Online Harms Observatory: Real-Time Analytics for Internet Safety - Session 4

Overview

Explore the Online Harms Observatory, a groundbreaking analytics platform developed by The Alan Turing Institute's Public Policy programme, in this 55-minute session from AIUK 22. Learn how this innovative tool combines large-scale data analysis and advanced AI to provide real-time insights into online harmful content. Discover the platform's potential to assist policymakers, regulators, security services, and civil society stakeholders in understanding the landscape of online harms. Gain insights into the Observatory's methodology, including AI training techniques, active learning, and adversarial data generation. Examine the first tracker, a live monitor of abuse directed at Premier League football players on Twitter. Join Dr Bertie Vidgen, Angus R Williams, and Hannah Kirk from the Alan Turing Institute as they discuss the need for such an observatory and its aims in addressing online safety challenges.

Syllabus

Intro
The Online Safety Team
The Online Harms Observatory
Methodology for OHO
How do we train AI?
Active learning
How does the model choose entries?
Adversarial generation of data
Research design
Test set: model accuracy
Test set: model F1
Model results: Precision-Recall Curves
Alan Turing Institute