Your data is being manipulated.
summary
Argues that training data is both biased and actively targeted: shows ImageNet dataset artifacts and Latanya Sweeney’s finding that Google returned criminal-record ads disproportionately for Black-sounding names, documents popular use of Twitter and Reddit open APIs to train models, and highlights research that crafts adversarial stop-sign attacks which humans still read as STOP but classifiers see as YIELD. Concludes systems are already being gamed, attackers are iterating, and simple filtering does not stop strategic data poisoning.
AI-written summary · about
related talks
46mA Short Introduction to Hacking Artificial Intelligence
21mAdversarial Attacks
35mHacking AI Systems: How to (Still) Trick Artificial Intelligence
25mTricking Neural Networks : Explore Adversarial Attacks
42mIs your AI plotting to kill you? Why AI and Machine Learning is dangerous
44mComputers are Stupid: Protecting "AI" from Itself
42mHow to attack AI systems (and how to defend them) !!!!
39mSecure Machine Learning: Overcoming Algorithmic Bias, Adversarial ML in the Wild West of AI