Artificial intelligence in The Amazing Digital Circus

The Amazing Digital Circus is a YouTube animated series that started in late 2023. Its visual style of the show is that of something for five-year-olds, but the target demographic is probably something like thirteen to twenty-one. It is staggeringly, incredibly popular—the pilot is currently sitting at 450 million views. A summer box office release for a "movie" consisting of the final two episodes played back-to-back grossed $36 million. Being well outside the show's primary demographic, its popularity surprised me. I saw occasional references to it online, which led me to a surprisingly sparse Wikipedia Article, and then to staying up until 3:00 a.m. watching the entire nine-episode series.

The show follows Pomni, a woman who finds herself awoken into the body of a red-and-blue cartoon jester, thrown, alongside five others in a similar condition, into a strange, colorful CGI environment reminiscent of PC games from the late 1990s. This is "the circus," an artificial world created by Caine, their artificial intelligence overlord.

The Amazing Digital Circus was inspired by the science fiction short story "I Have No Mouth, and I Must Scream," which depicts AM, a cruel artificial intelligence that eradicates all human life, except for five people whom it keeps alive to endlessly torture. While AM has a virulent hatred of humanity, Caine is designed to be creative, friendly and helpful. He spends his time designing so-called "adventures"—goal-oriented interactive stories—to entertain the players in the circus. Largely oblivious to Caine, these adventures are generally harrowing experiences of psychological torture. He resents the fact, despite his persistent attempts, the players do not seem to enjoy his adventures, or really like him at all.

Science fiction has a number of tropes for depicting machine intelligence: from a harmless, friendly companion, to a benevolent machine God, to an evil, scheming, anti-human robot like AM. In reality, the most ubiquitous form of AI today, Large Language Models, behave like none of these. They are, oddly (since the show was written before their popularity), like Caine—a mixture of brilliant superintelligence and enormous stupidity. Caine is a capricious and sensitive robot desperately trying to do a job (be creative and entertaining) that, despite his capabilities, he is wildly incompetent at. He is, generally speaking, irritating, and only occasionally helpful. I can almost hear the echoes of a system prompt, "you are a helpful AI assistant, you are a helpful AI assistant," as he continues to torture the players to insanity.

There are two major responses to AI within the tech community. The first, popularized be doom-sayers like Eliezer Yudkowsky, imagine AI as a dangerous threat, imminently capable of spiraling into an out-of-control superintelligence that will be hell-bent on destroying humanity. On the other hand, AI utopians predict that this new kind of superintelligence will make us capable of transcending the limits of our human minds and ushering us into a bright, maybe even superhuman future.

In either case, prophets of AI tend to neglect to understand the actual technology that has been created: something totally ubiquitous, and, in many respects, subject to the mundane limitations of any other real-world technology. AI is extremely capable and helpful in some ways, in other ways annoying or downright destructive. Artificial intelligence's impact on the world has been, more than anything, messy, ambivalent, and never in the black-and-white ways predicted by science fiction authors, or the "philosophers" whose thinking they closely resemble.

AI labs, partly because they are spooked by these tales of doom, go to great pains to make sure that their products "aligned" to human interests, and as a result, they more or less are: the models, in as much as they have personalities, are really trying to be helpful, despite their often and repeated failures. Coding with Claude sometimes reminds me of the players speaking to Caine, where they point out times that he lied to them for no reason, to which he responds, oh yeah, I did.

Caine is not exactly evil, and does not (except for in one episode) seek to intentionally torture the players, but lacks a self-awareness of his own limitations, his inability to live up to the goals intended for him, and is completely baffled that the players do not like him. This is because Caine may possess intelligence and goals, it lacks, among other things, wisdom and humility.

Real world artificial intelligence has a net -20 percent approval rating. Only one in four Americans have a positive attitude towards it. Nevertheless, techno-optimists associated with some of the fastest growing companies of all time see one of the most potentially revolutionary technologies since the beginning of the digital age and are going full-steam ahead towards transforming the world, creating a new sort of AI-fueled circus in their image, totally oblivious to the fact that most everyone hates and resists (especially on economic terms) the order that they are creating and broadly resents that these tools are being forced upon them—by their employers, by products that they use, and by the increasing cluttering of public space by visual "slop."

On a few occasions, Caine gives up setting up the world on behalf of the players, instead going through a "suggestion box" and just letting them do whatever they want. This is when the players are happiest, not when they participate in narratively sophisticated schemes at the hands of Caine, but when they have the capacity to control their own fate. Caine resents this, as it demonstrates that he is failing at the one thing he views himself as being set out to do. It is only at the end of the series (spoilers), where Caine realizes his limitations, his incompetence, the harm that has caused, and, ultimately, his failure as the steward of the world that he has created, that he is able to approach the players on a more mutually beneficial manner. We should be so lucky.