Embedded Files
October 28, 2024
8AM–12:30PM PT
11AM–3:30 PM ET
General-purpose AI systems are now deployed to billions of users, but they pose risks related to bias, fraud, privacy, copyright, CBRN, NCII, and more. To assess these risks, we need independent and community-driven evaluations, audits and red teaming, and responsible disclosure.
Our workshop on the future of third-party AI evaluation dives into these topics with experts on:
Third-party evaluations, red teaming, and disclosure in the real world
How to design technical infrastructure for identifying and reporting flaws in AI systems
The legal and policy infrastructure for advancing a healthy AI evaluation ecosystem
You can watch the event on YouTube here.
Speakers
Rumman Chowdhury
CEO,
Humane Intelligence
Ilona Cohen
Chief Legal and Policy Officer, HackerOne
Nicholas Carlini
Research Scientist, Google DeepMind
Lauren McIlvenny
Technical Director,
CERT Threat Analysis Director, AISIRT
Harley Geiger
Coordinator,
Hacking Policy Council
Amit Elazari
Co-Founder and CEO, OpenPolicy
Victoria Westerhoff
Director, AI Red Team, Microsoft
Avijit Ghosh
Applied Policy,
Hugging Face
Schedule
Opening Remarks, Keynote
8 AM – 8:40 AM PT
Why did we organize this workshop?
In our recent paper, A Safe Harbor for AI Evaluation and Red Teaming, we propose protections for third-party AI researchers.
Hear from the organizers about why we convened this group.
Session 1: In Practice
8:40 AM – 9:50 AM PT
Speakers will discuss third-party evaluations, audits, and red teaming in practice.
What types of evaluations are most informative?
What methods are most effective?
How are flaws systematically identified?
What are the key challenges in reporting these flaws?
Session 2: By Design
10:00 AM – 11:10 AM PT
Speakers will discuss the design of infrastructure for third-party AI evaluation, including vulnerability reporting.
How should reporting of AI flaws differ from reporting of security bugs?
How should disclosure be scoped and documented?
How are real flaws substantiated?
Session 3: Law & Policy
11:20 AM – 12:20 PM PT
Speakers will discuss policy incentives for and barriers to third-party AI evaluation.
How can third-party AI research be legally protected?
What benefits does such research have from a government perspective?
How can the AI community get involved in law and policy around these issues?
The workshop was held online on October 28, 2024 from 8 AM – 12:30 PM PT. Each session includes invited talks from experts and panel discussion.
Organizers
Kevin Klyman
Stanford University
Princeton University
Stanford University
Michelle Sahar
OpenPolicy
Arvind Narayanan
Princeton University
Rumman Chowdhury
Humane Intelligence
Percy Liang
Stanford University
Questions? Contact slongpre@media.mit.edu or sayashk@princeton.edu
Google Sites
Report abuse