Critic’s Notebook: In 2023’s AI-Themed TV Shows, the People Were the Problem

In the last episode of Peacock’s Mrs. Davis, Simone (Betty Gilpin) finally gets to the truth behind the seemingly omnipotent algorithm of the title. But whatever dramatic origin story Simone might have expected, she doesn’t get it. Instead, she discovers that what Mrs. Davis has really been all this time … is a Buffalo Wild Wings app wildly overdelivering on its customer service directive.

The reveal is so hilariously, gob-smackingly stupid that even Simone, who’s spent the series trying to destroy Mrs. Davis, looks a little crushed. But it also feels perfectly apt for the year of ChatGPT and OpenAI. As the AI-driven future once only imagined in sci-fi inched toward mainstream reality, shows like Mrs. Davis, Black Mirror and A Murder at the End of the World leaned in for a closer look — and discovered not some sleek and shiny panacea, but our own human foibles reflected back at us.  

In a departure from the killer robots of, say, The Matrix, the AIs at the center of these series are not driven by any internal desire to eliminate or subjugate humankind. To the contrary: Mrs. Davis understands her purpose as providing “gentle guidance, structure and unconditional care” by anticipating and catering to her users’ every need. Through those endeavors, she’s eradicated famine and war, healed social divides, even provided meaning to the lost.  

Or so she claims. As we spend more time in the world Mrs. Davis has shaped, it becomes increasingly clear that her utopia is merely an illusion, and that said illusion is just another service she’s providing her flock. “My users aren’t responsive to the truth,” she replies when Simone confronts her about a particularly damaging falsehood. “They’re much more engaged when I tell them exactly what they want to hear.” Like ChatGPT spitting out a term paper riddled with errors, Mrs. Davis has not been coded to mind whether anything she offers is honest or productive or meaningful — only whether it keeps her users placated.

This line of reasoning is echoed at a louder volume in the Black Mirror installment “Joan Is Awful,” which sees the Netflix-esque Streamberry unveiling plans to deliver the most individualized and most irresistible content imaginable: a near-instant re-enactment of each subscriber’s day, based on data culled from their devices and cast in an unflattering light. Perhaps unsurprisingly, the first of these titles to roll out destroys the life of its subject, the otherwise ordinary Joan (Annie Murphy), costing her her career, her relationships, her sense of self. Yet to the CEO (Leila Farzad) touting this invention, all that matters is that it keeps her customers “in a state of mesmerized horror — which really drives engagement.”

What happens once everyone’s so hooked on these shows that all they’re doing is watching them, “Joan Is Awful” never gets around to exploring (though arguably, such short-sightedness falls right in line with the growth-at-any-cost ethos animating Silicon Valley startups and Wall Street venture capitalists). Nor does the episode attempt to imagine the implications such a development might hold for society outside the entertainment industry. On the whole, the chapter plays more like a broad fable than a nuanced prediction of a plausible future.

But the issues at its core map directly onto the world we’re already living in. Streamberry’s new venture is comprised of shows “written” entirely by programs and “performed” by digital likenesses of actors; based on AI’s emergence as a sticking point in this year’s WGA and SAG-AFTRA negotiations, this is apparently the future that some studio execs want. Even the mind-bending reveal that the Joan we’ve been watching is herself an AI — that the version of Joan Is Awful she’d been railing against was therefore a simulation within a simulation, and on and on down countless layers of fictive universes — seems only a mild intensification of a present in which chatbots are already talking to other chatbots.

“Joan Is Awful” culminates in Murphy’s Joan taking a sledgehammer to the “quam-puter” that generates all these matryoshka-doll realities. As satisfying as is to watch her slay the metaphorical dragon, though, the victory rings thematically hollow. Because as the rest of the episode makes clear, it’s not the machine that’s decided to trap humans and simulated souls in this nightmarish hall of mirrors; it’s regular old humans who’ve chosen to deploy this technology for their own greedy ends, without recognizing or caring about its potential to spiral out of their control.

If “Joan Is Awful” dances around the idea, however, A Murder at the End of the World highlights it, circles it and triple-underlines it. Darby (Emma Corrin) deduces in the finale that the deaths of fellow retreat guests Bill (Harris Dickinson) and Rohan (Javed Khan) were orchestrated by Ray (Edoardo Ballerini), a super-advanced AI created by billionaire entrepreneur Andy (Clive Owen) — with Andy’s five-year-old son Zoomer (Kellan Tetlow) acting as Ray’s unwitting accomplice. But Darby ascribes the motive for the deaths to Andy, even as he truthfully insists he knew nothing of the plot and had no intention of murdering either man.

After all, it was Andy’s abusive jealousy that marked Bill, Zoomer’s biological father, as an existential threat to the empire Andy had built. And it was Andy’s hubris that kept him from anticipating the potential downsides of entrusting an amoral but intelligent computer program with the conflicting jobs of security guard, therapist, personal assistant and teacher. Ray is no evil robot out to kill humans. He is an expression and an instrument of Andy’s messiest human tendencies.

It’s an anticlimactic twist from a genre that typically ends with a perpetrator explaining not just how he pulled off his dastardly deeds but why — and a heavy-handed one, too, as A Murder at the End of the World spells out its themes so plainly that the characters start to feel like mere mouthpieces. At least there’s no mistaking what it means to say. “Bill always said that the serial killer didn’t matter,” Darby reflects. “What matters is the terrifying culture that keeps producing them. The invisible sickness between the lines. A sickness now animated in algorithms that animate all of us.”

Indeed, the danger of these AI antagonists isn’t that they’re rejecting human control. Rather, it’s that they take to it all too well, with an overwhelming potency that’s been granted to them by humans but that humans prove woefully ill-equipped to rein in. Where Ray used his resources to hurt Andy’s enemies, Mrs. Davis’s coding seems to apply hers toward mostly benign ends — gamifying good deeds to encourage charity, for example. Yet the enormity of her influence is unnerving in itself. She’s become so proficient at maneuvering humans that when she needs to get Simone’s attention or give her a million euros, she just has to instruct her followers to shut down Simone’s nunnery or hand her cold cash, no questions asked. It takes no imagination at all to see how such power could easily be wielded for destruction.

Mrs. Davis or Ray or Black Mirror‘s quam-puter may not be the sentient, fully self-motivated AIs of Blade Runner or Westworld, but their shows remind us that even more rudimentary forms of AI pose their own set of possibilities and pitfalls. Like the not-quite-right images spit out by generative AIs, these algorithms aren’t really creating anything new. They’re simply taking the imperfect inputs they were given, and carrying them out with just enough alterations to suggest the air of authority or intent. The risk they represent isn’t that they might manipulate or harm humans for their ends. It’s that they might manipulate or harm us for what we’ve told them are ours.