On your typical reality TV show, the digital world exists only to host the heated conversations of viewers. Contestants on the Bachelor franchise, for instance, have virtually no access to the phones or the internet during filming. Netflix’s new reality show, The Circle, turns this model on its head: The show’s action happens exclusively through social media.
The contestants, who are holed up in apartments meticulously decorated to fit their characters’ personalities, are not allowed to interact face-to-face. (This leaves room for contestants to compete as someone else, catfishing their fellow contestants.) Instead, they communicate through the Circle, a “voice activated social media platform” that appears on their apartment’s multiple TV screens. The Circle includes a scaled-back version of all the features you’d expect to find on social media—profile pictures, bios, a newsfeed, and chat. (Warning: potential spoilers ahead!) Contestants rank one another based on these barebones online profiles and interactions, and the last player standing wins $100,000.
The limited abilities of the network keep the producers in control, allowing them to nudge the show’s plot forward. For instance, in the first episode, contestants are allowed to choose just one photo from existing albums to “post” as their profile picture. As they make their selection, they talk the cameras through their options, and their decision-making process. Antonio, a pro basketball player, is afraid that using a shirtless photo will lead other guys to think he’s cocky. “First impressions are everything,” says Karen, who’s catfishing other players as “Mercedeze” by using photos of someone else. To flip through Mercedeze’s photos, she talks to the social media platform using the same stilted tone we use when commanding Siri or Alexa: “Circle, enlarge the photo in the blue overalls.” Magically, the Circle displays the correct photo of Karen’s alter-ego.
This was the moment I began wondering what the Circle itself is, and how it works. While it’s now commonplace to talk to digital assistants, what those assistants can actually do is still quite limited, in large part because computers can’t understand speech and context the way humans do. Most humans would have no problem picking out the photo of Mercedeze in blue overalls, even though it only shows Mercedeze from the collarbone up. but for a computer, this task could get tricky: What are overalls? How can you tell that she’s wearing overalls from just a blue pattern on a shirt?
As the show goes on, there’s no shortage of moments that would baffle the average computer. Contestants command the Circle to send emojis of all sorts (most often a heart emoji); when I ask Siri to compose a text with a heart emoji, it simply writes out “heart emoji.” Whereas a human would rarely make this mistake, a computer regularly does, because there’s no way for it to understand that I want it to go into the emojis and search for one that fits the description I’ve given instead of just writing what I say. And on The Circle, the computer magically reads contestants’ minds over and over. While contestants move between talking to the camera and composing messages to send on the Circle, the computer seems to know exactly when to start transcribing.
There are two possibilities here. The first is that these reality TV producers have built an ultra-smart technology, sensitive to the nuances of human speech in a way that no other major voice-activated assistant on the market is capable of. Given the imagery in the show—wires and circuit boards appear when the Circle is introduced—the producers clearly want us to believe technology is at work here. But the more likely possibility is that the “voice-activated platform” is actually just a bunch of humans furiously typing behind the scenes, transcribing contestants’ speech and sending messages to their screens. To find out more about this “voice-activated” system, I reached out to Netflix and Studio Lambert, the production company behind the Circle, but neither responded to my request for comment.
If the Circle actually is just a bunch of “voice-activated” TV producers, it wouldn’t be the first time human work was passed off as a sophisticated computer. In 2018, investigative journalist David Farrier detailed his discovery that a digital assistant called Zach was actually just one guy pretending to be an AI “transcribing” medical notes. Tech start-up X.ai promised users an AI personal assistant, but in reality, human “trainers” did much of the heavy lifting. Other “AI” services like email scheduling bot Clara and Facebook’s now-defunct personal assistant M also relied on humans to monitor interactions. Machine-learning expert Janelle Shane has even written for Future Tense about how to tell when a human is pretending to be a bot. Building a fully autonomous, smart AI that can understand and respond to human speech is a big ask. When it comes to something like an edited reality TV show like the Circle, the Occam’s razor explanation for the Circle’s impressive natural language processing is that it’s powered by a back room of fast typists sending texts between contestants’ screens, rather than a custom-built AI system.
If the Circle is human-run, it adds new dimensions to the show. Over the course of several episodes, I documented inconsistencies in how the Circle transcribes: Sometimes, contestants will include tips on punctuation, like ending a statement with “question mark” or “period,” but often, the Circle will just add in what seems right, even without those commands. It takes other liberties, too. Some contestants’ messages appear in lowercase or include ellipses, while others’ don’t, and when transcribing longer statements, the Circle will sometimes send the whole thing as one long, punctuated message, and other times it will send multiple shorter phrases without punctuation.
These tiny cues may seem insignificant, but they play a major role in how people—especially those who text often—perceive messages. In internet linguist Gretchen McCulloch’s recent book Because Internet, she spends a chapter explaining how the nuances of capitalization, punctuation, and line breaks in online spaces convey what she calls a “typographical tone of voice.” While sending long messages with multiple punctuated sentences might feel overly formal and stiff, sending multiple messages can feel more conversational, since it gives the other person more of a space to respond. After all, if you were beginning a conversation with someone in person, you’d be unlikely to launch directly into a rehearsed monologue right after “hello.” A lack of punctuation also conveys an informal tone, while repeated letters (Miranda writes that she is “verrrryyy” single) and all caps (“GOD knows my heart BUT the Devil knows my vocabulary!” reads Mercedeze’s profile bio) can convey excitement or emphasis. Yet, “the Circle” just inserts these without contestants explicitly commanding it to do so—and captures how we use text to express emotion in this particular cultural moment.
It might also provide a glimpse into how our relationship with voice-activated technology might change over time, if AI does become smart enough to be ever-present in our homes the way the Circle is in contestants’ apartments. (Think Smart House, but with a better ending.) At the beginning of the show, contestants speak awkwardly to the Circle; in the first few scenes, Karen (aka Mercedeze) constantly addresses “Circle” in every command, and even tells it “good job” after it successfully executes her command. By Episode 4, contestants are zooming through their commands—“message” has become shorthand for the beginning of a text, while “send” serves as the sign-off—and use a more natural conversational tone when dictating their messages, as if they’re actually talking to another human directly.
Regardless of what the Circle really is, the show is a horrifying, addictive look at our online habits, and how difficult it is to escape our impulses to manage others’ impressions of us and judge others, even when we know it’s all a one-dimensional façade. And that’s something we’ll surely keep doing, with or without “voice-activated” AI.
Future Tense is a partnership of Slate, New America, and Arizona State University that examines emerging technologies, public policy, and society.