Anticipates Very High Processing Power Requirements
One of the largest computational costs of AI-based NSFW detection is the requirement of serious computational power. AI models (particularly deep learning models) need a large amount of computational resources to analyse, and infer the image and video processing in real-time. One popular example is the typical deep learning model for image recognition based on deep convolutional neural networks, which requires (say) 50 TFLOPS (trillion floating operations per second) to run comfortably on modern hardware. This sort of raw processing muscle requires some serious GPUs, ranging from five grand for mid-range models on up to $10,000 or more for top-of-the-line versions.
Costs of Data Storage and Management
In connection the data storage is second the major role player in the computational requirements. Large datasets of millions of images and videos are used to train AI models for NSFW detection. It needs robust infrastructure to store this data. The new going rate for keeping one petabyte of data on a Tier 1 enterprise cloud as of 2023 is ~$23,000 per month. In addition to this, there is the operational overhead of managing this data, making it accessible, and backing it up securely.
Energy consumption and environmental concerns
Powerful AI detection systems are also energy-consumptive. To train a standard AI model for NSFW detection, the energy consumption is said to be approximately equivalent to the carbon footprint of 5 cars over their lifetimes. For orders of magnitude larger tech companies this can mean annual electrical bills in the millions of dollars, which can make energy efficiency quite a serious concern.
Cost of Maintenance and Upgradation
The continuous maintenance and the eventual upgrade of AI models also add to the price. It is important that AI systems are updated regularly to learn and detect novel types of unsafe (NSFW) content to increase their accuracy. Any new update means, the model will again have to be trained with new data which in turn means more computational resources and manpower are required. The cost of these updates changes immensely, however, usually end up spending at least tens of thousands of dollars per year supporting each model at scale, based on the model complexity and update rates.
Million Mile Warranty and Economic Rationale
Nevertheless, even though these systems are expensive, they make a lot of financial sense for the bigger platforms to invest on. These systems make it technically more innocent to detect content, which means that an extensive human moderation team - expensive in the long run - may not be as necessary. For instance, automating NSFW screening can save a major platform up to 70% in human moderation costs, equivalent to millions of dollars of savings per year.
To learn more about how AI is revolutionizing the detection and moderation of NSFW content, welcome to nsfw character ai.