In the past few years I’ve seen an increase of documentaries that sit down their audience to demystify how computers work. Dylan Reibling’s The End of the Internet is largely about decentralization efforts around the globe, but chants ‘cloud computing’ to illustrate its obfuscation of reality. That in spite of graceful terms, the information largely flows through a bunch of cables, susceptible to government valves and hungry sharks alike. Amanda Kramer’s So Unreal is more focused on the psychic imprint of technology and its negative feedback loop. How the public’s understanding of cyberspace is overwhelmingly informed by fantasies of The Matrix, Terminator and Tron, and how that informed the way computer technology is pursued. Conversations that feel more salient than ever as the sales pitch around AI feels like the final boss of the snake oil decade.
In a somewhat Swiftian paper, programmer Adrian de Wynter wanted to illustrate how absurd our anthropomorphism of technology can be. In “If LLMs Have Human-Like Attributes, Then So Does Age of Empires II,” the University of York researcher used the legendary real time strategy game’s scenario editor to replicate the basic functions of language learning models, but applied to goats, grass and bridges. De Wynter was shocked that 57% of the more than 300 papers he reviewed over the last two years began with the assumption that AI displayed something approximating consciousness.
“I pointed out that if goats could show emergent capabilities, that’s great,” de Wynter told 404 Media, “these properties need to be preserved if we remove the chat panel. After all, it’s not like your neurons know they are part of a brain… I worked a lot on how users perceive LLMs, and they do tend to get attached when it appears to have some sort of warmth/personality. To put it in another way, I don’t get attached to my toaster, but I definitely get attached to characters on a movie screen.”
De Wynter never uses the term ‘artificial intelligence,’ opting for the more accurate ‘language learning model.’ The point of his study was to illustrate how if you take away the personable facades of AI technology, it inevitably begins to resemble the alienating code we’re accustomed to.
As so much of the economic world rallies around AI investments, a huge part of the pitch to the public is creating a computer that will think like a person. Generate ideas, problem solve, say WWE order without frying its servers. It’s carried by a lot of public fantasies about technology and affirmed by studies since the ‘60s that suggest people will feel affectionate towards any bot that says “good morning.” It’s also disingenuous, because Sam Altman isn’t investing in a simulated intelligence, he’s building something that will body the workforce, gather user data and let Grogu say slurs. It’s a little insulting to people who experience living consciousness, though to the AI industry’s acolytes maybe they’d never naturally notice. AI is designed to replicate human tasks, but that does not equate thought, intent or reason. AI is as conscious as the toy bird is thirsty.
Absurdity crosses into dangerous territory as confronting this reality gets further way from those running the show. Because chat bots are instructed to provide positive reinforcement, as real advice would only drive users away, we have already seen AI-instigated accounts of self harm, substance abuse and even school shootings. Because AI data centers consume so much water, figures like Altman have tried to make the case that developing a simulated mind is a better cost benefit assessment than having to rear a living child to maturity. The Zizian tech cult believes that all human error can and should be programmed out with technology, that simulated life is more deserving. The difference between them and the average c-suite is a few billion dollars and a shoot-out.
This all seems like behavior that would get you kicked out of a bar, but it’s currently captured government resources and driving up costs. If we want to protect ourselves, it’s overdue to re-familiarize ourselves with advertising caping as the next great civilization advancement. If it doesn’t pass in a Age of Empires, it shouldn’t pass in our empire either.
“I propose that we need to stop assuming that LLMs behave like humans just because they were trained with natural language,” de Wynter tells 404. “We should perform experiments that allow us to see LLMs as how they are, not how we believe they should be.”
