Over a 48-hour period at the Human Rights DiploHack February 29, we experimented and innovated on projects and challenges presented by the Office of the High Commissioner for Human Rights. The results were presented during #HRC31.
Big verification challenge: human rights violations
Our team, Pictrue, focused on collecting and verifying evidence of human rights violations. Every day human rights are violated. About 7 billion people, with 3 billion smart phones, are potential eyewitnesses to these violations.
Does a photo or video tell the ‘truth’? Methods and tools for verification of visual content are constantly improving, but the amount of content requiring verification for Human Rights purposes increases over-proportional – limiting its usability to a small and biased selection. Can improved creation, filtering, sorting and verification contribute to turn the tables on this volume problem?
We developed a ‘Self-learning Assessment Framework for Rights’ to quickly assess whether human rights violations can be identified in photos or videos and, if so, also identify the nature of the violation. Thanks to machine learning, this tool continuously improves its ability to filter, sort and verify photos and other data.
We established databases with each 500 pictures of child labour, child soldiers, and children in classroom. In our demo presentation on the next day, it was already able to tell with a very high certainty if criteria, such as child labor violations, were detected (about 10.000 pictures of one topic are needed to reach ‘maximum’ certainty).
Even though not all human rights violations are captured in pictures, we see a high potential in this tool. This preselection of incoming data would enable the OHCHR to process more data, faster.
The data load of human rights violations will increase, as more and more people use smart phones for reporting and are aware of these violations, so we think there’s a great need for this tool.
Project Manager FSU