Home

He “Albert” Zhang

Fifth-year Ph.D. candidate (ABD), Informatics · College of Information Sciences and Technology, Penn State

I study what happens when AI systems become the interpreters of intimate human data — and who keeps control of what gets inferred.

1,120Citations
14h-index
17i10-index
31Students mentored
100+Papers reviewed

Scholar metrics as of 11 August 2026

01 — Research statement

I am He “Albert” Zhang, a fifth-year Ph.D. candidate in Informatics at the College of Information Sciences and Technology, Penn State University, advised by Distinguished Prof. John M. Carroll, and a student member of the Center for Socially Responsible Artificial Intelligence.

I build and empirically evaluate LLM- and LMM-based systems that infer meaning from sensitive material — qualitative interview transcripts, facial and physiological affect signals, ambient multimodal home sensing, and camera-mediated assistance for people with visual impairments — and I study how those systems reshape trust, disclosure, control, and agency for the people whose data they process.

When an AI system mediates the interpretation of personal experience, who retains control over what is inferred, retained, and disclosed?

I answer that in three registers at once — I build the systems, I study what people do with them, and I follow what that does downstream. Every paper below sits in one of the three.

Arc ISystems I buildTools, datasets, testbeds and defences — 15 works.
Arc IIEvidence I gatherStudies of what people actually do with them — 18 works.
Arc IIIConsequences I traceWhat it does to trust, ethics and institutions — 7 works.

My dissertation, Integrating Large Language Models into the Qualitative Research Process, takes that question where it bites hardest: qualitative research, where scholars routinely route confidential participant data through commercial models. I work closely with Prof. Syed M. Billah, Prof. ChanMin Kim, and Prof. Xinyi Fu. In 2025 the college named me the recipient of the IST Ph.D. Student Award for Research Excellence.

Before Penn State I spent a year as a full-time research assistant at The Future Laboratory, an interdisciplinary HCI lab at Tsinghua University, building the multimodal affect and smart-home testbeds that much of my systems work still draws on. I hold an M.Sc. in Computer and Mathematical Sciences with first-class honours from Auckland University of Technology (2021) and a B.Sc. with a double major in Computer Science and Data Science from Massey University (2019), both in New Zealand. On the industry side I have worked on applied LLM systems at Genentech and on large-scale behavioural modelling at Experian, Callaghan Innovation, and HKUST’s SyMLab.

Human-Centered AI · Human–AI Interaction · Trustworthy and Responsible AI · LLM/LMM Evaluation and Human Oversight · Affective Computing · Multimodal and Agentic Systems · CSCW · Accessibility · Ubiquitous Computing

02 — Research map

Systems I build Tools, datasets, testbeds, defences 15 works Evidence I gather Studies of what people actually do 18 works Consequences I trace Trust, ethics, institutions, theory 7 works 2022 2023 2024 2025 2026 Home platform[j.1] · smart home Patent: testbed[pat.2] · smart home QualiGPT[arXiv.2] · AI4Qual Sensor network[p.1] · smart home Patent: interaction[pat.1] · smart home VRMN-bD[c.4] · affect BubbleCam[c.6] · accessibility Image annotation[c.9] · LLM eval VR Calm+[p.6] · VR NaviGPT[p.4] · accessibility VR Calm Plus[c.14] · VR Wolfborn[demo.1] · embodiment AIoT home agent[p.7] · smart home PromptShield Home[p.10] · AI safety AI4Qual[tut.1] · AI4Qual Gov. communication[c.1] · social Decoding Fear[p.2] · affect Travel network[c.2] · social Smart-home review[j.2] · smart home QualiGPT in use[arXiv.4] · AI4Qual Fear & coping[j.3] · affect Twitch tool devs[c.5] · social YouTubers & GenAI[p.3] · social LMM practices[arXiv.3] · accessibility AI in qual. work[j.5] · AI4Qual Coding reliability[c.11] · AI4Qual Face emotion + LMM[c.12] · affect BVI & LMM use[c.8] · accessibility VRChat Discord[p.5] · social Learning Together[arXiv.1] · education Bot interviewer[c.17] · AI4Qual Human vs. AI help[c.15] · social Twitch devs II[c.16] · social AI governance[j.4] · trust Instrumental realism[c.3] · VR Students & LLMs[c.7] · education Trust & burden[c.10] · trust Follow-up ethics[p.9] · AI4Qual Parents & RAI[c.13] · education AI uncertainty[p.8] · accessibility
Systems Tools, datasets, testbeds, defences Evidence Studies of what people actually do Consequences Trust, ethics, institutions, theory 2022 2023 2024 2025 2026 Home platform[j.1] · smart home Patent: testbed[pat.2] · smart home QualiGPT[arXiv.2] · AI4Qual Sensor network[p.1] · smart home Patent: interaction[pat.1] · smart home VRMN-bD[c.4] · affect BubbleCam[c.6] · accessibility Image annotation[c.9] · LLM eval VR Calm+[p.6] · VR NaviGPT[p.4] · accessibility VR Calm Plus[c.14] · VR Wolfborn[demo.1] · embodiment AIoT home agent[p.7] · smart home PromptShield Home[p.10] · AI safety AI4Qual[tut.1] · AI4Qual Gov. communication[c.1] · social Decoding Fear[p.2] · affect Travel network[c.2] · social Smart-home review[j.2] · smart home QualiGPT in use[arXiv.4] · AI4Qual Fear & coping[j.3] · affect Twitch tool devs[c.5] · social YouTubers & GenAI[p.3] · social LMM practices[arXiv.3] · accessibility AI in qual. work[j.5] · AI4Qual Coding reliability[c.11] · AI4Qual Face emotion + LMM[c.12] · affect BVI & LMM use[c.8] · accessibility VRChat Discord[p.5] · social Learning Together[arXiv.1] · education Bot interviewer[c.17] · AI4Qual Human vs. AI help[c.15] · social Twitch devs II[c.16] · social AI governance[j.4] · trust Instrumental realism[c.3] · VR Students & LLMs[c.7] · education Trust & burden[c.10] · trust Follow-up ethics[p.9] · AI4Qual Parents & RAI[c.13] · education AI uncertainty[p.8] · accessibility
  • 2022
  • 2023
  • 2024
    • VRMN-bD[c.4] · IEEE VR 2024affective computingVRmultimodal↳ from [p.2]
    • BubbleCam[c.6] · CHI 2024accessibilityprivacy
    • LMM practices[arXiv.3] · arXiv:2407.08882accessibilitymultimodal↳ from [c.6]
    • QualiGPT in use[arXiv.4] · arXiv:2407.14925AI4QualLLM evaluation↳ from [arXiv.2]
    • Twitch tool devs[c.5] · CHI 2024social computing
    • Fear & coping[j.3] · IEEE Transactions on Games, 2024affective computingVR↳ from [c.4]
    • YouTubers & GenAI[p.3] · CHI EA 2024social computing
    • Instrumental realism[c.3] · iConference 2024VRembodiment↳ from [p.2]
    • Students & LLMs[c.7] · SIGITE 2024educationLLM evaluation↳ from [j.4]
    • AI governance[j.4] · Future Internet, 16(10), 2024trusteducation
  • 2025
    • Image annotation[c.9] · ICLR 2025 Workshop on Bidirectional Human–AI Alignment · Oral at CHI ’25LLM evaluationmultimodal↳ from [j.5]
    • NaviGPT[p.4] · GROUP 2025 Companionaccessibilitymultimodalagentic AI↳ from [c.9], [c.8]
    • VR Calm+[p.6] · IEEE ISMAR-Adjunct 2025VRembodimentaffective computing↳ from [j.3]
    • Learning Together[arXiv.1] · arXiv:2510.20123educationsocial computing↳ from [p.3]
    • Coding reliability[c.11] · AHFE 2025AI4QualLLM evaluation↳ from [j.5]
    • Face emotion + LMM[c.12] · MRAC 2025affective computingLLM evaluationmultimodal↳ from [c.9], [c.4]
    • BVI & LMM use[c.8] · CHI 2025accessibilitymultimodalLLM evaluation↳ from [arXiv.3]
    • AI in qual. work[j.5] · Computers in Human Behavior: Artificial Humans, 2025AI4QualtrustLLM evaluation↳ from [arXiv.4]
    • VRChat Discord[p.5] · CHI EA 2025social computingVR↳ from [c.5]
    • Trust & burden[c.10] · FAccT 2025trustsocial computing↳ from [j.5], [j.4]
  • 2026
    • VR Calm Plus[c.14] · CHI 2026VRembodimentaffective computing↳ from [p.6]
    • Wolfborn[demo.1] · DIS 2026 Companion · first author is an undergraduate menteeembodimentaffective computingVR↳ from [c.14]
    • PromptShield Home[p.10] · UbiComp 2026 · led a five-student team across five institutionsAI safetyagentic AIsmart home↳ from [p.7], [c.12]
    • AIoT home agent[p.7] · IUI 2026 Companionsmart homeagentic AImultimodal↳ from [p.1]
    • AI4Qual[tut.1] · Tutorial, IUI 2026, Limassol · lead organizer and instructorAI4Qualeducation↳ from [c.11]
    • Human vs. AI help[c.15] · IMX 2026social computingagentic AI↳ from [p.5]
    • Twitch devs II[c.16] · CSCW 2026social computing↳ from [c.5]
    • Bot interviewer[c.17] · HCOMP 2026AI4Qualtrustagentic AI↳ from [j.5]
    • Parents & RAI[c.13] · CHI 2026educationtrust↳ from [c.7], [arXiv.1]
    • AI uncertainty[p.8] · ASPIRE · 69th HFES Annual Meeting, 2026accessibilitytrust↳ from [p.4]
    • Follow-up ethics[p.9] · CHIWORK Adjunct 2026AI4Qualtrust↳ from [c.17]

Hover or focus any point to open its card. Arrows point from a project to what it made possible.

Figure — All 40 works, 2022 to 2026. Rows are contribution type — systems, evidence, consequences; colour is topic; arrows run from a project to what it made possible. Hollow points are preprints and patents; larger points are the six works that best define the research identity; the dotted ring marks a CHI 2024 Best Paper Honorable Mention.

03 — Topics

The three arcs above say what kind of contribution each paper makes. These six topics say what the papers are about. Every topic below is a sequence of questions rather than a pile of citations — the tiles read left to right, each answering a piece the previous one exposed, and the hatched tile is what is still open.

AI4Qual

11 works
AI4QualLLM evaluation

This is the dissertation. Researchers were already pasting confidential interview transcripts into ChatGPT before anyone had established whether that was defensible. I started by building the tool that made the question answerable, then spent three years finding out what actually happens: where an LLM's codes agree with a human's and where the agreement is coincidental, how trust gets renegotiated the moment a researcher re-scopes the model, and what changes when the machine stops assisting the interview and starts conducting it.

Scope. Covers the full qualitative pipeline — coding, inter-rater reliability, interviewing, follow-up generation, and the ethics of each. Does not cover automated content analysis at corpus scale, which is a different craft with different failure modes.

  1. 2023

    Can a general-purpose LLM code qualitative data at all?

    arXiv.2arXiv.4
  2. 2025

    Do researchers trust it once confidential transcripts are involved?

    j.5
  3. 2025

    Does it agree with human coders closely enough to count?

    c.11
  4. 2026

    And if the model runs the interview itself — what do people withhold?

    c.17p.9
  5. still open

    Where must a human stay in the loop, and how do we teach that?

    tut.1

Affective computing

9 works
affective computingVRembodiment

I build the emotion-inference pipelines that privacy and ethics scholarship needs to interrogate from the inside. It began with fear in VR horror games — an emotion strong enough to show up unambiguously in pose, physiology, and gaze — and produced a dataset of natural rather than staged behaviour. From there the question inverted twice: first, can a general-purpose multimodal model read emotion off a face without training, and where does it fail? Then, having measured affect thoroughly, can a physical object regulate it rather than only record it?

Scope. Covers multimodal affect capture, zero-shot emotion benchmarking, and tangible interventions in immersive environments. Does not extend to clinical affect assessment or diagnosis.

  1. 2023–24

    What does fear look like in multimodal signals, and why do players stay?

    p.2c.4j.3
  2. 2025

    Can an off-the-shelf multimodal model read emotion from a face?

    c.12
  3. 2025–26

    Can a physical object regulate affect instead of only measuring it?

    p.6c.14demo.1
  4. still open

    Which affect inferences should never ship at all?

Ambient agents & AI safety

10 works
smart homeagentic AIAI safety

Instrumenting a home to study it and giving that home an autonomous agent turn out to be the same engineering problem approached from opposite ends. At Tsinghua I built the multi-sensor testbed and the petabyte-scale pipeline behind it, which is also direct experience with the data-collection infrastructure that makes domestic privacy a live question. Years later, when multimodal models became capable enough to make decisions in that space, the obvious next question was adversarial: a home agent reads the room, and the room can be written on.

Scope. Covers domestic multimodal sensing, autonomous home-agent architecture, and ambient prompt-injection defence. Does not cover industrial IoT or robot manipulation.

  1. 2022–23

    How do you instrument a home richly enough to study what happens in it?

    j.1j.2p.1pat.1pat.2
  2. 2026

    Can a multimodal model make autonomous decisions in that home?

    p.7
  3. 2026

    What happens when the room itself carries an adversarial prompt?

    p.10
  4. still open

    Who audits an always-on agent living in domestic space?

Accessibility

5 works
accessibilityprivacy

Camera-mediated assistance is where interpretation and privacy collide most directly: to get help seeing, a blind user has to let someone — or something — look at their room. BubbleCam reframed that as a scoping decision rather than a blurring problem. When large multimodal models arrived, blind users adopted them faster than the design literature could keep up, so the next studies documented practice first and derived implications second. NaviGPT tested whether any of it survives contact with a real street, and the current work is about the hardest interface problem in the set: how a system that might be wrong should say so.

Scope. Covers remote sighted assistance, LMM assistants on mobile, real-time navigation, and uncertainty communication for people with visual impairments. Does not cover screen readers or motor accessibility.

  1. 2024

    How does a blind user share a camera feed without over-disclosing?

    c.6arXiv.3
  2. 2025

    What do blind users actually do with LMM assistants on a phone?

    c.8
  3. 2025

    Can real-time multimodal navigation hold up outside the lab?

    p.4
  4. 2026

    How should the system say “I am not sure”?

    p.8

Social computing

9 works
social computing

A recurring finding across these studies is that the systems people depend on are held together by labour nobody is paying for. Government pandemic messaging worked or failed on form, not just content. Stranded travellers rebuilt an entire route network through personal infrastructuring. Twitch's community-management stack is built by unpaid third-party developers, and those developers in turn run their own support economy on Discord. Generative AI is now entering all of it, which makes the comparison I care about tractable: put human and AI-powered support side by side in the same community and see what changes.

Scope. Covers platform and community studies on Twitch, Discord, VRChat, YouTube, and public-sector social media, using large-scale trace data alongside qualitative interpretation.

  1. 2022–23

    How do institutions and communities communicate under stress?

    c.1c.2
  2. 2024

    Who actually builds the tools communities run on?

    c.5p.3
  3. 2025

    How does generative AI enter an existing community’s support practices?

    p.5
  4. 2026

    Human support versus AI support — what changes in the exchange?

    c.15c.16

Trust, governance & education

10 works
trusteducation

The consequences arm of everything above. Universities wrote AI guidance quickly; the gap turned out to be between the policy language and what happens in a classroom, so the next study asked students rather than administrators. In public benefits, an LLM assistant does reduce the cost of learning a system — and relocates burden somewhere the evaluation was not looking. Parents applying their own responsible-AI criteria to children's learning do not reproduce the principles that published frameworks assume. Running under all of it is a philosophical commitment I made explicit early: an interpreting system is an instrument, and instruments mediate rather than transmit.

Scope. Covers institutional AI guidance, trust–burden dynamics in public services, learner and parent perspectives, and the philosophy of technology framing that holds the programme together.

  1. 2024

    What guidance exists, and what do students make of it?

    j.4c.7
  2. 2024

    What is the instrument doing to the experience it mediates?

    c.3
  3. 2025

    Does AI reduce administrative burden, or relocate it?

    c.10arXiv.1
  4. 2026

    Whose responsibility criteria count — the institution’s or the parent’s?

    c.13

04 — Recent news

  • I’ll serve as Associate Chair for the CHI 2027 Full Paper Track.
  • I designed and taught AI4Qual, a full tutorial on LLM-supported qualitative research, at IUI 2026 in Limassol, and presented our AIoT home-agent architecture in the companion track. [tut.1] [p.7]
  • PromptShield Home was accepted to UbiComp 2026; I’ll present it in Shanghai in October. I led the five-student team across five institutions. [p.10]
  • Our demo Wolfborn was presented at DIS 2026 in Singapore — first-authored by an undergraduate I mentored from the ground up. [demo.1]
  • I presented our comparison of human versus AI-powered support in VRChat communities at IMX 2026 in Athlone. [c.15]
  • I presented VR Calm Plus at CHI 2026 in Barcelona — the third iteration of a tangible-plus-VR system built with two undergraduate mentees. [c.14]
  • Two more acceptances: HCOMP on MLLM-led interviews and CSCW on Twitch developers, both presenting this autumn. [c.17] [c.16]
  • I was named runner-up for the IST Award for Excellence in Teaching Support.
  • I wrapped up an AI internship at Genentech, building and evaluating LLM systems with the Product and Data Science group.
  • I chaired a paper session at CSCW 2025 in Bergen, and presented two pieces of the affect work — VR Calm+ at ISMAR and our zero-shot facial emotion benchmark at MRAC. [p.6] [c.12]
  • AI Trust Reshaping Administrative Burdens appeared at FAccT 2025. [c.10]
  • I received the Ph.D. Student Award for Research Excellence and the Alumni Association Graduate Fellowship.
  • Two at CHI 2025 in Yokohama — smartphone interaction of blind users with LMMs, and generative AI in the VRChat Discord community — plus an oral on our human–LMM annotation framework. [c.8] [p.5] [c.9]
  • NaviGPT was presented at GROUP 2025 — real-time multimodal navigation for people with visual impairments. [p.4]
  • Our Twitch third-party developer study received a Best Paper Honorable Mention at CHI 2024, in the top 3.7% of 4,028 submissions. [c.5]
  • VRMN-bD, our multimodal fear-response dataset, was presented at IEEE VR 2024. [c.4]
  • I joined a Big Ideas Grant seed award at Penn State as co-principal investigator.

05 — Publications by contribution

Systems I build — Tools, datasets, testbeds, defences · 15 works

  • [c.14] · Conference paper · 2026

    VR Calm Plus: Coupling a Squeezable Tangible with Immersive VR for Stress Relief

    Zhang, H., Li, X., Zhou, X., & Fu, X.

    CHI 2026

    The full cycle — tangible hardware, immersive software, controlled evaluation — after two prior iterations built with undergraduate mentees.

    VRembodimentaffective computing
  • [demo.1] · Demonstration · 2026

    Wolfborn: Silent Emotional Catharsis and Empathy via Pseudo-Vocalized Embodiment

    Zhou, X., Zhang, H., Han, Y., Gao, H., Fu, X., & Zhao, S.

    DIS 2026 Companion · first author is an undergraduate mentee

    Pseudo-vocalized embodiment lets someone express and discharge emotion without making a sound — catharsis routed through the body instead of the voice.

    embodimentaffective computingVR
  • [p.10] · Poster / extended abstract · 2026

    PromptShield Home: Ambient Multimodal Prompt Injection Defense for Smart-Home Agents

    Zhang, H., Li, F., Long, D., Cui, Y., Zhang, P., Zhang, Y., Xu, Q., & Fu, X.

    UbiComp 2026 · led a five-student team across five institutions

    Ambient prompt injection is a new attack surface — text printed on a box in the kitchen can hijack a home agent. This builds and evaluates a defence against it.

    AI safetyagentic AIsmart home
  • [p.7] · Poster / extended abstract · 2026

    AIoT Smart Home Automation Architecture: Autonomous Decision-Making Powered by Multimodal Large Language Models

    Zhang, H., Zhang, Y., Li, B., Chen, W., Liu, Y., Xu, Q., & Fu, X.

    IUI 2026 Companion

    Hands domestic decision-making to a multimodal LLM and asks what an autonomous home-agent architecture actually needs to be safe to run.

    smart homeagentic AImultimodal
  • [tut.1] · Tutorial · 2026

    AI4Qual: A Comprehensive Field Guide to LLM-Supported Qualitative Research

    Zhang, H., Cai, J., Xie, J., Wu, C., Kim, C., & Carroll, J. M.

    Tutorial, IUI 2026, Limassol · lead organizer and instructor

    Turns the whole research programme into something teachable — the practices and failure modes, not just the findings.

    AI4Qualeducation
  • [c.9] · Conference paper · 2025

    Augmenting Image Annotation: A Human–LMM Collaborative Framework for Efficient Object Selection and Label Generation

    Zhang, H., Fu, X., & Carroll, J. M.

    ICLR 2025 Workshop on Bidirectional Human–AI Alignment · Oral at CHI ’25

    A division of labour for annotation: the model proposes objects and labels, the human arbitrates — faster and better calibrated than either alone.

    LLM evaluationmultimodal
  • [p.4] · Poster / extended abstract · 2025

    Enhancing the Travel Experience for People with Visual Impairments through Multimodal Interaction: NaviGPT, A Real-Time AI-Driven Mobile Navigation System

    Zhang, H., Falletta, N. J., Xie, J., Yu, R., Lee, S., Billah, S. M., & Carroll, J. M.

    GROUP 2025 Companion

    Puts real-time multimodal AI into the travel loop for blind users, testing whether the latency and the language hold up outdoors rather than on a bench.

    accessibilitymultimodalagentic AI
  • [p.6] · Poster / extended abstract · 2025

    VR Calm+: Furry Stress Ball as a Haptic Modality for Relaxation in Virtual Reality

    Zhang, H.*, Li, X.*, Zhou, X., & Fu, X.

    IEEE ISMAR-Adjunct 2025

    Adds a furry stress ball as a haptic channel in VR — physical squeezing as a relaxation modality rather than one more screen affordance.

    VRembodimentaffective computing
  • [c.4] · Conference paper · 2024

    VRMN-bD: A Multimodal Natural Behavior Dataset of Immersive Human Fear Responses in VR Stand-up Interactive Games

    Zhang, H., Li, X., Sun, Y., Fu, X., Qiu, C., & Carroll, J. M.

    IEEE VR 2024

    A multimodal dataset of natural fear behaviour — pose, physiology, gaze — captured in stand-up VR rather than staged lab conditions, plus a benchmark on top of it.

    affective computingVRmultimodal
  • [c.6] · Conference paper · 2024

    BubbleCam: Engaging Privacy in Remote Sighted Assistance

    Xie, J., Yu, R., Zhang, H., Lee, S., Billah, S. M., & Carroll, J. M.

    CHI 2024

    Privacy in remote sighted assistance is a framing problem, not a blurring problem: the user decides what enters the shared bubble.

    accessibilityprivacy
  • [arXiv.2] · Preprint · 2023

    QualiGPT: GPT as an Easy-to-Use Tool for Qualitative Coding

    Zhang, H., Wu, C., Xie, J., Kim, C., & Carroll, J. M.

    arXiv:2310.07061

    Packages GPT into a purpose-built coding tool, so the question stops being “ChatGPT versus a human” and becomes a question about tool design.

    AI4QualLLM evaluation
  • [p.1] · Poster / extended abstract · 2023

    Multi-channel Sensor Network Construction, Data Fusion and Challenges for Smart Home

    Zhang, H., Ananda, R., Fu, X., Sun, Z., Wang, X., Chen, K., & Carroll, J. M.

    ChCHI 2023

    Multi-channel sensing and fusion in a real home — RGB, depth, IR, audio, gait — with an honest account of what breaks at petabyte scale.

    smart homemultimodal
  • [pat.1] · Patent · 2023

    A Human-Machine Interaction System for Smart Home Environments, and Its Application Method

    Fu, X., Xu, Y.-Q., Zhang, H., Xue, C., Sun, Z., Gao, Y., & He, S.

    Patent, China, CN115291718A

    The human–machine interaction layer for the smart-home environment, filed as a patent.

    smart home
  • [j.1] · Journal article · 2022

    Design Research and Application Practice of Integrated Experimental Platform for Smart Home

    Fu, X., Zhang, H., Xue, C., Li, X., Sun, Z., & Xu, Y.-Q.

    Packaging Engineering, 43(16), 2022

    The instrumented smart-home platform everything downstream runs on — sensing hardware, capture pipeline, and the experimental protocol around it.

    smart homemultimodal
  • [pat.2] · Patent · 2022

    Smart Home Comprehensive Experiment System and Data Processing Method

    Fu, X., Xu, Y.-Q., Zhang, H., Xue, C., He, S., Sun, Z., & Gao, Y.

    Patent, China, 2022-09

    The testbed and its data-processing method, filed as a patent.

    smart home

Evidence I gather — Studies of what people actually do · 18 works

  • [c.15] · Conference paper · 2026

    Comparative Analysis of Human vs. AI-powered Support in VRChat Communities on Discord: User Engagement, Response Dynamics and Interaction Patterns

    Zhang, H., Kim, B., Carroll, J. M., & Cai, J.

    IMX 2026

    Human and AI-powered support in the same communities, side by side: engagement, response dynamics, and what changes in the exchange itself.

    social computingagentic AI
  • [c.16] · Conference paper · 2026

    Twitch Third-Party Developers’ Support Seeking and Provision Practices on Discord

    Cai, J., Zhang, H., Liu, Y., Carroll, J. M., & Yu, C.

    CSCW 2026

    Where third-party developers go for help and how they give it — the support economy underneath the tooling economy.

    social computing
  • [c.17] · Conference paper · 2026

    When the Interviewer Is a Bot: Behavior, Breakdowns, and Trust in MLLM-Led Interviews

    Zhang, H., Chukwuma, K., Kim, C., & Carroll, J. M.

    HCOMP 2026

    When the interviewer is a machine, disclosure changes shape — what participants tell a bot, what they withhold, and how breakdowns erode trust mid-interview.

    AI4Qualtrustagentic AI
  • [arXiv.1] · Preprint · 2025

    “Learning Together”: AI-Mediated Support for Parental Involvement in Everyday Learning

    Li, Y., Xie, J., Ling, Y. F., Zhang, H., Wang, G., Huang, G., Yu, R., & Chen, S.

    arXiv:2510.20123

    AI-mediated support for parents helping children learn — the mediation sits between two people, not between a person and a task.

    educationsocial computing
  • [c.11] · Conference paper · 2025

    Exploring Inductive and Deductive Qualitative Coding with AI: Investigating Inter-Rater Reliability between Large Language Model and Human Coders

    Zhang, H., Wu, C., Xie, J., Rubino, F., Graver, S., Cai, J., Kim, C., & Carroll, J. M.

    AHFE 2025

    Measures human–LLM inter-rater reliability across both inductive and deductive coding, separating real agreement from coincidental agreement.

    AI4QualLLM evaluation
  • [c.12] · Conference paper · 2025

    Zero-shot Emotion Annotation in Facial Images Using Large Multimodal Models: Benchmarking and Prospects for Multi-Class, Multi-Frame Approaches

    Zhang, H., & Fu, X.

    MRAC 2025

    Benchmarks what general-purpose multimodal models can and cannot infer about emotion from faces — the empirical ceiling for a technology already deployed in hiring and education.

    affective computingLLM evaluationmultimodal
  • [c.8] · Conference paper · 2025

    Beyond Visual Perception: Insights from Smartphone Interaction of Visually Impaired Users with Large Multimodal Models

    Xie, J., Yu, R., Zhang, H., Lee, S., Billah, S. M., & Carroll, J. M.

    CHI 2025

    Interaction traces of visually impaired users with LMMs — what they ask, what they trust, and where perception-first design assumptions fail them.

    accessibilitymultimodalLLM evaluation
  • [j.5] · Journal article · 2025

    Harnessing the Power of AI in Qualitative Research: Exploring, Using and Redesigning ChatGPT

    Zhang, H., Wu, C., Xie, J., Lyu, Y., Cai, J., & Carroll, J. M.

    Computers in Human Behavior: Artificial Humans, 2025

    Traces the full arc from exploring ChatGPT, to using it on live projects, to redesigning it — and shows trust is negotiated at the point of re-scoping, not at first contact.

    AI4QualtrustLLM evaluation
  • [p.5] · Poster / extended abstract · 2025

    Generative AI in Virtual Reality Communities: A Preliminary Analysis of the VRChat Discord Community

    Zhang, H., Zha, S., Cai, J., Wohn, D. Y., & Carroll, J. M.

    CHI EA 2025

    First look at how an established VR community absorbs generative AI into its everyday support practices.

    social computingVR
  • [arXiv.3] · Preprint · 2024

    Emerging Practices for Large Multimodal Model (LMM) Assistance for People with Visual Impairments: Implications for Design

    Xie, J., Yu, R., Zhang, H., Lee, S., Billah, S. M., & Carroll, J. M.

    arXiv:2407.08882

    Documents the practices blind users were already inventing around LMM assistants, well ahead of any design guidance existing for them.

    accessibilitymultimodal
  • [arXiv.4] · Preprint · 2024

    When Qualitative Research Meets Large Language Model: Exploring the Potential of QualiGPT as a Tool for Qualitative Coding

    Zhang, H., Wu, C., Xie, J., Rubino, F., Graver, S., Kim, C., Carroll, J. M., & Cai, J.

    arXiv:2407.14925

    Puts QualiGPT in front of working researchers and reports where it saves labour and where they refuse to delegate.

    AI4QualLLM evaluation
  • [c.5] · Conference paper · 2024

    Third-Party Developers and Tool Development for Community Management on Live Streaming Platform Twitch

    Cai, J., Lin, Y., Zhang, H., & Carroll, J. M.

    CHI 2024 Best Paper Honorable Mention

    Community management on Twitch runs on tools built by unpaid third-party developers — the platform depends on labour it does not acknowledge.

    social computing
  • [j.3] · Journal article · 2024

    Understanding Fear Responses and Coping Mechanisms in VR Horror Gaming: Insights from Semi-Structured Interviews

    Zhang, H., Li, X., Fu, X., Qiu, C., Zhang, J., & Carroll, J. M.

    IEEE Transactions on Games, 2024

    How players actually cope with fear in VR — the reasons behind the signals that VRMN-bD records.

    affective computingVR
  • [p.3] · Poster / extended abstract · 2024

    A Preliminary Exploration of YouTubers’ Use of Generative AI in Content Creation

    Lyu, Y., Zhang, H., Niu, S., & Cai, J.

    CHI EA 2024

    How creators fold generative AI into production, and how much of that they choose to show their audience.

    social computing
  • [c.2] · Conference paper · 2023

    Reconnecting An International Travel Network: The Personal Infrastructuring Work of International Travelers in A Multi-facet Crisis

    Lyu, Y., Zhang, H., & Carroll, J. M.

    ChCHI 2023

    International travellers rebuilt a broken travel network through personal infrastructuring work — labour that was invisible, uncompensated, and load-bearing.

    social computing
  • [j.2] · Journal article · 2023

    A Review of the Frontier Research on Future Smart Home

    Fu, X., Zhang, H., Xue, C., Sun, T., & Xu, Y.-Q.

    Science & Technology Review, 41(8), 2023

    Maps the frontier of smart-home research and locates the open problems — the survey that framed the later agent work.

    smart home
  • [p.2] · Poster / extended abstract · 2023

    Decoding Fear: Exploring User Experiences in Virtual Reality Horror Games

    Zhang, H., Li, X., Qiu, C., & Fu, X.

    ChCHI 2023

    First pass at what fear feels like from inside a VR horror game, in players’ own accounts — the study that motivated collecting the signals.

    affective computingVR
  • [c.1] · Conference paper · 2022

    Tips, Tidings, and Tech: Governmental Communication on Facebook During the COVID-19 Pandemic

    Haq, E. U., Braud, T., Lee, L. H., Mogavi, R. H., Zhang, H., & Hui, P.

    DG.O 2022

    Regression modelling and multilingual sentiment analysis on official pandemic messaging — which forms of government communication actually landed.

    social computing

Consequences I trace — Trust, ethics, institutions, theory · 7 works

  • [c.13] · Conference paper · 2026

    Understanding Parents’ Perspectives on Responsible AI for Children’s Self-Directed Learning

    Xie, J., Wu, C., Wang, G., Yu, R., Zhang, H., Metoyer, R., & Chen, S.

    CHI 2026

    Parents’ own criteria for responsible AI in children’s learning, which do not line up neatly with published responsible-AI principles.

    educationtrust
  • [p.8] · Poster / extended abstract · 2026

    Communicating AI Uncertainty in Assistive Navigation for People with Visual Impairments

    Shaffer, H., Shaikh, A., Zhang, H., & Xie, J.

    ASPIRE · 69th HFES Annual Meeting, 2026

    An assistive navigator that hides its uncertainty is more dangerous than one that admits it; this asks how a system should say “I am not sure” usefully.

    accessibilitytrust
  • [p.9] · Poster / extended abstract · 2026

    Ethics and Social Responsibility in AI-Assisted Interviewing: An LLM-in-the-Loop Study of AI-Generated Follow-Up Questions

    Zhang, H., Liu, Y., Guan, X., Cai, J., & Carroll, J. M.

    CHIWORK Adjunct 2026

    AI-generated follow-up questions are where interviewing ethics concentrate — what an LLM in the loop should and should not be permitted to ask.

    AI4Qualtrust
  • [c.10] · Conference paper · 2025

    AI Trust Reshaping Administrative Burdens: Understanding Trust-Burden Dynamics in LLM-Assisted Benefits Systems

    Jo, J., Zhang, H., Cai, J., & Goyal, N.

    FAccT 2025

    Trust and administrative burden are coupled: an LLM assistant can cut learning costs in a benefits system while quietly relocating burden somewhere else.

    trustsocial computing
  • [c.3] · Conference paper · 2024

    Exploring Virtual Reality through Ihde’s Instrumental Realism

    Zhang, H., & Carroll, J. M.

    iConference 2024

    Reads VR through Ihde: the headset is not a window but a mediating instrument, and that reframing changes what counts as a faithful experience.

    VRembodiment
  • [c.7] · Conference paper · 2024

    The Future of Learning: Large Language Models through the Lens of Students

    Zhang, H., Xie, J., Wu, C., Cai, J., Kim, C., & Carroll, J. M.

    SIGITE 2024

    LLMs in education from the students’ side rather than the institution’s — what they use them for, and what they think it is costing them.

    educationLLM evaluation
  • [j.4] · Journal article · 2024

    AI Governance in Higher Education: Case Studies of Guidance at Big Ten Universities

    Wu, C., Zhang, H., & Carroll, J. M.

    Future Internet, 16(10), 2024

    Compares AI guidance across Big Ten universities; the governance gap sits between policy language and instructional practice, not in the absence of policy.

    trusteducation

✦ marks the six works that best define the research identity. * denotes equal contribution.

06 — Mentoring and teaching

Since 2021 I have recruited and mentored 31 students — 18 undergraduate, 12 master’s, 1 research assistant — across 14 institutions in China, the United States, the United Kingdom, and Hong Kong, including four supervised through the Penn State IST Summer Intern program.

They joined as research interns on 12 projects I initiated and led, where I set the research questions, directed study and system design, supervised data collection and analysis, and mentored manuscript preparation. 21 mentees have co-authored peer-reviewed papers with me — several as second author, one as co-first author, and one I mentored all the way to first authorship. Five have stayed with me across multiple projects.

  • 2026AI4Qual, IUI 2026Lead organizer and instructor, full tutorial, Limassol, Cyprus · July 2026
  • 2025 – 2026Invited guest lecturer5 lectures at 4 institutions across 3 countries — Penn State, San José State, Tsinghua, FGV EBAPE (Brazil)
  • 2025 – 2026DS 435: Data EthicsTeaching assistant — Spring 2025, Fall 2025, Fall 2026
  • 2022 – 2023IST 302 · IST 402IT Project Management; Data, Environment, and Society
  • 2026Runner-up, IST Award for Excellence in Teaching Support

07 — Service

  • 2027ACM CHIAssociate Chair, Full Paper Track
  • 2026ACM CHI · ACM CSCWAssociate Chair, Poster Tracks
  • 2025ACM CHI · ACM CSCWAC for User Experience and Usability and for Late-Breaking Work; Ninja AC, Paper Track, and AC, Poster Track
  • 2025Program committeeICHEC · ICLR Workshop on Bidirectional Human–AI Alignment · LAW at NeurIPS
  • 2024Program committeeACM Learning at Scale
  • 2022 – 2024Program committeeChCHI · ACII · DG.O
  • 2025Session chairCSCW 2025, Bergen — Making Work Meetings Better. CHI 2025, Yokohama — Co-ideation; Creativity Support
  • 2022 – 2026Conference reviewerACM CHI, CSCW, UIST, DIS, HCOMP, L@S, SIGGRAPH, Multimedia, MobileHCI, UbiComp/IMWUT, VRST, CHI PLAY, ASSETS · AAAI, COLM, ICLR and NeurIPS workshops · IEEE VR, ISMAR, ISWC · AIS ICIS
  • 2023 – 2026Journal reviewerIJHCI · Computers in Human Behavior and CHB Reports · Sociological Methods and Research · International Journal of Qualitative Methods · Policy Studies Journal · Journal of Learning Analytics · Multimedia Tools and Applications · Cogent Arts & Humanities
  • 2024 – 2026Special Recognition for Outstanding ReviewsCHI ’24, ’25, ’26 · CSCW ’24, ’25
  • 2024 – 2026Student volunteerCSCW and IMX 2026 · CHI 2025 · UIST 2024
  • 2026 –Board member, ICACHIInternational Chinese Association of Computer Human Interaction
hpz5211@psu.edu · University Park, PA, USA ACM SIGCHI · IEEE Computer Society · Center for Socially Responsible AI, Penn State