The sameness problem behind those unappetizing AI-generated menus
While restaurant owners might look to generative AI as a shortcut to sprucing up their menu, customers can viscerally sense that something is wrong with the food.
TL;DR
- AI-generated menus often exhibit an eerie flawlessness and unnatural smoothness, making customers feel something is wrong.
- This aesthetic is a result of AI models being trained on vast datasets that favor a "pleasing" and non-offensive style, leading to homogenization.
- The issue is exacerbated when AI models are trained on their own generated content, a phenomenon that can lead to "convergence" or "model collapse."
- The AI's optimization for "pleasingness" shaves off the natural imperfections, making food look unrealistic.
- Repeated edits to AI-generated images, even minor ones, can further smooth and round the food visuals, increasing the uncanny effect.
- This "uncanny valley" effect, where images are almost real but not quite, causes unease and disgust.
- The shift towards AI-generated content challenges traditional notions of believing what is seen and heard, impacting everything from menus to evidence.