X (formerly Twitter) has introduced an advanced artificial intelligence system designed to identify and manage NSFW (Not Safe For Work) content on its platform, significantly reducing the need for human moderation. This development comes as part of Twitter’s broader strategy to enhance user safety and content moderation through cutting-edge technology.

The initiative began with the acquisition of Madbits, a startup specializing in AI and machine learning, founded by Clément Farabet. Madbits had developed sophisticated systems capable of identifying objects in photos and videos. Under the leadership of Twitter’s head of engineering, Alex Roetter, Madbits was tasked with creating an AI system specifically to identify pornographic and other objectionable images on Twitter.

After a year of development, the AI system is now operational. Farabet explains that the system can identify about 99% of all pornographic and objectionable images, with a false positive rate of only 7%. This allows Twitter to warn users with interstitials or warning screens before displaying potentially inappropriate content. This technology represents a significant advancement for social media platforms struggling with content moderation.

Traditionally, companies like Twitter and Facebook have relied on human moderators to review and flag inappropriate content. This work often places a significant mental and emotional burden on the workers, many of whom are based in countries like the Philippines. The new AI system can alleviate much of this burden, offering faster and cheaper moderation without the associated human toll.

The AI capabilities developed by the Madbits team extend beyond content moderation. The technologies, now part of a central AI operation called Twitter Cortex, are being used to enhance various aspects of the platform. These include identifying users to follow, curbing spam and abuse, and improving the relevance of tweets, ads, and other content displayed to users.

Roetter notes that Twitter Cortex mirrors similar efforts at other tech giants like Google and Facebook, which use deep learning algorithms for tasks such as facial recognition, speech recognition, and real-time translation.

While the AI system shows promise, it is not without challenges. Identifying NSFW content involves nuanced distinctions that can be difficult for algorithms to make, such as differentiating between nudity in an artistic context versus pornographic material. David Luan, a specialist in photo identification for other companies, highlights the complexity of this task, especially given the need for real-time processing on a platform as large as Twitter.

Despite these challenges, Twitter’s investment in AI technology underscores its commitment to improving user experience and safety. By automating the identification of inappropriate content, Twitter can provide a more secure and enjoyable platform for its users.

Twitter’s efforts are part of a larger trend among tech companies to leverage AI for content moderation. With the acquisition of AI startup WhetLab, Twitter aims to further refine its neural networks using techniques like “Bayesian optimization,” which allows for continuous improvement of the AI models.

This approach, described by WhetLab founder Ryan Adams as using “machine learning to improve machine learning,” promises to enhance the efficiency and effectiveness of AI systems across various applications.

Twitter’s new AI system for identifying NSFW content represents a significant step forward in the realm of content moderation. By reducing reliance on human moderators and leveraging advanced machine learning techniques, Twitter is setting a new standard for safety and user experience in social media. This development not only addresses immediate concerns but also paves the way for future innovations in AI-driven content management.