I study & design the relationship between people and AI.
Co-writing
Youcontrol
AIreliance
You 68AI 32
Prototype: the two minds are wired into the control. Drag the seam or use ←→: push toward AI and watch the human mind disperse as control flows across. Scroll ↓ for the next page.
Research · the through-line
Three threads, one question: when people and AI think together, who stays in control?
Select a thread to read its papers
Writing & argument2025–present
ABE: AI for Brainstorming & Editing
A scaffolded AI writing coach, built into Google Docs, that positions generative AI as a critical reader, not a writer, guiding students through argumentation, perspective-taking, and iterative revision.
Extending Bandura's proxy agency to machines: a student's confidence that a GenAI tool can do their academic work on their behalf, measured across a national survey of 900 U.S. undergraduates.
73 children, ages 4–8, co-creating oral stories with an AI agent or a human partner: those with the AI gave briefer replies, those with a person told richer ones, mapping how a partner shapes a young storyteller.
Press play. The transcript writes itself as I talk.
About · why both
Most people either study AI or build with it. I do both, because it's the same question.
How much of our thinking do we hand to a machine, and how do we stay in the loop while we do it?
Before any of this, I spent two years as a Spanish translator and simultaneous interpreter, the human proxy standing between people who couldn't understand each other.
Now I study AI as that proxy.
Yuqing (Kyra) Xing is a PhD student at USC's Rossier School of Education, and a mixed-methods researcher and UX designer. Her work follows one through-line: the human–AI relationship in learning: agency, reliance, control, and proxy efficacy, a learner's confidence in getting AI to act on their behalf. She moves fluently between statistics and thematic coding, and between a research question and a working prototype.
UI design and interaction prototyping (Figma, Webflow), plus UX research: surveys, user interviews, usability, thematic synthesis. Then builds the thing: ships AI-native products across iOS, macOS, and web: Swift/SwiftUI, React/TypeScript, Python, front-end HTML/CSS/JS, and the Claude API.
Teaching assistant for EDUC 599: AI and Learning Technologies, a graduate course in USC's M.Ed. in Learning Design & Technology, Fall 2025. Co-designed course materials (lecture slides, in-class activities), delivered class presentations, mentored students through AI-augmented learning-environment prototypes, and ran grading and weekly office hours.
The path
2016–20
B.A. Spanish· Shanghai Int'l Studies University
2020–22
Spanish translator & interpreter· Shanghai Electric
An AI cooking & grocery assistant for Chinese kitchens: dish ideas from what's already in the fridge, recipe capture by photo, voice, or OCR, auto-built shopping lists, and a step-by-step cook mode.
A no-login, cross-timezone meeting scheduler: everyone paints their availability in their own timezone; it finds the overlap, suggests alternatives, and drafts the invite email.
A reading coach for English learners: inline word lookup, a growing vocabulary notebook, AI reading help, and newsletter subscriptions pulled straight from the inbox.
A research-reading companion: syncs with Zotero, pulls in Google Scholar alerts, and reads papers alongside you with AI.
Proxy Efficacy · a field explainer
When a student hands work to a machine, what kind of agency is that?
The field keeps asking whether AI helps or harms learning. That's the wrong question. The answer depends on a belief we don't yet measure.
Sixty years of Bandura's social-cognitive theory give us the vocabulary for it. This is the construct at the centre of my research, AI Proxy Efficacy, and why "does AI help learning?" has no answer until you know a student's calibration.
01 · The taxonomy
You can act three ways.
Bandura distinguishes three modes of agency, and each runs on its own efficacy belief. Act yourself, act through others, or act together. Tap each:
02 · The construct
Self-efficacy asks can I? Proxy efficacy asks can it?
Self-efficacy · the human
Can I write this essay?
A belief about my own capability.
Proxy efficacy · the machine
Can it write this for me?
A belief about the AI's capability, on my behalf.
AI Proxy Efficacy (APE)
A student's task-specific confidence that a generative-AI system can perform a delegated academic task effectively on their behalf.
Defined here, building on Bray et al. (2001, p. 426) and Hanham et al.'s technological proxy efficacy (2014, p. 4).
It is not trust, not perceived usefulness, not confidence in yourself. It is a capability belief about a third party, pointed at a task you have delegated.
03 · The same tool, five relationships
Proxy is a relationship, not a piece of software.
The same ChatGPT is a tool when you check spelling, a proxy when it writes the essay you submit, a partner when you argue with it to sharpen a thesis you then write yourself. What moves it along the line is how much of the work (and the authorship) you delegate.
Toolamplifies your own actionself-efficacy
Agentexecutes a delegated, bounded task↓
Proxyacts in place of your competenceproxy efficacy
Partnerreciprocal, you stay engaged↑ shared
Co-authorauthorship genuinely sharedcollective
04 · The core insight
It's not the level of the belief. It's the calibration.
A high proxy-efficacy belief is neither good nor bad in itself. What matters is whether it tracks what the AI can actually do on this task. Pick a task; drag your belief against the AI's real capability.
The task you're delegating
AI can
you believe
Calibrated
05 · The fork that decides everything
The same belief can grow you or erode you.
Bandura never resolved this. In 1982 he warned that leaning on a proxy "reduces opportunities to build the requisite skills." By 1997 he allowed the opposite: a proxy can "free time and effort to enhance personal efficacy in other areas." Which branch a student lands on is not fixed. It turns on conditions we can measure and change:
Belief
Engagement
Task type
Complementarity: your own capability grows.
The central, testable claim (H1). APE → self-efficacy is positive or negative depending on measurable moderators, which is exactly why the belief is worth measuring rather than assuming. The same construct has opposite downstream signs; interventions are predicted to move students from the substitutive branch to the complementary one.
06 · Discriminant validity
What proxy efficacy is not.
A reviewer's first objection: isn't this just trust, or perceived usefulness, relabeled? No. The difference is structural, not cosmetic.
Trust in AI
An attitude spanning ability, benevolence, integrity. APE is the ability facet only, made task-specific. A non-social proxy has no motives to trust. Glikson & Woolley (2020)
Perceived usefulness (TAM)
A belief that using the tool improves my outcomes. APE is the belief the agent can do the task for me, a capability attribution, not an adoption judgment. Davis (1989)
Self-efficacy
Belief in your own capability. APE is belief in a third party's. Proxy agency, not personal. Bandura (1997)
Reliance / dependency
A behavior: the act of delegating. APE is the belief that drives it, separable from the act. Ferrario (2025); Mai et al. (2026)
07 · Why the field needs this
A learner-centred construct AIED doesn't yet have.
The GenAI-in-education literature is rich in trust, usefulness, literacy, and dependency, but has no validated construct for a student's belief in the AI's capability to act on their behalf, and no model of how that belief builds or erodes their own competence. APE is that missing piece: purpose-built, domain-specific, embedded in a moderated model of the substitution–complementarity fork.
APE→reliance→over-reliance · offloading→skill ↓
APE→engagement · SRL→deep learning→skill ↑
Which branch fires is set by measurable moderators: calibration · task type (machine- vs human-better) · critical thinking & autonomy · novice status.
That turns a vibe ("does AI help learning?") into something decidable and intervenable. AI literacy, calibration training, capability transparency, and task design become the levers predicted to move students from the substitutive branch to the complementary one.
The work
This is new. I'm building the instrument and the model.
Proxy efficacy for a generative-AI proxy: defined, measured across the domains students actually delegate, and tested with real learners. If you work on human–AI interaction, AI literacy, or learning at scale, I'd like to talk.
