter bits about it.
Otherwise, I am in the dissertation salt mines. But it’s going great, thanks for asking.
This Was Originally a “Weekly Reading List” Segment but I Decided It Was Too Long and Made It a Separate Essay about Generative AI — not written by AI
https://www.vulture.com/2024/02/george-carlin-ai-comedy-podcast-dudesy.html — Over the past couple of weeks, there has been a news item haunting RSS feeds. The item in question? A George Carlin special, posthumously released, posthumously recorded, allegedly created by AI. First it was the promotion and righteous indignation (the two, obviously, indistinguishable) directed toward podcast Dudesy and the podcast’s hosts Chad Kultgen and Will Sasso. Then the suit filed against Kultgen and Sasso by the Carlin estate, claiming “copyright infringement and violation of Carlin’s right of publicity caused by Defendants’ unauthorized use of Carlin’s works to create their AI-generated ‘George Carlin Special’”. This would all be very interesting, and even precedent defining, if this case had anything to do with AI. Hilariously, however, it doesn’t.
I am familiar with Chad Kultgen’s work as the host of another podcast, Game of Roses, that I have written about in the past. There, Kultgen presents himself by the moniker BachelorClues and recaps The Bachelor franchise shows from the perspective of a sports commentator. Above all, though, Kultgen is a comedian (by way of serial podcasting). And it was with morbid intrigue that I tuned in to the first episode of Dudesy. The premise of the podcast makes the claim that Dudesy is named for an AI that writes the show. The Dudesy AI, supposedly, consumed an exhaustive archive of all of Kultgen and Sasso’s information: their previous published work, their private emails, text messages, and media consumption habits through various streaming services. This is, of course, ludicrous. Kultgen explains to Sasso in the first episode that Dudesy accesses a wide range of services with no API for consumers, apps, or AIs to access usage records.
What’s even more telling about the introduction of Dudesy is the terminology Kultgen uses to introduce it. It is always nebulously an “AI,” no greater specificity than that. They rarely refer to it as a “model” or the model’s “training.” Instead, it “reads,” “knows,” etc. That seems strange to a modern audience given the proliferation of the idea of the large language model (LLM) and what it’s capable of. But LLMs didn’t enter public consciousness until the public launch of ChatGPT on November 30th, 2022. The first episode of Dudesy aired on March 8th of the same year.
My indignity regarding the capabilities of the supposed Dudesy AI may have been a little naive given how much the cultural understanding of what AI can do changed from March to November. But there are still parts of Dudesy’s supposed capabilities that are beyond even the most advanced LLMs. And, I was right. In the Vulture article above:
In a development that should shock no one, a rep for Sasso told the New York Times that the AI central to Dudesy’s conceit is not actually an AI, but a “fictional podcast character created by two human beings.”
Of course, it is easy to pretend to be unsurprised by something after the revelation of the facts. Dudesy (the fake AI) was convincing enough to trick the Carlin Estate into expending legal fees to sue its custodians. It was convincing enough to rile up a laundry list of pearl clutching social media posters and celebrities. Even without the added context of the absurd capabilities of Dudesy discussed on the podcast, LLMs and other kinds of AI can’t make coherent hour long comedy specials. You are welcome to try it. Train your own LLM on Carlin’s work. You aren’t going to get anything even resembling the output of human creativity.
Dudesy is supposedly the AI that created the George Carlin special. But if Dudesy is a character created by Kultgen and Sasso, and not an AI… who wrote the special? In a development that should shock no one, but probably does shock a lot of people, Kultgen is the author. From Vulture again:
The rep also noted that the Carlin special was in fact written by Kultgen himself, though they have apparently yet to clarify whether the synthetic Carlin voice heard on the special was produced by AI technology.
Why was I so convinced even before the suit and these admissions? I listened to the special. I’ve been listening to Kultgen’s comedy for long enough. The special was full of his familiar topics, gags, and jokes.
Society, as a whole, learns nothing. Knowledge, culture, and even consciousness of factual events are far too siloed and individuated for anyone to make declarations about shared comprehension. The “we” in any “we should” statement, with the implication that the “we” is a broader collective, is absurd. Nonetheless, we should take a lesson from the George Carlin AI affair1. The rapid ascendency of AI technology has inspired mass panic. It is overblown.
What the current state of LLM technology does is precisely the opposite of what it is perceived as doing. There are endless discussions about the dangers to academic writing, students using LLMs in writing classes, AI replacing TV writers. Though ChatGPT may be a menace in the first year composition class, currently available LLM technology is nowhere close to being able to reliably produce good writing of anything but the most schematic kind.
On the contrary, what the LLM can do now is close the gap between soft and hard skilled individuals. LLMs can flawlessly write various kinds of code, for instance, given the right instruction. They can reproduce the work of computer and data scientists and engineers far more consistently than they can the work of creative writers. The only people who should be afraid of it are the ones with computer science degrees or who paid for an expensive coding bootcamp.
As for the case against Dudesy, I’m still curious about the outcome. With minimal legal knowledge, my point of interest is whether the usage of Carlin’s voice reading Kultgen’s jokes constitutes any kinds of copyright infringement. The artificial voice, at least, likely involved AI software. The trajectory from the fanciful invention of the fictional Dudesy to the emergence of the LLM captures the essence of Kultgen’s appeal. He may have cooked up a ridiculously powerful fake AI as the conceit of a podcast, but he anticipated the direction of AI technology with astonishing foresight. I wonder if his attitude on cryogenic freezing will age equally well.
Weekly Reading List
https://www.forbes.com/sites/paultassi/2024/02/12/palworld-has-lost-two-thirds-of-its-players-in-two-weeks/ — I have quit playing Palworld. Not sure if I will ever pick it up again, although I enjoyed my time with it. It seems like I might not be the only one.
https://www.cnn.com/2023/02/11/us/he-gets-us-super-bowl-commercials-cec/index.html — I’m sure readers haven’t missed the “He Gets Us” ads that have pervaded television over the past couple of years. Here’s a primer on their backgrounds.
https://www.bookforum.com/print/3003/john-a-williams-s-unsung-novel-25303 — I just finished my dissertation chapter on the recently republished The Man Who Cried I Am. It’s funny to see a wave of press about a novel I have been studying for (I admit, not proudly) the better part of a decade. But, it is a good one. Seymour’s words are a fine starting point.
Until next time.
“We” won’t, “we” can’t. But any individual person can.
Also, if you are curious, this is the rewritten AI title I got from that stupid website:
AI still working hard to learn what “funny” is.