生成式AI在餐厅菜单设计中的应用暴露出一个棘手问题——算法生成的食物图像呈现出不自然的单调性和光滑感，让消费者感到不适1。这种现象源于AI模型在训练数据中学到的狭隘审美标准1。Reality Defender首席技术官Alex Lisle用一个生动的比喻描述了这一困境："这就像外星人在制作披萨而不理解其核心原则"1。AI系统往往会生成"完美圆形冰淇淋球"和"遗传改良虾咬尾巴"这类过度规则化的食物形象1。
问题在于图像的反复编辑与重处理加剧了这种劣化。Lisle解释了其中的机制："当你把一个模型的输出反馈到自身，最终近亲繁殖会过度，整个系统崩溃"1。这种被称为"收敛"的现象已经有用户实际验证——Labtec在社交媒体X上展示了对ChatGPT菜单编辑100次后的效果1。Elon大学数字未来中心主任Lee Rainie指出，"AI在图像和语言中都倾向于磨平边角...优化是为了取悦而非冒犯"1，这导致算法生成的内容失去了真实食物的多样性与个性。德国杜伊斯堡-埃森大学的研究进一步证实，AI生成的食物图像会产生"恐怖谷"效应——那些接近真实但明显虚假的图像会引发更强的厌恶感1。
Generative AI models produce unsettling food imagery for restaurant menus due to narrow aesthetic standards embedded in their training data, according to recent analysis 1. The AI-generated images display unnatural consistency and smoothness that appears distinctly unappealing to viewers, with the models tending to generate peculiar food forms such as "perfectly spherical ice cream balls" and "genetically modified shrimp biting their own tails" 1.
The problem intensifies when restaurants repeatedly edit these AI-generated menus. As images cycle back through the system for refinement, a degradation phenomenon known as "convergence" occurs, in which the outputs become progressively more unrealistic and distorted 1. Reality Defender's Chief Technology Officer Alex Lisle explained this compounding effect: "When you feed a model's output back into itself, eventually inbreeding becomes excessive and the entire system collapses" 1. Lisle further characterized the disconnect between AI generation and culinary reality by stating, "It's like an alien making pizza without understanding its core principles" 1.
Research from Duisburg-Essen University in Germany has identified an underlying psychological mechanism: AI-generated food images trigger an uncanny valley effect, in which images that approach realism but remain obviously artificial provoke stronger disgust responses 1. This perception problem reflects a broader tendency in generative AI systems, according to Lee Rainie, Director of the Center for the Digital Future at Elon University, who noted that "AI tends to smooth out edges in both images and language...optimization is designed to please rather than offend" 1.
评论
还没有评论，欢迎留下第一条。