Title

AWS re:Invent 2023 - Harnessing 110 years of insights: Using AI in content classification (PRO303)

Summary

The British Board of Film Classification (BBFC) has been providing age ratings for over 110 years, adapting to changes in media consumption from cinema to online streaming.
The BBFC, in collaboration with AWS, is developing an AI solution to scale human compliance processes for content classification.
The AI proof of concept is trained on a rich archive of video data and metadata to identify compliance issues and their severity levels, aiding human moderators.
The AI tool does not directly produce age ratings but supports human decision-making, aiming to streamline processes, reduce costs, and provide consistent age labeling online.
AWS ProServe and the BBFC have built a model that considers video, audio, and text components of content, with special considerations for language, animation, profanity, and sound.
The model has been trained on over 1,600 movies and takes 8-12 hours to train, with an MLOps pipeline for continuous improvement and expansion to other categories.
The AI model has shown proficiency in identifying no issues and extreme cases but struggles with nuances between moderate and strong issues due to lack of context.
Future work includes model tuning, expanding the training dataset, cost optimization, adapting to different styles like animation, and understanding model decisions better.

The BBFC's move to AI-assisted content classification reflects the broader industry trend of leveraging machine learning to handle large volumes of data efficiently.
The AI model's ability to identify no issues and extreme cases with high accuracy but struggle with nuanced cases highlights the current limitations of AI in understanding context and subtleties.
The use of an MLOps pipeline indicates a commitment to continuous improvement and adaptability, which is crucial in the fast-evolving field of AI and content classification.
The collaboration between a traditional organization like the BBFC and a tech giant like AWS demonstrates the potential for technology to enhance and extend the capabilities of established institutions.
The focus on an AI tool that aids rather than replaces human decision-making underscores the importance of human expertise in areas requiring nuanced judgment.
The project's approach to handling different content types, such as animation and non-English language films, suggests a comprehensive and inclusive strategy for content classification that can cater to diverse media.
The session's emphasis on protecting children and vulnerable groups online aligns with broader societal concerns about the impact of digital media consumption on mental health and well-being.