Meta accelerates shift to AI content moderation

Meta is accelerating the deployment of generative AI (Gen AI) to replace human content moderation across its platforms. This year, the social media group has substituted large language models (LLMs) for approximately 50% of human review requests, with plans to reduce human involvement by more than 90% for specific content types by year-end.

The shift coincides with a broader cost-cutting initiative to offset the high costs of artificial intelligence (AI) infrastructure spending. While staff report that the company recently transitioned from Google's Gemini to its own foundational model, Muse Spark, executives state that the increasing pace of change is driven by efficacy rather than savings. Meta claims internal testing shows that large language models commit 13% fewer errors than humans and identify 10% more violations.

However, some employees warn that the rapid rollout lacks sufficient oversight, noting that the AI technology continues to make errors by incorrectly removing or hiding perfectly acceptable posts from feeds, hashtag searches, and recommendation algorithms, a practice known as 

shadow-banning. In contrast, supporters argue the transition enhances response speeds, broadens language coverage, and mitigates the mental health risks human reviewers face when exposed to graphic material. Meta maintains that robust governance remains in place, though human oversight will continue via sample testing.

£ - These articles require a subscription.

A version of this article is available without subscription in investing.com.

Training Announcement: Find out more about our range of independent accredited qualifications from BCS and IAPP. These include professional certifications for practitioners covering data protection and AI governance, information security and risk management, along with access to public information.