Перейти к содержимому

Data, AI and Predictive Tools in Profiling: ViCAP, PredPol and Genetic Genealogy | Ep 31

Joshua Moses | Cybertrace Foundation

0:00 / 0:00

Data, AI and Predictive Tools in Profiling: ViCAP, PredPol and Genetic Genealogy | Ep 31

7 просмотров · 12 дн. назад
Joshua Moses | Cybertrace Foundation
210 подписчиков
7 просмотров · 12 дн. назад
Algorithms, massive databases, and even our own biology are changing how crimes get solved and predicted. In this episode we look at what these tools can do and the human flaws that can undermine them. Part one looks at the case linkage databases built to connect violent crimes across jurisdictions, including ViCAP in the US and ViCLAS in Canada and the UK. Participation is mostly voluntary, ViCAP captures only a small share of violent crimes, and analysts have sometimes had to enter cases from newspaper clippings. Garbage in, garbage out. Part two covers predicting with machines. We compare clinical prediction (gut feel) with actuarial prediction (statistics and machine learning) and walk through automated linkage: behavioral checklists, base rates, and weighting rare behaviors. Some models report accuracy as high as 90 percent, but they were built on solved cases, which may inflate how well they would do on messy unsolved ones. Part three tackles predictive policing. PredPol uses location and time data to tell officers where to patrol, and early use in Santa Cruz was linked to a drop in burglaries. We also explain Bernard Harcourt's "ratchet effect," the feedback loop where more policing produces more arrests that justify even more policing. Part four looks at biology overtaking traditional profiling: epigenetics, forensic DNA phenotyping, and microbiome analysis that may hint at appearance, habits, and travel history. Part five lays out guardrails: preset variables to stop data dredging, transparency and auditability instead of secret databases, independent peer review, and courts refusing to let testimony hide behind an "aura of science." CHAPTERS 0:00 Introduction: Data, Algorithms and Biology 0:29 Roadmap 0:51 Part 1: Case Linkage Database Flaws 1:38 Garbage In, Garbage Out 2:34 Part 2: Predicting Profiles With Machines 3:22 How Automated Linkage Works 4:00 The 90 Percent Accuracy Claim 4:18 Validation Limits 4:49 Part 3: Predictive Policing and Bias 5:39 The Ratchet Effect 6:38 Part 4: Genetic Genealogy and Phenotyping 7:28 What a Drop of Blood Can Reveal 7:58 Part 5: Establishing System Guardrails 8:54 The "Aura of Science" Warning 9:21 Final Thought