Title

AWS re:Invent 2023 - Customer Keynote Thorn

Summary

Tanya Cushman from Thorn shared a story about a child named Maria who was rescued from sexual abuse due to Thorn's Safer products.
Safer uses machine learning, specifically a CSAM classifier, to detect child sexual abuse material (CSAM) on content hosting platforms.
In 2022, the National Center for Missing and Exploited Children received over 88 million files of suspected CSAM.
Thorn is a nonprofit that uses AWS services to power their machine learning tools for combating child sexual abuse at scale.
The classifier was trained on-prem due to the illegal nature of CSAM and then distributed using Amazon's ECR.
Amazon S3 is used to store non-abuse material, which is crucial for training the classifier.
Amazon EC2 and EKS are used for R&D with benign data.
Regular model maintenance is essential to address staleness and bias, with performance monitoring and retraining.
User feedback is collected via an API in AWS-hosted Safer Services to improve the classifier.
Deployment is privacy-forward and human-in-the-loop, ensuring that humans make the final decision on reporting.
Thorn's impact includes over 2.8 million potential files of CSAM found and the launch of Safer Essential for quick detection of known CSAM.
Tanya Cushman calls for collaboration among various stakeholders and the community to join Thorn's mission.

The use of machine learning and AI in the detection of CSAM is proving to be a critical tool in the fight against child sexual abuse.
The sheer volume of suspected CSAM files reported highlights the scale of the problem and the need for efficient and scalable solutions like Thorn's Safer.
Thorn's approach to training the classifier on-premises due to the sensitive and illegal nature of the data is a unique solution that respects legal constraints while leveraging cloud technology for distribution.
The importance of regular model maintenance to address bias and improve accuracy is emphasized, showing that machine learning models require ongoing attention and refinement.
Thorn's deployment strategy prioritizes privacy and human oversight, which is crucial in sensitive areas such as CSAM detection.
The call to action for collaboration and community involvement underscores the multifaceted nature of combating child sexual abuse, where technology is a part of the solution but not the sole answer.
The launch of Safer Essential indicates Thorn's commitment to constant innovation and providing tools that can be quickly integrated and used by various platforms to protect children.