No. 01
Human Agency × AI
How can AI systems support — not suppress — the human capacity to choose, voice, and act?
Vol. III · Spring 2026
I will be here
with you,
my friend.
Hi, I'm Xianglu — most call me TJ.
I explore how culture shapes agency — and how AI might illuminate rather than obscure the paths we choose.
Now
PhD student (Fall 2026),
Social Psychology · University of Michigan
PI: Dr. Catherine Thomas
Past
Lab Manager, Stanford HAI
w/ Dr. Geoffrey Cohen
Columbia Business School
w/ Dr. Michael Morris
Focus
How AI can deliver culturally wise interventions — and building the theory to support it.
Methods: from NLP modeling of global narratives to field studies. Scope: East, West, Global South.
I care about fairness, and helping people find their agency.
My research question: how can AI deliver culturally wise interventions? I'm building the theory of culturally wise interventions — spanning from NLP modeling of global narratives to field studies across East, West, and the Global South. I care about fairness, and helping people find their agency.
↓ keep reading
§ One
Inquiries
Four threads I keep pulling on. They overlap more than they diverge — each one an attempt to ask what makes a system genuinely human-centered, not just human-shaped.
No. 02
Voice & Power Dynamics
The psychological architecture of speech, status, and silence inside algorithmic systems.
No. 03
Value Alignment
Designing timely, tailored, targeted interventions that respect ethical and cultural context.
No. 04
Culture & Identity
How narratives and perceived scarcity shape behavior across cultural contexts.
§ Two
Recent dispatches
- 01 · Meta-Analysis
Stanford: AI Depression Detection Meta-Analysis
Meta-analytic evaluation of depression detection from social media data — pooled predictive accuracy and ethical considerations for real-world deployment. Screened 4,361 records, analyzed 67 eligible studies.
Stanford HAI
Accepted at AAAI 2025
- 02 · AI & Psychology
Stanford: Ubiquitous Well-being & Patchable AI
Runtime-adaptive architecture enabling timely, targeted psychological interventions that can be injected without retraining LLMs. Synthesizing ubiquitous computing and intervention science.
Stanford HAI
Feb 2025 – Present
- 03 · AI & Education
Stanford: SheRockets Child-Safe AI
Child-safe, web-based GenAI experience featuring a female engineer tutor for K-12 girls. Discrete-choice conjoint study with 125 parents on AI-powered STEM learning.
Stanford HAI
Feb 2025 – Present
- 04 · AI & Crisis Intervention
Stanford: LLMs for Crisis Detection
Evaluating 30 models across suicide risk detection and dynamic intervention timing. Two-phase evaluation with expert blind review and stakeholder-centered assessment in high-risk contexts.
Stanford HAI
Feb 2025 – Present
- 05 · AI & Negotiation
Columbia: ACE Negotiation Coaching
LLM-based Assistant for Coaching nEgotiation (ACE) with specialized feedback for underrepresented groups. Examining identity salience effects on AI coaching gains.
Columbia Business School
Mar 2024 – Present
§ Three
Chat with Xianglu
Ask me anything about my research, psychology, AI, or just have a casual chat.
Hi there! 👋 I'm Xianglu. Ask me anything about my research, psychology, or just say hi.
§ Four
From the journal
Research Toolkit
Building Your Own Research Toolkit — My Playbook
A personal research diary about creating your own research opportunities — from spotting real-world problems to fearless outreach and cultivating sisu.
December 15, 2024
Travel
Falling Hard for Finland
A love letter to the land of a thousand lakes, where kindness meets authenticity. From Santa Claus Uber drivers to gentle shop owners.
December 20, 2024
AI Research
My AI-in-Research Workflow
How I use AI in my research pipeline — and why I never trust AI overall. Practical guide with the “AI duel” technique and Dao/Shu framework.
January 2025
Subscribe
New essays in your inbox.
Occasional letters when something new is published. No spam, unsubscribe anytime.