摘要
以扩散模型和经扩散模型训练而成的海量的、个性化的预训练模型和微调模型为代表的图像生成式AI在图像生成、修改与替换等领域一枝独秀。核心技术开源、“一键式”便捷操作并且兼顾“质”与“量”是现阶段图像AI生成的重要特征。作为生成式AI在淫秽图像制作领域的产物,淫秽图像不仅数量上泛滥,淫秽程度和引发的风险也持续提升,继而使得技术滥用的危害不断扩大。而预训练模型和微调模型正是目前图像生成式AI被广泛应用于淫秽图像制作领域的关键,现行行政法规将具有图像生成能力的模型及相关技术定义为生成式人工智能技术,根据部分预训练模型和微调模型的特点以及在虚拟淫秽图像生成过程中所发挥的作用,应当承认直接认定为淫秽电子信息的合理性,从而摆脱技术中立原则的干扰,更好地遏制特定内容淫秽图像的泛滥、维持网络信息的健康和安全。
Image generation AI, which is represented by diffusion model and massive, personalized pretraining model and finetuning model trained by diffusion model, is unique in the fields of image generation, modification and replacement. Open source core technology, “one-click” convenient operation and taking into account “quality” and “quantity” are important features of image AI generation at this stage. As a product of generative AI in the field of obscene image production, obscene images are not only flooded in number, but also the degree of obscenity and the risk caused by it continue to increase, which makes the harm of technology abuse continue to expand. The pretraining model and finetuning model are the key to the widespread application of image-generating AI in the production of pornographic images. The current administrative regulations define models and related technologies with image generating ability as generative artificial intelligence technologies. According to the characteristics of some pretraining models and fine-tuning models and their roles in the generation of virtual pornographic images, We should recognize the rationality of directly identified as obscene electronic information, so as to get rid of the interference of the principle of technical neutrality, better curb the proliferation of pornographic images of specific content, and maintain the health and safety of network information.
出处
《争议解决》
2023年第6期3418-3427,共10页
Dispute Settlement
