Runqi Zou
I'm a researcher in human-computer interaction and computational cognitive science. I'm interested in cognition, behavior, and the ways intelligent systems influence how people perceive, decide, and interact. Across my projects, I focus on connecting technical implementation with questions about human cognition.
I'm a research assistant with Julian Togelius in the Game Innovation Lab at the NYU Department of Computer Science and Engineering, and with Laurence Maloney in the Decision Lab at the NYU Department of Psychology and Center for Neural Science.
News
-
Oct 2026
Language as Trait, Politeness as Strategy, a follow-up to LURK Werewolf on linguistic effects in LLM social-deduction agents, was submitted to NAACL 2027.
-
Jun 2026
LURK Werewolf was accepted for an oral presentation at the IEEE Conference on Games 2026.
-
Jun 2026
A manuscript on structure detection and event prediction in binary sequences, with M. Dal Martello and L. Maloney, is in preparation.
-
Apr 2026
Aligned Information Bottleneck for Multimodal Social Sentiment Analysis was published in IEEE Transactions on Computational Social Systems.
-
Dec 2025
PureMetric received an Honorable Mention at the International Design Awards.
Publications & Manuscripts
† Equal contribution · * Corresponding authors
-
Language as Trait, Politeness as Strategy: Controlled Tests of Linguistic Effects in LLM Social-Deduction Agents
Manuscript submitted to NAACL 2027
-
What good is classifying the generator when the best prediction never changes?
Manuscript in Preparation
-
LURK Werewolf: Evaluating Linguistic Bias in Large Language Models
IEEE Conference on Games (CoG 2026) · Oral Presentation
-
Aligned Information Bottleneck for Multimodal Social Sentiment Analysis
IEEE Transactions on Computational Social Systems
Research
Coming Soon
Language as Trait, Politeness as Strategy: Controlled Tests of Linguistic Effects in LLM Social-Deduction Agents
A follow-up study on the LURK framework. It treats language as a fixed trait of an agent and politeness as a strategy an agent can choose, and runs controlled tests of how each changes credibility and voting when LLM agents play social deduction games.
Coming Soon
What good is classifying the generator when the best prediction never changes?
Behavioral experiments and Markov chain models of how people detect structure in binary sequences and predict what comes next.
LURK Werewolf: Evaluating Linguistic Bias in Large Language Models
→A Werewolf framework for testing whether LLM agents react to meaning or to wording. Moving agents from logic predicates to natural language changed 52.6% of later decisions in 7-player games.
Aligned Information Bottleneck for Multimodal Social Sentiment Analysis
→A text-anchored information bottleneck for sentiment analysis across video, audio, and text. Improves over strong baselines on the MOSI and MOSEI benchmarks and stays more robust under modality noise and imbalance.
Selected Design Work
Floating Umbrella
A robotic umbrella that hovers overhead and follows a walking user, no free hand required.
PureMetric
Embedded sensing that tracks beverage freshness after opening and forecasts the drinkable window.
ZKP Dynamics
Web3 platform that proves asset requirements with zero-knowledge proofs, without exposing wallet history.
Sound of Body
Biometric installation that turns heart rate and blood pressure into personalized music and visuals.
Sentverse
Wearable device that captures scents and plays them back through electrical nasal stimulation.
Love Utopia
Satirical RPG set in a town where romance runs on timed tasks and coin rewards.
Awards
-
2025
International Design Awards, Honorable Mention
Product Design (Smart Home Products)
-
2025
A' Design Award, Winner
Installation Design
-
2024
Muse Award, Silver Winner
Product Design (Wearable Technologies)