OF THE
TIMES
Employing artificial intelligence
At the Online Hate Prevention Institute (OHPI), we have spent the past six years both tackling specific cases – including Bornstein’s – and working on the problem of measurement using world-class crowdsourcing and artificial intelligence approaches.
Others are also looking at identification and measurement as the next step. The Antisemitism Cyber Monitoring System (ACMS) – a new tool to monitor antisemitism on social media – has been under development by Israel’s Diaspora Affairs Ministry since October 2016. It will be launched at the 2018 Global Forum for Combating Antisemitism in Jerusalem later this month.
The tool uses text analysis – a form of artificial intelligence – and works by searching social media sites for words, phrases and symbols that have been identified as indicators of possible antisemitic content. The tool then reviews the content and generates interactive graphs.
Similar approaches have been used by the World Jewish Congress and by Google’s Conversation AI project, but the approach has limited effectiveness, particularly when applied to large social media sites.
Data from a one-month trial of ACMS was released ahead of the system’s launch. While the software is being promoted as a major step forward in the fight against cyberhate, the data itself highlights serious methodological and technological limitations making it more of a distraction.
Limitations of the technology
One limitation ACMS has is detecting abuse that uses the coded language, symbols and euphemisms that are increasingly favoured by the far right.
Another is that ACMS only monitors content from Facebook and Twitter. YouTube, which accounted for 41% of the online antisemitism identified in a previous report, is not included. The automated system also only monitors content in English, Arabic, French and German.
What’s more concerning is the Ministry’s claim that the cities that produce the highest volume of racist content were Santiago (Chile), Dnipro (Ukraine) , and Bucharest (Romania). These cities have primary languages the software is not programmed to process, yet they have somehow outscored cities whose primary languages the software does process.
Of particular concern to Australia is a graph titled Places of Interest: Level of Antisemitism by Location that shows Brisbane as the highest-ranked English-speaking city. This result has been explained by a later clarification suggesting the number is an amalgamation of global likes, shares and retweets that engaged with content originally posted from Brisbane. The data is therefore subject to a large degree of randomness based on which content happens to go viral.
Lawyers and data scientists must work together
There is a place for AI-based detection tools, but their limitations need to be understood. Text analysis can identify specific subsets of online hate, such as swastikas; language related to Hitler, Nazis, gas chambers and ovens; and antisemitic themes that are prominent among some far right groups. But they’re not a silver bullet solution.
They're the last specimens we'd waste energy on in a race to save the human species