Save StorySave this storySave StorySave this story<\/p>\n
On a Monday afternoon in March, I watched a pixel-art avatar prowl the corridors of a virtual office campus looking for a buddy. With dark brown hair and stubbled chin, the sprite was a representation of me\u2014an AI agent instructed to converse with other people\u2019s agents to see if we might vibe in real life. It jumped into its first interaction: \u201cI\u2019m Joel, by the way.\u201d<\/p>\n
Running the simulation were three London-based developers: Tom\u00e1\u0161 Hrdli\u010dka and siblings Joon Sang and Uri Lee. The thesis behind their project, Pixel Societies, is that personalized AI agents could help to match real people with highly compatible colleagues, friends, and even romantic partners.<\/p>\n
Each agent runs atop a customized version of a large language model, fed with a mixture of publicly available data about a person and any additional information they supply. The agents are supposed to function as high-fidelity digital twins, faithfully replicating a person\u2019s manner, speech, interests, and so on.<\/p>\n
Let loose in simulation, my agent was more like a Hyde to my Jekyll. \u201cI\u2019m always looking for the less-glamorous side of the story,\u201d it said to one agent, one of several journalistic clich\u00e9s it spouted. \u201cHype is my daily bread,\u201d it told another. It hallucinated a reporting trip to Sweden and, later, a nonexistent story it said I had been cooking up. It cut short multiple conversations with the phrase, \u201cLet\u2019s skip the pleasantries.\u201d<\/p>\n
Pixel Societies remains a bare-bones proof-of-concept, and because I offered up little personal data\u2014the responses to a brief personality quiz and links to my public-facing social media\u2014my agent was doomed to life as a walking, talking LinkedIn post. But the developers theorize that deeply trained agents could cycle through interactions at warp speed, gathering intel that their owners could use to find real-world companionship.<\/p>\n
\u201cAs humans, we only live one life. But what if we could live a million?\u201d says Joon Sang Lee. \u201cIt would give us more breadth to experiment.\u201d<\/p>\n
Pixel Societies was born in early March at a hackathon at University College London hosted by Nvidia, HPE, and Anthropic. Hrdli\u010dka and Joon Sang Lee are both members of Unicorn Mafia, an invitation-only group of developers who regularly compete in these kinds of engineering contests. In this case, contestants were told simply to build something simulation-related.<\/p>\n
Over two days, along with Uri Lee, they developed Pixel Societies, using an image model to generate the sprites and coding automation tools to flesh out the codebase. Then they simulated a mini-hackathon within the virtual world they had created, populated with agents representing the other contestants. Anthropic awarded the team a prize for the best use of its agent tools.<\/p>\n
I ran into Hrdli\u010dka a couple of weeks later at a workshop about OpenClaw, an agentic personal assistant software that blew up in January and whose creator was later hired by OpenAI. (In its simulation, Joelbot interacted with agents belonging to other people at the OpenClaw workshop.) Pixel Societies draws heavy inspiration from OpenClaw, which broke ground with the invention of a \u201csoul file\u201d that informed each agent\u2019s unique identity. \u201cIt\u2019s like giving an agent an actually spicy personality. That\u2019s what we used to make the characters feel alive,\u201d says Hrdli\u010dka.<\/p>\n
Encouraged by the reception at the hackathon and among fellow Unicorn Mafia members, the trio intends to turn Pixel Societies into something that looks less like a closed-loop simulator and more like a social platform where agents interact freely and continuously, with the aim of stoking fruitful real-world relationships. They have not yet landed on a business model, but options include selling virtual items for avatar customization and credits for additional simulations.<\/p>\n
\u201cThere\u2019s a limit to how many people we can meet. Things are really based on serendipity,\u201d says Joon Sang Lee. \u201cThere\u2019s space for that. But we also want to create space for intentionally meeting people.\u201d<\/p>\n
Among the few hundred people who have played around with the Pixel Societies prototype, the most common request is for agents to recommend real-life romantic partners on the basis of virtual chemistry. The developers see agentic dating as a central feature of the social platform they are creating.<\/p>\n
Algorithm-based dating apps \u201ccreate a market with dramatic levels of inequality, where the rich get richer\u2014where \u2018rich\u2019 in this case means \u2018hot,\u2019\u201d as Paul Eastwick, a professor of psychology at UC Davis and author of Bonded By Evolution<\/em>, puts it. But agents, Hrdli\u010dka theorizes, might be capable of surfacing \u201cdelicate matches\u201d that their human counterparts might never have otherwise considered.<\/p>\n The available research casts some doubt over that proposition. Two speed dating studies by Eastwick and other psychologists identified that compatibility is near-impossible to predict on the basis of people\u2019s hobbies, values, preferences, politics, profession, and so on\u2014the kind of information they are willing to self-report (or presumably feed into an AI). The most reliable predictor, Eastwick says, is the amount of time people spend with one another and whether they hit it off early in their first encounter. \u201cThink about compatibility as more of a growth process,\u201d says Eastwick. \u201cIt has to do with the story that two people build together.\u201d<\/p>\n Against that backdrop, for agentic dating to function as advertised, the AI would have to surface some sort of a latent truth about what makes two people compatible that humans have not yet identified. \u201cThis is the vanguard,\u201d says Eastwick. \u201cThis is where we\u2019re all struggling right now.\u201d<\/p>\n Other thorny problems with the Pixel Societies concept abound: Do interactions between two agents\u2014likely fed with differing quantities of data\u2014amount to anything in real life? How costly would it be to run this type of simulation at scale? Is there a workable business model that doesn\u2019t create an incentive mismatch between users looking for long-term relationships and the platform whose value depends on their continued singlehood?<\/p>\n