- Awarded Best Paper at CHI 2025, the Premier International Conference on Human-Computer Interaction
- Implements "Future Self" Agents Utilizing Large Language Models to Reflect Users' Personal Values and Identities

▲ (From left) Hayeon Jeon (PhD student, Interdisciplinary Program in Artificial Intelligence, Seoul National University; first author), Professors Hajin Lim (corresponding author) and Eun-mee Kim (corresponding author) from the Department of Communication at SNU, and Professors John Zimmerman and Laura Dabbish (co-authors) from Carnegie Mellon University's Human-Computer Interaction Institute

Seoul National University College of Engineering announced that a joint research team from SNU and Carnegie Mellon University (CMU) has developed a human-centered artificial intelligence (AI) technology that enables young adults to explore career paths through conversations or letter exchanges with their "future selves."

Going beyond simply providing career-related information, this innovative AI technology offers personalized feedback tailored to users' values and identities, thereby helping young adults make career decisions grounded in self-understanding and meaningful reflection.

This research, conducted by the “SNU-CMU Human-Centered AI Research Center” established jointly by SNU and CMU in February of this year, is being recognized as an early success demonstrating the societal applicability of AI technologies. The research team includes Professors Hajin Lim and Eun-mee Kim from the Department of Communication at SNU, doctoral student Hayeon Jeon from SNU’s Interdisciplinary Program in Artificial Intelligence, and Professors John Zimmerman and Laura Dabbish from CMU’s Human-Computer Interaction Institute (HCII).

This research paper has been accepted for official presentation at CHI 2025 (ACM Conference on Human Factors in Computing Systems), the world's premier international conference in the field of human-computer interaction (HCI), scheduled to be held from April 26 over six days in Yokohama, Japan. The paper, titled "Letters from Future Self: Augmenting the Letter-Exchange Exercise with LLM-based Agents to Enhance Young Adults' Career Exploration," was awarded the Best Paper Award, an honor given only to the top 1% of submitted papers.

The "Future Self Letter" exercise is widely utilized in career-planning classes around the world. In this self-reflective activity, students either write letters to their future selves or assume their future perspectives to communicate back to their present selves. Extensively employed in counseling and mentoring within psychology and education, this technique helps students vividly envision their futures and create more concrete career plans. However, students often face significant cognitive burdens when trying to independently imagine their future selves. Additionally, the imagined future self frequently lacks realism and vividness, limiting students' deep engagement in career exploration.

To address these challenges, the joint research team designed a novel user experience (UX) in which young adults interact with their future selves through AI-powered letter exchanges or real-time conversations. Leveraging large language models (LLMs), the team developed personalized "future-self agents" reflecting each user's unique characteristics and integrated these agents into the traditional "Future Self Letter" exercise.

In a week-long study involving 36 young adults, participants first wrote a letter addressed to their future selves three years from now. They were then randomly assigned to one of three groups, each experiencing interaction with their future selves differently. The control group employed the conventional method, independently composing replies from their imagined future selves. The two experimental groups either read AI-generated letters from a future-self agent or engaged in real-time chats with the agent.

The findings revealed that participants who read letters from the future-self agent demonstrated the highest levels of engagement and overall satisfaction with the exercise, significantly benefiting from more vividly visualized and concrete future selves. Additionally, the chat-based interaction, which enabled quick feedback and flexible communication, was particularly effective in facilitating active career exploration and practical information exchange.

This study has been recognized for its originality in both its technical approach—introducing a novel form of self-reflection through AI-enabled interaction with one’s future self—and its empirical approach, which systematically compares the impacts of different interaction modalities.

To develop personalized future-self agents reflecting participants’ personalities, values, and current life situations, the research team utilized their self-developed framework called "SPeCtrum." By integrating users’ multidimensional identities and career contexts into large language models (LLMs), this framework enabled the creation of agents that authentically embodied individualized self-concepts and emotional experiences. The paper detailing this framework was also accepted by NAACL 2025, one of the premier academic conferences in natural language processing, affirming its scholarly significance.

Furthermore, the research provided meaningful implications for human-centered AI design by empirically comparing the effects of two distinct interaction modalities—letter writing and real-time chatting—on career exploration. According to the findings, the letter-based interaction allowed participants to read and reflect at their own pace, promoting deeper emotional introspection through iterative contemplation. In contrast, the chat-based interaction facilitated more active information seeking and flexible career exploration through rapid feedback and dynamic dialogue. These outcomes highlight that the mode of interaction can substantially shape the quality and trajectory of self-reflective experiences.

The LLM-based future-self agent developed in this study is anticipated to support not only self-directed career exploration among young adults but also various other aspects of life requiring self-reflection. For instance, engaging in conversations with a future-self agent could effectively aid tasks such as setting academic goals, cultivating better daily habits, managing emotions, and maintaining mental well-being. Particularly for young people in environments with limited access to formal career counseling infrastructure, personalized interactions with a future-self agent could provide meaningful emotional support and motivation beyond mere career guidance.

Furthermore, the framework and interaction design developed through this research hold promise for broader applications in designing diverse social agents, including—but not limited to—future-self agents. These contributions may also serve as foundational resources, offering new possibilities and establishing benchmarks for the ongoing advancement of human-centered AI systems.

Hayeon Jeon, the first author and PhD student, explained the direction of the research, saying, “This study began with the belief that AI should not replace human decision-making, but rather become a companion that helps individuals engage in deeper self-reflection. . We prioritized designing AI not as an 'answer provider,' but as a facilitator of users' internal conversations to support meaningful career exploration." She expressed optimism for the future application of this technology, adding, "I hope young adults can use interactions with their future-self agents as a valuable experience to manage anxieties and concretely envision their possibilities."

Meanwhile, Professor Hajin Lim, the lead investigator, highlighted the ethical considerations raised by the future-self agent’s immersive and persuasive qualities. She observed, "The advice generated by AI was so vivid and realistic that some participants began to interpret it as a 'predetermined future.' This underscores the importance of carefully calibrating the agent’s communication style and level of intervention to preserve user autonomy and encourage open interpretation."

As the study confirmed the strong psychological influence of interacting with a future-self agent, concerns were also raised about potential over-reliance on AI or reduced self-determination, depending on how the AI's advice is interpreted and used. Furthermore, since the process of integrating a user's identity and values into the agent involves collecting various types of personal data, ethical considerations such as privacy protection and transparency in data usage will be essential for future implementation

Accordingly, the joint research team plans to conduct follow-up studies with a broader and more diverse group of users, aiming to refine AI interaction methods that promote self-reflection, preserve autonomy, and foster user trust. They will also focus on developing sophisticated human-centered AI technologies that offer practical support for personal growth and life direction.

▲ Letter Exchange with AI-Based Future Self: After the participant (left) sends a letter to their future self, the AI-generated “future-self agent” (right), representing the participant three years from now, replies via letter or real-time chat. This aims to alleviate anxieties and uncertainties commonly encountered during career exploration.

[Reference Materials]
- Paper Title/Conference : “Letters from Future Self: Augmenting the Letter-Exchange Exercise with LLM-based Agents to Enhance Young Adults' Career Exploration”, CHI 2025
- Paper Link: https://arxiv.org/pdf/2502.18881

[Contact Information]
PhD Student Hayeon Jeon, Interdisciplinary Program in Artificial Intelligence, Seoul National University / jhy94520@snu.ac.kr