Sexy AI Chatbots Are Creating Thorny Issues for Fandom

But despite the removal of what many feel to be both a core capability and function of any internet chatbot, large numbers of people continue to talk to the “characters” of Character.AI—a term the platform uses loosely, even encompassing things like AI assistants, which answer queries just as ChatGPT might, but with humanoid names and faces. There’s extensive guidance for character creation—essentially teaching users to do the work of training bots themselves—and the terms of service makes it clear that everything on both the training side and the chatting side is the intellectual property of those who input it, leaving the platform itself as a mere middleman, though not a particularly transparent one.

Even if Character.AI might want you to get emotionally attached to its coding bots (your fellow “pair programmer”) or its grammar bots (your “English teacher”), it’s the characters you’ve heard of, real or fictional, that have sparked the most interest across the social web. “Billie Eilish” currently has six times the amount of interaction of “Joe Biden”; both of them eclipse “Alan Turing.” “Remember: Everything Characters say is made up!” reads a cheerful message atop every chat, and which evokes memories of Historical Figures, the supposedly -educational app that went viral earlier this year when users’ chats with, well, historical figures spit out utter nonsense (and not even interesting nonsense).

But the app’s fictional characters have also garnered a fair amount of attention from fandom, where the idea of chatting with your actual favorite character might hold more affective appeal than chatting with a fake English teacher. The #characterai tag on Tumblr is awash with screenshots from the platform, many of them also tagged “self-insert” or “x reader,” a subgenre of fan fiction in which you engage with known characters (often—but not always—romantically and/or sexually) via the second-person narration of an unnamed “reader,” sometimes written as Y/N, or “your name.”

X reader fic is regularly invoked in discussions of Character.AI and fandom, as is chat-based roleplaying, which fans have been engaging in for decades. But these parallels only resemble what’s happening here on the surface—and for fandom, Character.AI is already proving a complex, sometimes thorny space, from fans’ relationships with the companies that own the characters to fandom’s wide range of opinions about AI to what it means to directly interact with a character you love.

“Chatbots have existed in the context of fandom for the past 10 years, and gained more traction around five years ago,” says Nicolle Lamerichs, a senior lecturer in creative business at the University of Applied Sciences, Utrecht. “Often these chatbots were initiated by companies to market to fans specifically, and allow for more interaction with their brand.” Most of these pre-programmed bots offered a limited number of responses and interactions, like Disney’s Facebook Messenger–based Zootopia chatbot, or Marvel’s Conversable, also via Facebook as well as X (previously known as Twitter), which let you DM Marvel characters. But the rise of generative AI has utterly altered the top-down, corporate-sanctioned way fans were previously able to chat with characters. “These tools have become democratized,” Lamerichs says. “This is leading to new types of fanworks and fan interaction, which is very interesting to observe.”

This democratizing element opens up complicated questions about copyright and AI, but right now, like most questions about copyright and AI, there are no clear answers. “We’re still very much in the vocabulary-building phase,” says Meredith Rose, senior policy counsel at Public Knowledge, a consumer advocacy organization that focuses on tech issues. “You have copyright specialists who now have to learn specifically about the tech that underlies this stuff—and because things like fair use determinations, which are crucial to AI discussions, are very, very fact-specific, you have copyright experts who need to understand all the intermediate steps that go on under the hood in a generative AI platform, and that kind of learning takes a lot of time.”