About me
I'm a Ph.D. candidate in Psychology at Princeton University, with a minor in Statistics & Machine Learning. I work with Professors Tania Lombrozo and Tom Griffiths, whose research spans psychology, cognitive science, and computer science.
I'm broadly interested in how people reason, learn, and make decisions, particularly as AI becomes increasingly integrated into everyday life. My current research investigates how people evaluate information and explanations provided by AI, when they defer to AI judgments, and how these interactions affect critical thinking and learning, as well as decision-making in moral contexts and high-stakes domains such as law and medicine. I'm also interested in understanding where human and LLM reasoning differ, and what these differences reveal about the nature of intelligence.
To study these questions, I combine behavioral experiments with computational modeling. Ultimately, I hope this work can inform the design of AI systems that support, rather than replace, human reasoning and learning.
Prior to starting my Ph.D., I studied young children's curiosity, learning, and cognitive development, including their intrinsic motivation to explore their own competence, using Bayesian models to formally tease apart what motivates exploration. This work also shaped how I think about studying intelligence more broadly. One unique challenge of developmental research is that we are trying to study cognition in agents who can't report what they know — a similar challenge to what AI researchers face today. My training in developmental research provided me with tools for exactly this challenge: designing novel paradigms that dissociate competence from performance, rule out alternative explanations, and probe underlying mechanisms rather than taking behavior or performance at face value.
Outside of research, I enjoy spending time in nature (I LOVE camping!), reading, practicing yoga, doing soothing art projects and hanging out with my furrends (check out their photos in "For Fun"!). From 2021 to 2023, I also served as a co-host for the Stanford Psychology Podcast – it was one of the coolest experiences I've ever had! If you are interested, check it out here!
Current Research
- Explanations and deference to AI: Across three preregistered studies (N = 3,907), we found that LLM rewriting increased people's evaluations of explanations and their willingness to defer to them, even when the underlying content remained unchanged, and that changes in linguistic properties, particularly lexical complexity, accounted for much of these effects. Manuscript under review. With Tania Lombrozo.
- Critical thinking and learning from AI: Can the linguistic changes introduced by LLM rewriting make flaws in an explanation's reasoning harder to detect, even when the underlying reasoning remains unchanged, and what are the consequences for people's perceived understanding, critical evaluation, and learning? With Tania Lombrozo and Sunnie Kim (Microsoft Research).
- Flexible reasoning in humans and LLMs: Using matched pairs of familiar riddles and “riddle riddles” that resemble them but require literal reasoning, we found opposite patterns of failure across nine LLMs and 100 adults: LLMs often applied inventive reasoning when it was unnecessary, whereas humans tended to rely on literal reasoning when an inventive solution was needed. Preprint · With Tom Griffiths.
- Moral deference to AI: When do people defer to AI recommendations in moral dilemmas, and how does their willingness to defer depend on their confidence, the extent to which the AI agrees with their own judgments, and the moral domain? Part of a multi-institution, DARPA-funded collaboration on algorithmic trust.
- AI in legal decisions: Examines how LLM-rewritten and LLM-generated legal arguments influence people's evaluations of evidence, judgments of guilt, and decision-making in legal contexts. With Dave Lagnado (UCL) and Tania Lombrozo.
- Validity of LLM judges as proxies for human evaluators: Examines whether LLM judges exhibit the same linguistic preferences and evaluation biases as humans, how these preferences affect their ability to evaluate reasoning, and whether they emerge or change through model post-training. With Tania Lombrozo and Tom Griffiths.
