Project Description
Artificial intelligence is becoming increasingly widespread, with hundreds of millions of users worldwide. Recognising the potential risks of AI-generated hate speech, this project investigated the prevalence of AI-driven antisemitism on social media, evaluated the potential harm to users, and developed policy recommendations for government officials and regulators to mitigate the spread of harmful content.
The research comprised a comprehensive analysis involving thecollection and detailed annotation of 50 images and one video. By developing the first multidimensional annotation scheme for antisemitic deepfakes, the study considered elements such as form, content, and discourse, thus advancing existing frameworks in this field.
Outputs
- 50 AI-generated images and 1 deepfake video were identified and analysed.
- 5 trained classifiers were assessed for their accuracy in distinguishing between real and fake images.
- The research resulted in seven policy recommendations designed to offer a comprehensive framework for regulating AI-generated content. The researchers aim to engage with policymakers in the UK and the European Union to present the findings and propose these actionable recommendations.
Findings
- The research demonstrates that antisemitic content can be easily created using AI technologies, even with existing safety mechanisms in place.
- 50% of the images contained antisemitic stereotypes portraying an ‘evil’ Jew. Many of these images have the potential to spread harmful disinformation that incites hatred against Jews.
- Despite the prevalence of AI-generated antisemitic content online, there is a significant lack of effective detection mechanisms. Most current classifiers exhibit poor performance when identifying and analyzing antisemitic deepfakes.
- The findings underscore the urgent necessity of making generative AI technologies safer by preventing the creation and dissemination of illegal discriminatory content through enhanced transparency, moderation, and awareness.