Research
My earlier research examined information manipulation, political communication,
and coordinated editing on Wikipedia. My current work investigates how interactions
with conversational AI may influence human reasoning and judgment, and why their effects
might differ across users. Across these topics, I emphasize rigorous measurement,
causal identification, and reproducible evaluation.
Ongoing research program
When AI Agrees: Sycophancy, Human Reasoning, and Adult Development
This research investigates whether and how conversational AI influences human
reasoning, judgment, and agency. A central question is whether these effects
vary with adults' developmental structures of meaning-making. Combining
ego-development measures, experimental manipulations of AI behavior, and
measures of integrative complexity, the work examines heterogeneity in
responses to sycophantic AI.
The program integrates preregistered human experiments with computational
behavioral evaluation and research on human-calibrated simulation.
Open-source experimental prototype · In development
Sycophancy Survey Agent
A locally hosted conversational AI prototype for investigating how
different assistant behaviors may shape human reasoning and meaning-making.
Built on Qwen3-14B, it implements neutral and sycophantic conversational
conditions using the same underlying model, with differences in evaluative
affirmation and alignment with the user's framing.
The current prototype supports discussions of AI regulation and clean energy,
adjustable conversation duration, optional web search, and transcript export.
Treatment calibration and evaluation are ongoing.
View the open-source prototype →
·
Methodology →
Open-source research tooling
Controllable Sycophancy Evals
An open-source behavioral-evaluation framework developed alongside my research
on heterogeneous responses to conversational AI. It supports controlled
variation in synthetic-user reasoning profiles while holding topic, stance,
and confidence fixed, with separate generation, constraint checks, and
stage-blind evaluation. Validation results and limitations are documented
in the repository.
Explore the code & methodology →
Talks & collaboration
I welcome research collaborations at the intersection of AI behavior, human reasoning,
adult development, computational social science, and causal inference. I am
particularly interested in experiments on heterogeneous responses to AI,
behavioral evaluation of language models, and AI-mediated reasoning and agency.
I co-organized SICSS Zürich at ETH in 2019 and have presented work in academic
and industry settings. I also welcome invitations for talks and guest lectures.
Get in touch →