Greške umjetne inteligencije: od podataka do pogreške

Dokumentarac prati stvarne slučajeve u kojima su sustavi strojnog učenja donijeli pogrešne ili štetne odluke: od lažnog uhićenja na temelju prepoznavanja lica, kroz pristranu selekciju životopisa i netočne medicinske preporuke, do algoritmskog određivanja rizika od recidivizma. Gledatelj razumije kako demografska neravnomjernost u podacima za učenje postaje sustavna pogreška modela, zašto interni pokazatelji izvedbe mogu biti zavaravajući i zašto pravna odgovornost za pogreške AI sustava ostaje nedefinirana. Poglavlja 00:00 Čovjek na koga je pokazao algoritam 03:10 Sirovine: što AI konzumira 08:01 Kako stroj uči: gubitak, gradijent, prilagođavanje 12:04 Amazonov algoritam za zapošljavanje: pogrešni učitelj 15:48 COMPAS: kada algoritam sudi 20:57 Mjerni propust: Epicov model sepse 25:58 Rodne sjene: isti algoritam, različiti podaci 32:56 IBM Watson i rak: pogrešni savjet 38:05 Tko odgovara: nadzor, propis i otvorena pitanja 42:38 Povratak k Robertu Williamsu Izvori - ACLU of Michigan — I Was Wrongfully Arrested Because of Facial Recognition Technology. It Shouldn't Happen to Anyone Else: https://www.aclumich.org/news/i-was-wrongfully-arrested-because-facial-recognition-technology-it-shouldnt-happen-anyone-else/ - PBS NewsHour — detroit police challenged over face recognition flaws, bias: https://www.pbs.org/newshour/amp/nation/detroit-police-challenged-over-face-recognition-flaws - AI Incident Database — Incident 74: Detroit Police Wrongfully Arrested Black Man Due To Faulty FRT: https://incidentdatabase.ai/cite/74/ - ACLU — Williams v. City of Detroit: https://www.aclu.org/cases/williams-v-city-of-detroit-face-recognition-false-arrest - PBS NewsHour — Detroit Police Challenged Over Face Recognition Flaws, Bias: https://www.pbs.org/newshour/amp/nation/detroit-police-challenged-over-face-recognition-flaws - CBS News / 60 Minutes — Police departments adopting facial recognition tech amid allegations of wrongful arrests: https://www.cbsnews.com/amp/news/facial-recognition-60-minutes-2021-05-16 - Civil Rights Litigation Clearinghouse, U-M Law School — Williams v. City of Detroit 2:21-cv-10827 (E.D. Mich.): https://clearinghouse.net/case/44401/ - Hoodline — Detroit Police Overhaul Facial Recognition Policy After Wrongful Arrest; Settles for $300,000: https://hoodline.com/2024/07/detroit-police-overhaul-facial-recognition-policy-after-wrongful-arrest-settles-for-300-000/ - ACLU — Civil Rights Advocates Achieve the Nation's Strongest Police Department Policy on Facial Recognition Technology: https://www.aclu.org/press-releases/civil-rights-advocates-achieve-the-nations-strongest-police-department-policy-on-facial-recognition-technology - Exposing.ai — Exposing.ai: Adience Benchmark: https://exposing.ai/adience/ - University of Central Florida, Complex Adaptive Systems Laboratory — LFW: Labeled Faces in the Wild – Complex Adaptive Systems Laboratory: https://complexity.cecs.ucf.edu/lfw-labeled-faces-in-the-wild/ - MIT News — Study finds gender and skin-type bias in commercial artificial-intelligence systems: https://news.mit.edu/2018/study-finds-gender-skin-type-bias-artificial-intelligence-systems-0212 - ACM FAT* Conference — ACM FAccT Conference 2018: https://facctconference.org/2018/ - Proceedings of Machine Learning Research (PMLR) — Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification: https://proceedings.mlr.press/v81/buolamwini18a.html - MIT DSpace / Massachusetts Institute of Technology — Gender Shades: Intersectional Phenotypic and Demographic Evaluation of Face Datasets and Gender Classifiers (MIT Master's Thesis): https://dspace.mit.edu/handle/1721.1/114068?show=full - Google — Linear regression: Loss — Machine Learning Crash Course: https://developers.google.com/machine-learning/crash-course/linear-regression/loss - Google — Linear regression: Gradient descent — Machine Learning Crash Course: https://developers.google.com/machine-learning/crash-course/linear-regression/gradient-descent - Google — Linear regression: Hyperparameters — Machine Learning Crash Course: https://developers.google.com/machine-learning/crash-course/linear-regression/hyperparameters - Cambridge University Press / d2l.ai — Dive into Deep Learning — Chapter 1: Introduction: https://d2l.ai/chapter_introduction/index.html - Nature Publishing Group — Learning representations by back-propagating errors: https://www.nature.com/articles/323533a0 - Google — Machine Learning Glossary: ML Fundamentals: https://developers.google.com/machine-learning/glossary/fundamentals - IBM — What is Overfitting? — IBM Think: https://www.ibm.com/topics/overfitting - … Napomene - Naracija: sintetski (AI) glas. Glazba: originalna, sintetizirana za ovaj video. - Sheme su pojednostavljeni prikazi. #dokumentarac #umjetnainteligencija #algoritamskapristranost

Pogledaj na YouTubeu