The Man Machine
On a hot July afternoon, a friend sent a video, a TikTok livestream, to a group chat, followed by a question: “What is this genre called?” A young man in a Spider-Man suit and hoodie was rhythmically swaying back and forth in a parking garage as he stared into the camera. He frequently broke his silence when he received a “gift” sent by one of the stream’s 2,000 viewers. Depending on the gift type, he repeated one of a few phrases, each with the same delivery: “Hey, thanks for the rose”; “Too much ice cream makes me cold”; “Nothing like a New York hot dog.” If he received a new gift before he could finish his sentence, he cut his answer short and moved to the next gift’s phrase: “Hey, thanks… Hey, thanks… Hey, thanks for the rose.” When the gifts stopped coming, he went back to his idle animation, swaying back and forth.
I thought it was genius. “He is webcamming as a video game character, specifically a non-player character; I suggest the name NPC camming,” I wrote to the group. But while the video was eerie and dystopic at first impression (one viewer even commented, “This is not Miles Morales, this is Mike Myers”), the man’s rhythmic moves and imperfect efforts to efface human emotions and gestures were also charming and comforting. Soon after, I realized that many people shared my enthusiasm: This genre quickly became an internet hit and already had a name, though with a little less of a 2000s vibe than my proposed term. It’s called “NPC streaming,” and it originally became popular when TikTok user Pinkydoll’s videos went viral that same month. Hearing rumors that she made $7,000 per stream, many people started imitating her, adapting her methods into their own NPC characters. Most recently, Nicki Minaj did an NPC stream. We learn something new about ourselves every day, and it has turned out that watching other people act as AI agents is something we enjoy. Though not completely surprising, it definitely is curious.
Around the same time as the rise of NPC streaming, New York Times Op-Docs released Hardly Working, a film by artist collective Total Refusal that observes the daily routines of the NPCs in the game Red Dead Redemption 2. The filmmakers followed and recorded several NPCs—a carpenter strolling the docks, meticulously driving two nails into separate locations; a stablehand, even though seemingly drunk, always chopping wood right through the middle; a laundress tirelessly scrubbing garments for hours, even in the rain—over many game days. The film meticulously shows and describes these characters—just a fraction of the game’s over-1,000 NPCs—and finds beauty in their repetitive actions, which in turn points to the labor of the artists and programmers who spent thousands of hours to give these characters life. There is also a familiar sadness in their inscrutable interiority, the mismatch between them and the game world and their inability to influence the world or be influenced by it. Hardly Working leverages its observations for political critique; the film portrays these characters as people stuck in the infinite loop of capitalist labor practices, people whose time belongs not to them but to their employer. When they glitch, they stop working and are capable of standing idle for hours. The narrator of the film says that this is when they claim their time back and become subjects of potential change. He asks, “Can we start glitching?”
The NPC streamers of TikTok glitch, too, though in different ways. Repeating exact phrases and putting on a continuous idle animation of swaying is hard work; the performance requires constraint and stamina, as well as an environment without any distractions. Streamers often break character: when they need to attend to the needs of their kids, when a stranger walks into a frame, when they get tired or disheartened. Viewers wait for these moments, which are critical points in these endurance games but also offer revelations of the humans behind the streamers. After breaking character, a streamer shared her thoughts with the four viewers of her stream: “I tried to do it for money, but it is so hard to do it after one hour. I don’t understand how people do it for eight hours.”
As with Hardly Working, questions around labor are dominant in NPC streaming. Attempting to attract more viewers, many people imitate franchise characters. I came across a Deadpool impersonator who was patiently anticipating that his channel would eventually garner a larger audience than its current two viewers; two people in Scream masks looking swiftly left and right, searching for their victims; a young man wearing a pink ski mask dancing in front of a Barbie background; and several Grand Theft Auto characters in their white undershirts. A young woman said that her NPC goal was to pay her father’s $4,000 medical bills; a mother and daughter duo received as many gifts as insults as they swayed in unison. In contrast to Hardly Working’s NPCs, the streamers were not engaging in repetitive activity because of labor demands made by others. They were themselves inventing repetitive activity as labor to make money. Labor is not the source of repetitive activity; repetitive activity is the source of labor. In NPC streaming, the purpose of the labor is erased, leaving only its form.
New visual–sonic languages emerge and fade on social videos, video games, livestreams. This process simultaneously happens so rapidly and on so many fronts that it allows only an intuitive understanding and partial articulation, but a common thread is the logic of the digital seeping into our real-world lives. For example, for years we have been restructuring reality around the metaphor of the computer. Video game aesthetics are dominant. Computational explanations of the self (the brain is a computer) and the universe (reality is a simulation) are common. Thus, watching people act as simulated AI agents satisfies this model of the world. It provides us with a reflexive moment in which to contemplate this process, even unconsciously.
A month before the NPC streaming wave went viral, chip maker NVIDIA announced a new technology, “Avatar Cloud Engine” (ACE), and released a demo video. In the video, a player walks up to the chef in a ramen shop. Unlike conventional games relying on multiple-choice responses, the player here uses their microphone to pose an open-ended query, “How are you?” In poorly written, wooden NPC fashion, the chef replies, “I am worried about the crime around here. It’s gotten bad lately.” The voice is deep but lacks personality. As they pose more questions, the player gets more answers from the chef, and the implications of this new technology become clear. With a boost from generative AI, an NPC can stop being a definitively pre-programmed character with a limited number of responses. The NPC can start to have its own unique mannerisms and even “thoughts,” akin to the “hallucinations” currently found in ChatGPT conversations. According to NVIDIA, using pre-trained generative AI language models, ACE provides tools to “develop fully autonomous, intelligent avatars.”
Well, I don’t know if I need to say it out loud, but here is the obvious irony: As AI-driven video game characters are becoming more complicated, layered and unique, humans are imitating simple AI avatars, and viewers are demanding it. Of course, NPC streaming is just a performance; humans have not turned into NPCs. However, the NPC streaming meme can serve as an early sign of future shapes the labor market and even interpersonal relations may take. Over the past few decades, the public space of social media has already become more accessible and richer than the public spaces of most cities. Will the same thing happen to people? Will the AI NPCs be more genuine and authentic than real people? Will they compete for our intimacy?
Jonathan Crary, author of seminal art history book Techniques of the Observer, argues in his latest book, Scorched Earth, that the pervasiveness of the internet and AI technologies creates a loss of interiority, a dispossession of thought. NPC streaming is almost a ritual dance for that dispossession. I see similar patterns in myself and also in my peers, a sort of erasure of social techniques. For example, more and more, I tend to not remember exact information but rather the method to access it: “If I Google this term, then I will find it in this website.” This type of memory outsourcing extends the capabilities of the mind but also removes access to one’s own thoughts and memories. Occasionally, I find that I’d need a device to look up things to make my point in a conversation. The digital shadow of myself is slowly taking over the real me.
As I research AI agents and watch real people LARPing NPCs on my feeds, I think of Kiyoshi Kurosawa’s 2001 horror film Pulse. The internet opens a door to another world. Through this opening, ghosts start to take over the world and the real people disappear. A one-way stream that connects both worlds.