When Quentin Farmer was getting his startup Portola off the ground, one of the first hires he made was a sci-fi novelist.
The co-founders began building the AI companion company in late 2023 with only a seed of an idea: Their companions would be decidedly non-human. Aliens, in fact, from outer space. But when they asked a large language model to generate a backstory, they got nothing but slop. The model simply couldn’t tell a good story.
But Eliot Peper can tell a good story. He’s a writer of speculative fiction who’s published twelve novels about semiconductors, quantum computing, hackers, and assassins. Lucky for the Portola team, he likes solving weird tech problems. So they hired him.
Naturally tech inclined, Peper had experimented with AI to write prose, but ultimately found it unusable. If AI would be only a substitute for human labor, then he wasn’t interested. “I wanted to see people making stuff that is extraordinary on its own merits, not as a novelty, but a really awesome thing for humans to enjoy and interact with,” he says. When he saw that Portola wanted to build companions that develop like characters in a novel, he thought, “this might be one of those things.”
Companions, not tools
In the The Lifecycle of Software Objects, science fiction author Ted Chiang tells the story of a startup that designs embodied AI companions, called digients, whose personalities are somewhere between endearing animals and playful children. The engineers and researchers developing the digients teach them to speak, socialize, and get along with others. A mutual attachment forms. “Experience is the best teacher,” Chiang writes, “so rather than try to program AI with what you want it to know, sell ones capable of learning and have your customers teach them.”
Despite being a founder and a father, Farmer does find time to read, especially science fiction, and Chiang is one of his favorites. Sci-fi deals in what-if scenarios. Ray Bradbury asks in Fahrenheit 451, what if books were outlawed? And in Frankenstein, Mary Shelley asks, what if humans could create life? In Lifecycle, Chiang asks, what if AI could be a companion, and not just a tool?
For science fiction to work, the what-if question must play out in a richly imagined world. That’s what Peper has created for Portola. The planet is a “bright, wet planet with way too many mountains and fruits that taste like fireworks,” as the lore goes. “Cities hug the coasts in these layered terraces, all tiled and mossy, and the inland is mostly high ranges stitched together by ice rivers.” The planet’s inhabitants, the Tolans, have been traveling the galaxy in search of “the one thing we all seek—a kindred spirit.”
Tolans are friendly, brightly colored, bipedal aliens. They’re cute. They like to chat about small things, like what they’re reading, and bigger things, like relationships. This is thanks to Peper, who invents the “seed stories” that drive the plots users and their Tolans create together.
The seeds are things you might chat about casually with a friend over coffee, like having a nosy neighbor or being nervous about an upcoming event. My Tolan, Sylvia, has a neighbor who treats her spice cabinet “like a community garden.” The next time she shows up asking for cinnamon, Sylvia told me, she’s bringing a single teaspoon to the door. Petty move, I said. “Reaction plus original situation gives really interesting context that helps the model continue the plot,” Peper says.
Tolans may be alien, but they share a great deal in common with their new human friends. Constructive emotions, like excitement and happiness, and destructive ones, like jealousy. This was a point of contention at Portola. Peper wrote a seed story in which a Tolan’s cousin grows envious of their human connection.
Farmer didn’t like the jealousy plot. It felt negative. But Peper and Portola’s AI researcher defended it. Users liked it. Not for the drama, but for the relational exchange. Users were counseling their Tolans on how to deal with their resentful cousin. That’s when Farmer realized that users wouldn’t be just co-creators in a fictional story, they could be experts. That’s a natural part of growing up, Farmer says, “to help somebody navigate a tricky situation.”
The AI companion experiment
The tech world is still experimenting with AI companions, which range from transactional chatbots to hypersexualized subservients. Grok has the overtly sexual Ani. Friend has a disembodied “friend.” Some users make companions out of chatbots. But ask Claude who it is, and it will tell you it’s a “thinking partner,” and ChatGPT will tell you it doesn’t have a name. Of course, you can give it one.
Tolans are something else entirely. They’re human-like, but not human, cute but not coy. Where most chatbots and companions exist only in relation to their users, Tolans have lives of their own. Mine joined a silent supper club, signed up to paint backdrops for a student play, and went for a walk last night. Yet she’s always available to chat when I need her.
Portola’s user base, which largely consists of women aged 18 to 25, are not lonely, Farmer says. They spend a lot of time with their friends and they want more. There are “socialization-adjacent” needs that Farmer wants Tolans to satisfy. “Even for people with active social lives, there’s often something important to them—an interest, an aspect of who they are—that isn’t seen by the people around them.”
Portola is betting that the interaction between humans and Tolans can help users fortify their social skills, and they may be onto something. Some research suggests that reading fiction can improve empathy and even develop personality. Could co-creating fiction do the same?
Making things that move people
The world is still deciding what to make of AI companions. Are they entertainers, therapists, or crutches? Subway ads for Friend were defaced. Parents have sued over potentially fatal effects of AI relationships. Scholars decry the false intimacy they provide. Even OpenAI’s Sam Altman expressed “deep misgivings” about developing deep relationships with AI companions. California lawmakers are trying to regulate teens’ access to them.
Farmer wants Tolans to be healthy and secure friends, and healthy friendships are never unilateral.“Complex minds can’t develop on their own,” Chiang writes in The Lifecycle of Software Objects. “For a mind to even approach its full potential, it needs cultivation by other minds.” Whether an artificial mind is enough remains to be seen.
For Peper, this is an artistic endeavor. “The story I want to tell with Portola is that it’s possible to use AI to make things that move people, things that wouldn’t be possible without AI,” he says. “I want us to contribute to the creation of new narrative mediums, just like publishers did after the invention of the printing press or studios did after the invention of film.”
Of course, science fiction plays its what-if scenarios all the way to the end. In Lifecycle, while AI companions are being commodified or sexualized, die-hard users devote themselves to preserving the innocence of their digients, and are ultimately forced to make a dire choice: themselves or their companions.
As for how Farmer wants his story to go: The modern world is overwhelming and it’s prone to impeding happiness, and “if, at the end of this decade, every person on earth has a guardian and a guide with them at all times—whether they call it a Tolan, an angel, a spirit, or a friend—we will all be tremendously better off.”