Even the most autonomous NSFW AI systems still require quite a bit of human intervention. By 2024, more than 70% of the largest social media platforms will use AI to filter out nudity and other explicit content. In Facebooks case, their AI handles 15m posts a day and does so with as little human oversight as possible. Even so, about one in ten of flagged content is still reviewed by human moderators to ensure accuracy and context.
AI technologies like machine learning models use large datasets to recognize inappropriate content. For example, Google’s AI has been taught with over 100 million images to boost the accuracy in their content moderation. Even though those systems can efficiently analyze visual media, there are still misclassifications happening. In 2023 YouTube announced a case where its AI mis-applied to wrongly screen out as unsafe, there are c…
According to industry experts such as Dr. Mark Thompson from MIT, “AI systems can process content quickly but lack the detailed knowledge that human moderators provide.” This echoes how hybrid models work today: with the AI functioning as an extended tier one solution doing much of filtering, human-powered supervisory roles making final decisions for corner-cases and preventing errors.
And it can be costly even in financial terms to implement fully autonomous NSFW AI systems. Advanced systems, it turns out, cost companies like a cool $5 million per annum for an AI-based content moderation. That said, it saves up to 30% of overall moderation costs compared with manual processes helping augment human labor. That being said, fully automating this process remains difficult as a consequence of the restrictions adherence to current AI technologies.
However, incidents in real world applications bring out these problems. That led the company in 2022 to increase human review after its AI was added too slow and flagged about a fifth, or 18%, of content incorrectly. The prior example brings to light how humans consistently need be current with AI in order not to rely solely upon it.
Overall, the lessons learned is that NSFW AI systems can filter at scale even if they are not perfect on certain types of content; however human review will still be necessary to handle cases where things don’t behave as expected. nsfw ai technologies are continuing to advance in ways that strive for the best of both worlds — automation when possible, and human understanding as a form of control for managing content.