Appupen and Laurent Daudet’s work of fiction, which intersperses fact with fiction and academic papers with a cyberpunk plot, feels like a time capsule from the future
On page 26 of Appupen and Laurent Daudet’s graphic novel Dream Machine (Context), there’s a scene where the protagonist Hugo, co-founder of an AI startup, is talking to his wife Anna over dinner and they’re discussing the applications and potential dangers of AI-based systems. Hugo says, “I don’t like to think of the use of AI in war. Anyway, we don’t need AI to kill humanity. We do that very well ourselves.” This triple-decker stack of dialogue is deceptively straightforward and, in fact, contains multitudes.
The first line reveals that Hugo, like a lot of otherwise well-meaning scientists, is reluctant to think about the possible military applications of his research. The second line is, ironically, most often used by those who seek to downplay the potential dangers of AI. The second and third lines in conjunction amount to saying, ‘There’s nothing uniquely dangerous about AI’. Cumulatively, it presents a philosophical quandary — which is more dangerous, a ‘smart’ weapon in the hands of not-so-smart people, or the other way round?
At just over 150 pages, interspersing fact with fiction and academic papers with a cyberpunk plot, Dream Machine (originally published in French) is one of the most accomplished graphic novels of the last few years. Appupen is known for allegorical, largely wordless ‘funnysad’ books like Aspyrus, Legends of Halahala and The Snake and the Lotus (all set in the land of ‘Halahala’, a consumerist tech-dystopia).
At stake: the future of democracy and freedom
His versatile technique allows him to incorporate several different styles in these graphic novels — from the straightforwardly photo-realistic to the outright cartoonish and everything in between. In Dream Machine, Appupen and his co-writer Laurent Daudet (a professor of physics) have achieved something rare and special — a work of fiction that somehow feels like a time capsule from the future, even when it’s describing things that really happened (IRL, as the kids say) in the last 24 months or so.
The narrative follows Hugo, whose AI firm KLAI has made cutting-edge LLMs (language learning models) and is about to sell the tech to the all-conquering conglomerate called ‘REAL’, led by an Elon Musk-like CEO who claims that the tech will be used inside of a new virtual reality game.
Right from the beginning, Hugo and his colleagues suspect that there may be a more sinister motive behind the proposed acquisition. The rest of the book follows Hugo as he tries to make sense of REAL’s true intentions and what this might mean for the future of democracy and freedom (yes, the stakes are that high), especially since REAL is literally in business with dozens of governments around the world.
A primer on the basics of AI and LLMs
Dream Machine has plenty of longish stretches where the reader is basically being taught the basics of AI and LLMs. These expository bits are never dry or boring, though, because the way Appupen’s visual works with them. A scientist’s lecture about the history of innovation becomes the perfect pretext to draw Newton and Einstein in the same panel, the latter in his iconic tongue-out pose.
The moment an LLM starts ‘generating’ text after being fed trillions of data point is represented by giving the humaniform LLM-dummy a twisted, malevolent smile, as though this were a James Cameron villain and not a gigantic bean counter going through 1s and 0s at warp speed. The aforementioned scientist’s lecture also contains this very astute passage, which ends with an excellent question:
“AI models are opaque and it is sometimes completely unclear why a specific black box model works. The tools of our scientific method will keep evolving. But from Babylonian times and before, there has been one constant driver of mankind’s progress in science. That is the scientist’s intuition. I conclude my address with a question to you all. Are these powerful AI systems encouraging our intuition, or killing it?”
The psychological terror of the end-times
The ‘Super-Hugo’ comic strip that appears at the end of every chapter (and also kick-starts the book) is a particularly clever narrative device used in Dream Machine. From a comics point-of-view, it’s also a classical move, one that has been deployed by some of the greats of the medium.
In Alan Moore and Dave Gibbon’s ‘anti-superhero’ graphic novel Watchmen (1987), there’s a comic-within-the-comic called Tales of the Black Freighter, the story of a sailor under attack and driven to insanity by pirates while trying to return home. A little boy is shown reading Black Freighter while the city is being destroyed; the psychological terror of the end-times is amplified by this story of a mind under ferment.
Chris Ware did something similar in his book Jimmy Corrigan (2000), wherein the titular character is an unremarkable, socially awkward man nearing 40 years of age. In his visions, however, he is “Jimmy Corrigan, the Smartest Kid in the World”, a child prodigy. The smartest-kid storyline is Ware’s way of depicting the mental illness of this middle-aged character.
Similarly, Daudet and Appupen use the ‘Super-Hugo’ comics to depict the state of mind Hugo is in, usually as a response to a plot twist he’s dealing with. Some of the more optimistic, escapist strips happen towards the beginning of the story, when Hugo is still hopeful about the REAL.E deal. As the story progresses and the deal begins to look more and more like a Faustian bargain, the ‘Super-Hugo’ strips become moodier and their dream-logic takes a hard right turn into nightmare territory.
Intense close-up panels
The last chapter contains the book’s USP, a kind of dovetailing of form and content. You see, the last chapter was written with a little help from ChatGPT and Stable Diffusion (a text-to-image deep learning model). Appupen and Daudet prompted ChatGPT with five different ‘futures’ based on a synopsis of the book. Meanwhile, Stable Diffusion was fed with Appupen’s hand-drawn illustrations from the rest of the book. The result is a multi-trajectory Chapter 10. The Afterword in book clarifies:
“The illustrations were generated by the Stable Diffusion model, refined on over 500 drawings taken from the first six chapters and captioned by hand. The prompts that can be seen under the images were written by the authors based on the text. While these illustrations are far from perfect, they seem to capture some of the characteristics of Appupen’s graphic style in the book. A larger training set would help improve their finish.”
I’m not quite sure what to make of this, to be honest — it’s something that avant-garde artists would admire. In terms of methodology, this narrative choice is closer to performance art than anything books (comics/graphic novels or otherwise) have to offer. The art was much better than I expected, I admit. The text in this concluding chapter still falls in that uncanny valley where you cannot quite make out whether a human hand wrote this — I suppose that’s part of the point.
On the whole, though, I absolutely loved Dream Machine. The art reminded me of stone-cold masterpieces like Charles Burns’ Black Hole (the covers have a certain resemblance, especially in the way you can’t see the eyes) and Daniel Clowes’ Patience (Appupen’s Kubrickian ‘close-up’ panels are every bit as intense as Clowes’). The writing is fast-paced and does an impressive job breaking down the science for a generalist audience. As the Foreword says, “Human or machine, whoever you are, read this book!”
