现代人工智能 (AI) 的迅猛发展, 使寓言性的“人工生命”神话具象化为现实技术, 催生出一种新的视觉生态与媒介环境。不过, 影像生成机制也感染了AI影像的神经回路, 这种感染体现为训练集里偏见、碎片与意识形态残留所导致的媒介病理性症候, 进而引发一场前所未有的公共卫生焦虑。通过分析ImageNet训练集、特雷弗·帕格伦的艺术作品以及Sora的影像生成实验, 还可进一步见出, 算法内部自我生成的“合成数据”携带某种结构性缺陷, 使模型陷入“过拟合”的幻觉再生产。我们仿佛被卷入一个由“幻觉失乐园”缔造的舒适的恐怖谷中。在此境况下, 对AI逻辑的抵抗显得愈发无力, 而尚未被算法污染的“净土”亦将在可预见的未来难以寻觅。

The rapid development of modern artificial intelligence (AI) has transformed the allegorical myth of “artificial life” into tangible technological reality, giving rise to a new visual ecology and media environment. However, the very mechanisms of image generation have also infected the neural circuitry of AI-generated images. This infection manifests as a kind of media pathology caused by the biases, fragments, and ideological residues embedded in training dataset, which further generates an unprecedented public health anxiety. By analyzing the ImageNet training set, the artwork of Trevor Paglen, and image generation experiments with Sora, it becomes apparent that the “synthetic data” self-generated within algorithms carries inherent structural flaws, which drive the models into a state of overfitted hallucinatory reproduction. We thus seem to be caught up in a comfortable valley of terror created by the “hullucinated paradise lost”. In such a scenario, resistance to AI logic appears increasingly futile, and any “pure zone” untainted by algorithmic contamination will likely be difficult to find in the foreseeable future.