Lunch Bytes Lecture Series
Lecture Series on Law, Technology & Society
Lunch Bytes: AI in the Admissions Room: What Happens When AI Both Writes and Evaluates Recommendation Letters?
The lecture series Lunch Bytes opens up an interdisciplinary dialogue between law, technology, and society. Mirac Suzgun (Stanford University) speaks on:
AI in the Admissions Room: What Happens When AI Both Writes and Evaluates Recommendation Letters?
Every year, US colleges ingest nearly 10 million letters of recommendation. These letters are a central narrative component of the holistic college admissions process. But as teachers increasingly adopt generative AI systems to draft recommendation letters and admissions offices deploy LLMs to evaluate them, applicants enter a "dual-AI" pipeline. While human-written letters have long exhibited systematic demographic asymmetries of their own in length, tone, and framing, it remains unknown whether cascaded AI systems attenuate, replicate, or amplify these biases. This talk presents early results from our ongoing systematic empirical investigation of the complete AI writer-to-evaluator workflow. By pairing identical candidate profiles with race- and gender-signaling names across controlled name-blinded and post-writing swapped conditions, we seek to isolate whether biases emerge during drafting, evaluation, or their interaction. Given recent legal shifts governing college admissions (SFFA v. Harvard), our findings offer empirical insights and outline potential safeguards for the algorithmic scoring and evaluation of applicants in higher education.
Information
Date: 04 November 2026
Time: 12:30-14:00
Venue: Seminarraum SEM 34 Schottenbastei 10-16, 3rd floor (Juridicum)
Registration is required. Please send your registration to georgia.pantazi(at)univie.ac.at no later than 3 November, 14:00 CET.
Speaker
Mirac Suzgun is a sixth-year PhD candidate in Computer Science at Stanford University, co-advised by Dan Jurafsky and James Zou, and holds a JD from Stanford Law School. His CS research explores the capabilities and limitations of large language models, focusing on reasoning, hallucination detection and mitigation, and societal applications. He also conducts legal scholarship on constitutional law, federal courts, and AI governance & regulatory policy, working closely with Daniel E. Ho at the Stanford RegLab. Previously, Mirac graduated from Harvard College with a joint degree in Mathematics and Computer Science and a secondary field in Folklore & Mythology, earning the Thomas T. Hoopes Prize for his senior thesis. His work has appeared in leading venues including Nature Machine Intelligence, The Lancet Digital Health, Journal of Legal Analysis, ACL, EMNLP, ICLR, and NeurIPS. His graduate studies have been supported by the Google PhD Fellowship, Stanford HAI-SAP Fellowship, and Stanford Law School Fellowship.
Format
The series takes place during lunchtime. In a relaxed setting, participants only need to bring their curiosity. Rather than the traditional lecture hall atmosphere, the format creates space for dialogue across disciplines. The presentations are concise, the language used is comprehensible and discussions are open. Each session lasts around 90 minutes and includes snacks, making the exchange not only intellectually but also physically nourishing. The events are free of charge and open to anyone interested in digital innovations and their governance.