Meta, the parent company of Facebook, has unveiled plans to implement more sophisticated AI systems for content enforcement, aiming to decrease its dependence on third-party vendors. This move comes as Meta seeks to enhance its capabilities in detecting and removing harmful content, such as terrorism-related posts, child exploitation, drug sales, fraud, and scams.
The decision to transition to advanced AI systems is based on Meta’s belief that these technologies can effectively identify violations with higher accuracy, combat scams more efficiently, promptly respond to real-world events, and minimize instances of over-enforcement.
According to Meta, the new AI systems will gradually replace existing content enforcement methods across its platforms once they consistently outperform the current processes. Simultaneously, Meta plans to reduce its outsourcing of content moderation tasks to third-party vendors.
Meta explained that the AI systems will handle tasks that are best suited for technology, including repetitive reviews of graphic content and addressing evolving strategies employed by malicious actors, such as illicit drug sales and fraudulent schemes.
Initial tests of the AI systems have demonstrated promising results. Meta reported that the new systems are capable of detecting significantly more instances of adult sexual solicitation content compared to human review teams, with a notable error rate reduction of over 60%. Furthermore, these AI tools are effective in identifying and preventing impersonation accounts linked to celebrities and other prominent figures, as well as thwarting account takeovers by flagging suspicious activities such as login attempts from unfamiliar locations or sudden profile changes.
Meta highlighted that the AI systems have successfully foiled approximately 5,000 daily scam attempts aimed at deceiving users into divulging their login credentials.
Meta emphasized that human experts will remain integral in supervising and evaluating the AI systems’ performance, particularly in making critical decisions regarding high-risk content.
Source: TechCrunch