#algorithmicfairness#impossibilitytheorem#aiethics
8.2K subscribers • 4 videos on Hallway
Intermediate. Assumes working knowledge.
1.5K views7 years ago2 recommendations
AI and machine learningComputer scienceAI safety and impactComputation and complexity
Chris Stucchio explains the mathematical impossibility of simultaneously satisfying multiple fairness criteria in algorithmic decision-making systems.
In analytical philosophy, armchair philosophers ask theoretical questions like "is it wrong to push one fat man onto train tracks in order to stop a trolley from smashing into 5 italian grandmothers?" As AI ethics has become a concern, these problems have suddenly become practical and quantitative. In this talk, I'll present some western liberal ethical principles that are important to algorithmic decision making. I'll provide a number of examples in lending, criminology and education which illustrate how it's impossible to simultaneously satisfy all these ethical principles. If time permits, I'll also discuss how these western principles are far from universal, and how most of the literature on the topic is relatively useless in non-western contexts (e.g., most literature focused on the US takes N=2, I don't even know how to count N in India). - Captured Live on Ustream at ustream.tv/crunchWhat people say. Engineers recommend this talk for its rigorous, mathematical framing of the inherent trade-offs in algorithmic fairness.
dvtrnon Hacker News • Recommends
On “Alexandria Ocasio-Cortez Is Absolutely Right About Racist Algorithms”
PeterisPon Hacker News • Recommends
On “Who Should Stop Unethical A.I.?”
joshmarlowon Hacker News • Mentions
On “Building Inclusive AI at Facebook”
nostrademonson Hacker News • Mentions
On “600k Images Removed from ImageNet After Art Project Exposes Racist Bias”
missosoupon Hacker News • Mentions
On “New York Regulator Probes UnitedHealth Algorithm for Racial Bias”
missosoupon Hacker News • Mentions
On “Why are we using black box models in AI when we don’t need to? (2019)”