Panel on Interpretability in the Law
summary
Panel documents specific failures and partial fixes: Facebook ad delivery shows discriminatory outcomes while LinkedIn results rely on a workaround for missing gender data; Domino’s vs Instacart studies flag unobserved confounders like pay and policies. Law and policy lack an independent validator for criminal‑justice ML; NIST and NHSTA offer limited models. Census’s shift to differential privacy exposes tradeoffs between transparency and verifiability. Legal barriers (trade secrecy, CFAA, contracts) and business incentives block full audits, so the panel urges platform‑assisted, adversarial auditing plus statutory safeguards and funding for independent research.
AI-written summary · about
chapters (9)
- 0:00Start
- 7:48The Formalization Factor
- 15:01The Absence of a Formal Solution Is Also Something That Legal Actors Can Exploit for Nefarious Ends
- 16:26The Societal and Individual Impacts of Algorithms
- 17:16The Societal Impact of these Algorithms
- 25:15If There Was an Independent Agency That Reviewed and Validated Ml Systems Used in the Criminal Justice System
- 31:25Penalties in the Law
- 40:31The Knowledge Required from Judges
- 53:05What Do You Think Are the Major Properties of Social Media Algorithms That Should Be Covered Um in an Outside Audit and What Kinds of Entities Do You Think Are Best Situated
related talks
55mPanel: Main legal risks and challenges with LLMs / generative AI models
01h 20mUnlocking societal benefits of AI with privacy protective technologies.
56mEthics in Machine Learning Panel
01h 19mAcademic sessions 1.
43mPanel: Privacy Design Patterns for AI Systems: Threats and Protections
01h 12mInvestigating algorithmic systems - Algorithm auditing.
46mAlgorithmic Transparency and Accountability
32mData Ethics Panel