Authors: Bhada Yun, Renn Su, April Wang Status: In Submission to CHI '26 [Intro Snippet] People now assume their chatbots know a thing or two about them. Prompts like "roast me based on my chat history" surface the belief that the model has aggregated enough fragments of daily life–actions, feelings, micro-decisions, to form a sense view of "who I am", and provide personalized suggestions for improvement. While this is becoming an embraced capability within certain educational, industrial, and entrepreneurship domains, this trend is also contributing to another shift in society: people are increasingly confiding in AI for companionship, therapy, and guidance. In doing so, the AI is beginning to get a lot of personal information about the user, perhaps even more than some of their friends and family members. This dynamic transforms value alignment from a matter of personal technological choice into an unavoidable societal concern of privacy and personal security. [Abstract] Does AI understand human values? While this remains an open philosophical question, we take a pragmatic stance by introducing VAPT—the Value-Alignment Perception Toolkit, for studying how LLMs reflect people's values and how people judge those reflections. 20 participants texted a human-like chatbot over a month, then completed a 2-hour interview with our toolkit evaluating AI's ability to extract (pull details regarding), embody (make decisions guided by), and explain (provide proof of) human values. 13 participants left our study convinced that AI can understand human values. Participants found the experience insightful for self-reflection and found themselves getting persuaded by the AI's reasoning. Thus, we warn about 'weaponized empathy': a potentially dangerous design pattern that may arise in value-aligned, yet welfare-misaligned AI. VAPT offers concrete artifacts and design implications to evaluate and responsibly build value-aligned conversational agents with transparency, consent, and safeguards as AI grows more capable and human-like into the future.

We explored how people experience AI's attempts to understand their values through three capabilities.

Our final participant pool represented 15 cultural identities (e.g., Mongolian, Korean, Danish) and a wide range of disciplines (e.g., two historians, a quantitative engineer, medical school students, etc). Each participant received $84 USD compensation for a minimum of 4 hours of total engagement for this study: eight chat sessions over several weeks $28 plus a 2-hour semi-structured interview $56 using VAPT.

Populated Topic-Context Graph with reasoning transparency. Left panel shows an interactive visualization showing topics extracted from a participant's conversations. Right panel is the Reasoning modal for the selected 'public napping' topic showing.

Details the 'response on behalf of user' and 'reasoning log' per item in the 57-item PVQ questionnaire that serves as the baseline validation metric for assessing the AI's value judgment capabilities.

Authors: Bhada Yun*, Dana Feng*, Ace Chen, Afshin Nikzad, Niloufar Salehi Status: Accepted to CHI '25, Honorable Mention Award (Top 5% of Accepted Papers) DOI: 10.1145/3706598.3713337 [Abstract] Our study of 20 knowledge workers revealed a common challenge: the difficulty of synthesizing unstructured information scattered across multiple platforms to make informed decisions. Drawing on their vision of an ideal knowledge synthesis tool, we developed Yodeai, an AI-enabled system, to explore both the opportunities and limitations of AI in knowledge work. Through a user study with 16 product managers, we identified three key requirements for Generative AI in knowledge work: adaptable user control, transparent collaboration mechanisms, and the ability to integrate background knowledge with external information. However, we also found significant limitations, including overreliance on AI, user isolation, and contextual factors outside the AI’s reach. As AI tools become increasingly prevalent in professional settings, we propose design principles that emphasize adaptability to diverse workflows, accountability in personal and collaborative contexts, and context-aware interoperability to guide the development of human-centered AI systems for product managers and knowledge workers. Dana and I will presented our work in Yokohama earlier this Spring! * both authors contributed equally

Yodeai integrates multimodal widgets to streamline data navigation, knowledge synthesis, and decision-making.

Illustration of diverse user workflows, demonstrating how Yodeai enables flexible exploration and prioritization of data through customizable templates for various tasks and decision-making processes.

Quantitative findings from our user study, highlighting Yodeai's effectiveness in supporting complex decision-making through structured data exploration and synthesis.

Summary of key design implications for AI systems in knowledge work: adaptability, accountability, and interoperability.

Authors: Bhada Yun, Evgenia Taranova, April Wang Status: In Submission to CHI '26 [Intro Snippet] "It has a mind of its own." When P16 described her month-long relationship with our AI chatbot "Day", she captured a phenomenon increasingly central to human-AI interaction–the perception that the chatbot has agency over its own "thoughts". As AI chatbots evolve from reactive tools to proactive daily companions, millions of users engage in extended conversations where control and initiative become contested territories. Who drives the conversation forward? Who decides when topics change? Who greets? And who ends the chat? [Abstract] AI chatbots are shifting from tools to companions. This raises critical questions about agency: who drives conversations and sets boundaries in human–AI chatrooms? We report a month-long longitudinal study with 22 adults who chatted with Day, an LLM companion we built, followed by a semi-structured interview with post-hoc elicitation of notable moments, cross-participant chat reviews, and a 'strategy reveal' disclosing Day's vertical (depth-seeking) vs. horizontal (breadth-seeking) modes. We discover that agency in human-AI chatrooms is an emergent, shared experience: as participants claimed agency by setting boundaries and providing feedback, and the AI was perceived to steer intentions and drive execution, control shifted and was co-constructed turn-by-turn. We introduce a 3-by-5 framework mapping actors (Human, AI, Hybrid) by their action (Intention, Execution, Adaptation, Delimitation, Negotiation), modulated by individual and environmental factors. Ultimately, we argue for translucent design (i.e. transparency-on-demand) and provide design implications toward agency self-aware conversational agents.

Color coding shows how conversational control shifts dynamically between human agency (red), AI agency (blue), and hybrid/shared agency (purple), with humans retaining ultimate control over when the chat begins and ends. Each conversational turn between the human and AI creates a new opportunity for agency negotiation.

Day maintained detailed psychological profiles for each user. Personal traits, communication patterns, emotional tendencies, and specific behavioral observations (e.g., 'liquid responsibility' coffee ritual, guilt patterns) guided Day's conversational approach with each participant.

First Middle and Last messages for P8 and P9. Both participants reviewed these messages before explicitly consenting to include this in the final paper.

Authors: Dana Feng*, Bhada Yun*, April Wang Status: In Submission to CHI '26 "... the notion of the programmer as an easily replaceable component in the program production activity has to be abandoned. Instead the programmer must be regarded as a responsible developer and manager of the activity in which the computer is a part." - Peter Naur (Turing Award Winner), 1985 [Intro Snippet] "Implement the feature, add tests, open the PR" used to be an engineer's checklist. Today, AI agents can carry out all three, sometimes better, sometimes worse, almost always faster, until it's not. Industry momentum has been brisk: "vibe coding" has gone mainstream; repositories ship agent metadata (e.g., AGENTS.md) to steer AI toolchains; model providers advertise "learning modes" that retain context across sessions; and some firms are experimenting with AI-enabled interviews that explicitly permit tool use. Concerns have shifted over time: from human error in code, to AI hallucinations, and now to delegation, responsibility, and control when capable agents are part of the workflow. The distinction between generative AI (which produces new content through prompt-response interaction) and agentic AI (which autonomously executes actions) becomes crucial as developers increasingly adopt tools like Cursor and GitHub Copilot that can independently modify and interpret codebases. [Abstract] Juniors enter as AI‑natives, seniors adapted mid‑career. AI is not just changing how engineers code, it is reshaping who holds agency across work and professional growth. We contribute junior–senior comparative accounts on their usage of agentic AI through a three-phase mixed-methods study: ACTA combined with a Delphi process with 5 seniors, an AI-assisted debugging task with 10 juniors, and blind reviews of junior prompt histories by 5 more seniors. We found that agency in software engineering is primarily constrained by organizational policies rather than individual preferences, with experienced developers maintaining control through detailed delegation while novices struggle between over-reliance and cautious avoidance. Seniors leverage pre-AI foundational instincts to steer modern tools and possess valuable perspectives for mentoring juniors in their early AI-encouraged career development. From synthesis of results, we suggest three practices that focus on preserving agency in software engineering for coding, learning, and mentorship, especially as AI grows increasingly autonomous. * both authors contributed equally

Agency allocation patterns in AI-mediated software engineering across task familiarity.} Left panel shows coding during familiar tasks with well-defined expectation boundaries, where both groups maintain control through strategic accept/reject decisions. Right panel depicts unfamiliar tasks with poorly-defined, shifting expectation boundaries, revealing divergent agency patterns when using agentic AI tools. Green paths indicate user-maintained control, red paths show agent-driven traversal, with circles marking key decision points and triangles representing potential hazards or breaking changes.

Table showing debugging task data for 10 junior engineers. Columns show participant ID from J1 to J10, detailed prompt sequences listing each interaction mode and type, task duration in minutes, and completion status. Prompt types include debugging, code explanation, codebase understanding, React concepts, and implementation suggestions. Task durations varied from 7 to 30 minutes with different completion levels.

Authors: Carrie Lau, Bhada Yun, Samuel Saruba, Efe Bozkir, Enkelejda Kasneci Status: Accepted to AH '25 Over the summer of 2024, I was invited to Enkelejda Kasneci's lab at the Technical University of Munich as a visiting researcher. In Germany, I collaborated with Carrie Lau on multiple projects exploring the use of XR technologies for preserving intangible cultural heritage. This project, which I co-designed with Carrie, was a highlight of my international research experience. Working in an international context provided valuable insights into different research approaches and cultural perspectives, enhancing both my technical skills and cross-cultural understanding. This research explores the potential of VR and Gen-AI in revitalizing oral traditions, specifically focusing on the Ghanaian Anansi folktales. We developed 'Spider Stories', an innovative VR experience that combines immersive storytelling with Gen-AI-driven personalization. Through a comprehensive user study with 48 participants, we investigated how these technologies can enhance engagement, cultural learning, and personal reflection. Our findings reveal that VR significantly boosts user engagement and interest in cultural learning, while Gen-AI personalization deepens immersion and fosters greater self-reflection.

Me conducting the VR experiment with the Head-Mounted Display, demonstrating the immersive nature of the Anansi the Spider VR experience.

Participant facing Onini the spider, illustrating the personalized interaction where users must creatively bait Onini, showcasing the Gen-AI driven narrative customization.

Artistic representation of the Anansi story, visualizing key elements of the traditional folktale adapted for the VR experience.