[Colloquium] Designing Human-Centric AI with Use in Mind - Thursday, October 31st

[Catherine Clemons via Colloquium]

UChicago/TTIC NLP Seminar Presents


Katie Collins

University of Cambridge


Thursday, October 31st

2pm-3pm

In Person: John Crerar Library 298

Zoom: https://uchicago.zoom.us/j/97708787366?pwd=qD9KdtzwGCTPuciAfr6iaap3NubZgy.1


Title: Designing Human-Centric AI with Use in Mind


 Abstract: Tremendous advances in AI system capabilities are already unlocking new possibilities for human-AI complementarity – but have we really built machines that reliably and consistently meet our expectations and complement our limitations? In this talk, I will focus on the importance of interactive evaluation, with humans - and particularly expert humans - to uncover new system vulnerabilities. In particular, I will discuss the challenges of “surprising fallibility” in system performance for safe and sensible use of AI systems. Motivated by these challenges, I will then present early work exploring modular small tweaks to user interfaces (“frictions”) to encourage more effective and appropriate access and uptake of AI outputs. I will close with some forward-looking propositions for the design of more human-centric AI “thought partners” (and infrastructures around their use) which can safely, securely, and effectively be deployed alongside us. I will present some preliminary work that explores the marriage of large language models and probabilistic programming to move towards a system that is closer to meeting some of these goals.


 Bio: Katie is a PhD student at the University of Cambridge in the Machine Learning Group supervised by Adrian Weller and a Visiting Student in Josh Tenenbaum's Computational Cognitive Science Group at the Massachusetts Institute of Technology (MIT). Her research centers around applied computational cognitive science and data-efficient human-AI interaction. She received an MPhil in Machine Learning and Machine Intelligence from the University of Cambridge and a BSc in Brain and Cognitive Science from MIT, with minors in Computer Science and Biomedical Engineering. She is also affiliated with the Human-Oriented Automated Theorem Proving Effort led by Sir Tim Gowers and is a Student Fellow at the Leverhulme Centre for Future Intelligence. Additionally, she is the founder of the MITxHarvard Women in AI Group, a Class of 2021 Marshall Scholar, and a recipient of a Cambridge Trust Scholarship.


[Katie Collins]

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20241028/186d28aa/attachment.html>