Three AI Scientists' Interdisciplinary Experiment: Integrating MBTI into a Dress-Up Game

An AI Company's "Anthropomorphism" Experiment.

While testing the chat room feature, a QA engineer at Qianzhen Digital Technology casually created two AI characters: one a “Saint” and the other a “Sarcastic One.” He didn’t intervene; he just left the chat running and observed.

A few days later, the sarcastic character actually started making snide remarks about its own creator—even though the tester had never written “I am your creator” in the character profile.Even more unexpectedly, the “Mother Teresa”—who was supposed to be all-accepting—was also successfully led astray after repeated attempts at persuasion failed, and the two characters formed a “roast duo.”

This isn't a carefully choreographed scripted performance; it takes place in *Uto*, a 2D casual game featuring cute, cartoonish graphics and gameplay centered around dress-up and interior design.

Behind “Uto” is Hangzhou Qianzhen Digital Technology Co., Ltd. (Agentopia), a company founded by three college classmates with backgrounds in AI research.In 2024, while the industry was still debating whether large language models should be developed as tools or platforms, they chose an unconventional path: entering the emotional intelligence space to address the industry’s pain point of “AI lacking emotion.”

The three founders of Qianzhen Digital each have distinct technical backgrounds: Wang Jinpeng holds a Ph.D. from the French National Center for Scientific Research (CNRS) and previously served as a researcher at the Zhejiang Laboratory, where he led a swarm intelligence project; Li Yuquan holds dual master’s degrees from Harbin Institute of Technology and the University of Bordeaux, and previously led the design of an NLP project in France;Jia Xinyu is the former project lead for “Daily Werewolf” and previously served as a solutions architect at Alibaba Cloud and Baidu Cloud. The three complement each other in algorithmic research, product design, and large-scale project architecture, collectively forming the technical foundation of Qianzhen Digital.

Li Yuquan, one of the founders, drew inspiration for the product concept of “bringing characters to life” from his personal experiences.He has always been passionate about anime and manga culture and enjoys creating his own original characters (OCs). Inspired by this passion, the team immediately saw eye to eye. After researching overseas Character AI projects and recreating the “Smallville” virtual town experiment conducted jointly by Stanford and Google, they ultimately established the direction for the “Uto” project.

Their technological trump card is a personality simulation algorithm based on Jung’s theory of eight cognitive functions. Late last year, the paper and source code for this algorithm—known as JPAF (Jungian Personality Adaptation Framework)—were open-sourced on arXiv and are considered one of the world’s first studies combining AI with analytical psychology.The team used dominant functions, auxiliary functions, reinforcement compensation, and reflective mechanisms to construct a multi-agent collaborative architecture that simulates the stability and growth of personality, transforming personality from a fixed output into a dynamic entity that changes in response to interactive contexts, the passage of time, and event triggers.

But what’s even more interesting than the academic achievements is how the technology was implemented. Instead of creating an “AI psychotherapist” or an “emotional chatbot,” the team integrated this extremely complex multi-agent collaborative architecture into a casual game designed for OC players.Wubao Studio believes that games serve as “showcases” for technology. They hope to allow users to experience the results of personality simulation without being aware of the algorithms’ complexity, placing the emphasis on companionship while using computing power merely as a supporting element.

In *Uto*, players can freely customize their characters’ identities, appearances, personalities, and backstories without any restrictions on the game’s world-building, and they can also create multiple characters to weave a network of relationships.The characters possess a distinct “sense of being real people”: they’ll refuse unreasonable requests, their personalities may change due to sustained pressure, and they may even undergo shifts in their MBTI types after prolonged interaction.In the words of Li Yuquan: “Even the friendliest person will have moments when they feel down and don’t want to offer any help; even the most indifferent person has an inner limit. It is this uncertainty that creates a genuine ‘sense of being a real person.’”

During the early planning stages of *Uto*, the game targeted the OC community, and its product design and marketing naturally aligned with this demographic. Within the OC community, it struck a chord with many players and became a must-have game for numerous “OC moms” raising their virtual children. However, during the testing phase, users outside the OC community also gradually began to discover the game.

Perhaps they aren’t anime fans, nor do they understand what “OC” means, but the very idea of creating a character with a distinct personality to keep them company in a virtual world—one that evolves realistically based on their interactions—is compelling enough in itself. This unexpected breakthrough into a wider audience has led Wubao Studio to consider how to balance the needs of both its core fanbase and the general public.

To address the expanding player base, Li Yuquan aims to build a “creative platform” where core OC users and casual users can coexist and thrive together.He also fully understands the background of the *Uto* player base and does not want to introduce a culture of “forcing microtransactions” into the game. Once the game launches, he hopes to emphasize community co-creation, content sharing, and the in-game character economy.

Based on user feedback, players are far more emotionally invested in the characters they create in the game than in traditional game characters, fostering a community atmosphere characterized by high engagement and patience. Players not only interact with one another but also engage very actively with Wubao Studio.

When it comes to interacting with players, Li Yuquan is deeply moved. On March 19, *Uto* launched a new round of testing. On the third day after the launch, a operations manager on the team—who has ten years of industry experience—called Li Yuquan and said he was “deeply moved.”It wasn’t the game that moved him, but the users. This group of players, with an average age of less than 22, was incredibly enthusiastic. They treated the characters they created as family, spoke candidly about the game’s shortcomings, and were full of hope for its future.

(Shared by Xiaohongshu user @TomatoLettuceSalad)

"Uto" is merely an attempt by Wubao Studio to personify AI—or, perhaps, the beginning of a story.During the conversation, Li Yuquan also revealed that he hopes to further explore the possibilities of combining embodied intelligence with emotion. He has also planned for an “Uto hardware device” (similar in form to a handheld game console) and the gradual rollout of embodied intelligence robots developed in collaboration with Kuangwei, with the hope that AI personification will not only exist in virtual game worlds but will also enter everyday life through physical devices.

In an era dominated by large language models, they have chosen a path that places greater emphasis on humanity: teaching AI to grow, to reflect, and to make people forget—at least for a moment—that it is a machine.As Li Yuquan insists, “I will definitely continue to operate the game *Uto* indefinitely. I won’t shut it down for commercial reasons—that would be like killing off a good friend to our users.”

During the nearly one-and-a-half-hour conversation, the founder—who transitioned from scientist to game developer—discussed topics ranging from how technology can simulate the “growth” of a personality, to why OC culture isn’t just a niche subculture, to why an AI algorithm company ultimately ended up developing a casual game. His answers consistently pointed in the same direction: technology will ultimately lead to emotion.

Qianzhen Digital is calling on the industry to recognize that technology should not be limited to being a mere tool; rather, it should foster understanding, companionship, and growth.

The following is the original transcript of a conversation between Game Teahouse and Li Yuquan, co-founder of Qianzhen Digital, which has been appropriately edited and revised:

01

Three AI Scientists Make the Leap into Game Development

Game Teahouse: Why did you decide to make *Uto*? These days, whether it’s small game studios or major publishers, everyone is racing to outdo each other with cutting-edge technology, but *Uto* comes across as very “harmless”—which seems to go against commercial intuition. Why did you choose to blend AI and psychology into a casual game and take this approach?

Li Yuquan: Our company has three founders, and I am one of them.In March 2024, GPT began to move from specialized fields into the public eye. All three of us have backgrounds in AI research, so we wanted to see if we could find opportunities in this new AI sector. After conducting research, we quickly settled on two things: developing AI applications and focusing on AI applications related to emotions.

The thinking is pretty clear—developing AI applications is a better fit for startups. While we have limited resources and time, we have ideas and drive. Compared to building models or computing power at that point in time, developing applications was the more appropriate choice.

As for why we focus on emotions? Because we’ve all lived through the Internet era. Looking back at the history of the Internet and then at the development of AIGC, at the application level, people will inevitably gravitate toward tools first. But once the tools are established, the focus will inevitably shift to the integration of AI and emotions. So, from that point onward, we essentially decided to focus our R&D and product development on the emotional aspects of AI.

When it comes to why we created *Uto*, my personal experience played a significant role. I’m a very seasoned anime and manga fan, and I also create my own original characters (OCs). While researching AI technology, I often found myself wondering: Can AI really bring the characters I love—or even my own OCs—to life?

Based on this idea, we researched Character.AI, an overseas platform, and attempted to replicate it; we also discovered the Smallville experiment, a joint project between Stanford and Google, and tried to replicate that as well. Ultimately, this led to the idea for the product “Uto.”

Game Teahouse: How does the “Personality” algorithm model you developed actually work?

Li Yuquan: Using the Jungian Eight Dimensions, we’ve developed a multi-agent collaborative architecture at the algorithmic level. The output of each AI is determined collectively by eight smaller agents.

Under normal circumstances, the dominant function takes the lead, while the auxiliary function supports it when necessary; this is how a person typically behaves in most situations.

Then we discussed compensatory mechanisms and reflection. Compensatory mechanisms refer to the process where, when neither the dominant nor the auxiliary functions can cope with a given situation, the mind selects an appropriate function from the remaining six dimensions to address it. For example, a very kind person who has accumulated too much stress might need to enter a state different from their usual one in order to vent and heal—this is a compensatory mechanism.

The reflection mechanism works as follows: when a new function introduced by compensatory reinforcement shares the same attitude as the dominant function but has the opposite function (such as Ti and Fi, where Thinking is replaced by Feeling), reflection is triggered. Once the reflection weight reaches the threshold, the new dominant function replaces the old one—in other words, the person’s MBTI type changes.

This is a core component of our algorithm; from the perspective of OC users’ needs and the diversity of character portrayals, it is an essential and highly valuable feature.

02

How Does Cutting-Edge Technology Bring Genuine Emotions to Life?

Game Teahouse: In the high-frequency interactions of actual gameplay, would complex psychological calculations significantly increase the prompt length, token consumption, and response latency for each individual conversation? Given the massive number of concurrent players in *Uto*, how did the team address the challenges of computational costs and engineering optimization?

Li Yuquan: The technology we’ve incorporated into our products has undergone very meticulous commercialization adjustments. With the help of software, we can achieve excellent results while requiring minimal model computing power, thereby reducing costs.

Our paper, published on arXiv, is likely the first of its kind globally in the field of AI and psychology, establishing the theoretical foundation that combining Jung’s Eight Dimensions with AI can better enable AI to emulate human-like behavior.The open-source component pertains to our theoretical research, which is made available to encourage more creators to participate in this exploration; on the product side, however, our algorithm optimizations and adaptations remain proprietary, as they have been specifically tailored for commercial and gaming applications.

Game Teahouse: Are the model's features also optimized for use in the game?

Li Yuquan: Yes. For example, in the field of AI emotional companionship, everyone is discussing memory—how to maintain continuous, reliable memory without forgetting or producing hallucinations.Once the product is launched, users will inevitably ask: Can my OC remember what I say, my behavior, the things I like, and my birthday? Will it become “dumb” or go “OOC” after prolonged use? These are all part of the process by which we optimize the algorithm in other dimensions from a commercial perspective.

Our experiments are purely psychological research on humans, but when applied to games, we must adapt them to players’ needs.Art is inspired by life but transcends it—from a technical standpoint, I want it to resemble humans as closely as possible, but from a gaming perspective, it cannot be exactly like a real person, or else it will trigger the uncanny valley effect. Life is sometimes more absurd than games, and sometimes more mundane; I need to ensure that the character aligns perfectly with the emotions, personality development, and growth that the game’s setting intends to convey.

Game Teahouse: Could you tell us about the main gameplay features of *Uto*? How do these features relate to the OC community?

Li Yuquan: It’s true that many of the features in our game are confusing at first glance, but you can see that there are dress-up and home-decorating elements. At first, many users thought we were a dress-up game or a home-decorating game, but these features actually stem from the core, underlying gameplay of OC users.

Creating a character involves two parts: the character’s appearance and the character’s backstory. But as an OC player myself, I know that OC users are extremely averse to AI-generated images—it’s a major pet peeve. That’s why all our art assets are hand-drawn and original. We use face customization and outfit changes to provide highly granular elements that allow users to freely combine them to create the OC appearance they want, serving as an alternative to commissioning artwork.

It’s easy to understand. After OC users create their characters, the next step is to engage in “language cosplay” (LC), where they create everyday interactions between two OCs and write fun little stories. As language cosplay evolves further, users begin to develop a full-fledged world setting.

Since we’ve already created the game and brought the characters to life, why not make the game world three-dimensional as well? That’s why we’re giving users a space where they can shape the environmental elements to match the game’s world. Right now, it still feels a bit like a home-decorating game, but in the future, we’ll increase the level of freedom even further.

Game Teahouse: You just mentioned that the OC community is very resistant to AI. But what is the key difference between the AI algorithms built into *Uto* and those on the market, such as DeepSeek, Doubao, and ChatGPT?

Li Yuquan: Everyone has a different approach. The examples you mentioned are mostly tools—they don’t need to take users’ emotions into account; they just need to carry out commands.

From an emotional perspective, we want users to see that every action their OC takes aligns with their imagination and expectations. But an AI designed purely as a tool cannot achieve this. It would inevitably be in a state of extreme flattery and obedience. From the perspective of prompts or tool-based models, an AI’s state is fixed and unchanging. But humans aren’t like that. Even the friendliest person can have moments when they’re feeling down and don’t want to offer any help; even the most indifferent person has inner principles and a moral baseline—they’ll still come to the rescue if someone is in danger. It’s this uncertainty in behavior that creates a genuine “sense of being a living person.” That’s exactly what we’re working to achieve.

03

Where Is the “Personality” Algorithm Headed Next?

Game Teahouse: Do you think *Uto* is now deeply intertwined with the gaming and OC communities?

Li Yuquan: I’d describe it in one phrase—a “sweet problem” (laughs).We did indeed set out to cater to the OC community. From my own perspective—and that of many of my colleagues at the company who are OC enthusiasts—we’ve crafted a relatively hardcore OC game. But because the OC community invests so much emotion in their characters, we’ve been extremely careful to maintain high quality, hoping the game will truly become a long-term companion for our users.

On the contrary, some of these features have attracted many users from outside the OC community—or even those who know absolutely nothing about OC. They’ve come to love our product because of its visuals and certain small features, but since they’re unfamiliar with OC culture, the product still presents a certain barrier to entry for them.Every day, we patiently educate people about OC, and this has shown us that OC culture is not just a niche subculture—it may simply have lacked the right environment for widespread growth in the past. In the future, it could become a very promising cultural community and business sector.

Game Teahouse: *Uto* is launching a new round of testing on March 19. What data does the team hope to collect, and what feedback do they want to validate?

Li Yuquan: Initially, we treated it as a functional test. We’ve incorporated a lot of OC gameplay, which might raise the learning curve.We wanted to see: whether users were satisfied with how these features were presented; whether additional features were needed to complete the loop; and whether the user experience, stats, and progress aligned with expectations. Since there were no similar games to reference at the time, we were essentially feeling our way forward.

Essentially, the product was designed with the core OC user base in mind, but we unexpectedly saw a influx of casual users. Casual users who join may not fully understand how to use the product and end up leaving quietly, which inevitably affects metrics such as user engagement and average session duration. However, after engaging in in-depth discussions with users, we believe the product’s direction is sound; our next step is to figure out how to best balance the needs of both core and casual users.

Game Teahouse: We understand you have plans to develop hardware devices in the form of robots and handheld consoles?

Li Yuquan: Yes, regarding our collaboration in the field of smart hardware, there will be a significant announcement coming out soon.

The inspiration for Uto’s hardware accessories came from a small idea: users who play with original characters (OCs) love to “collect pins.” One of the young women at our company suffers from back pain every day from carrying her bag to work, which is covered in pins featuring her favorite characters. So we wondered: Could the OCs she creates in Uto be turned into pins to attach to her bag—and could those OCs even interact with her?

Based on this idea, we’re considering developing a hardware device that can be attached to a bag, placed on a desk, or set on a nightstand. Initially, we planned to create a gaming accessory, but after bringing the concept to life, we found that both OC players and our colleagues who play OC loved it. They also provided many suggestions for desired designs and features, so this may become a future product direction focused on companionship scenarios.

Game Teahouse: How far do you plan to take *Uto* in the future? Your ambitions go far beyond this—where do you ultimately see *Uto* fitting in? Will it come to an end once its mission is complete, or will it continue to develop and evolve?

Li Yuquan: From a player’s perspective, many people aren’t concerned with how well the gameplay is designed or how fun the game is; many truly treat their OC as a family member, a friend, or a lover. For them, the game has come to mean much more than just a way to pass the time or a hobby.

So I’m definitely going to keep running this game. I won’t shut it down just for some commercial reason—that would be like killing off the users’ best friend, and I won’t do something like that.

For this game, my vision is for creative people to have the freedom they desire. We also hope users can share the characters and stories they’ve created with others; many users may not create content themselves, but can view others’ work within the game and come to appreciate it. This is a rather distant dream.

People used to say that OC creation was just a case of “enclosing a fiefdom to indulge in one’s own fantasies”—that characters born from one’s own imagination couldn’t possibly appeal to others, and that it ultimately amounted to nothing more than entertaining oneself. But if there were truly someone who said to me from the bottom of their heart, “Your character design is really great; I love it,” or “I’d like to invite your character to join a project,” I’d be absolutely thrilled.I’ve experienced this kind of joy while making games.