Artificial Intelligence endangers evidence of human rights violations

Social media icons Social media icons Indipendent.ie

19 June 2020

Erroneous video takedowns by automatic software risks to delete social media documentation of human rights abuses

This month in Syria, more than a dozen videos have been mistakenly removed from the Youtube channel of the student journalist Baraa Razzouk, who documents Syrian events, including protests and attacks by regime forces. However, Youtube is not the only social media that is experiencing erroneous content removal. Also Facebook and Twitter have mistakenly taken down content due to the replacement of Artificial Intelligence (AI) for policy control in the COVID-19 pandemic emergency.

Youtube's policy foresees the elimination of videos with pro-terrorist content. The most common mistake of the removals would be due to the inability of the automatic software to distinguish between terrorist content and human rights documentation. As the researcher Jeff Deutch said, "AI is notoriously context-blind," which means it is unable to assess the historical, political or linguistic purpose of the video.

According to Syrian Archieve, the rate of video takedowns of Syrian human rights violations on Youtube has risen from 13% to 20% in 2020. Many international organisations, including Amnesty International, accuse the suppression of video documentation which seems to be common also in other war-torn countries, like Yemen or Sudan. These data are of vital importance because they can also be used as evidence for crimes against human rights by the International Criminal Court and the United Nations.

 

To know more, please read:

https://www.reuters.com/article/us-global-socialmedia-rights-trfn/lost-memories-war-crimes-evidence-threatened-by-ai-moderation-idUSKBN23Q2TO

https://syrianarchive.org/en/tech-advocacy/september-takedowns.html

 

Author: Carla Pintor

Read 138 times