https://claude.ai/chat/01902982-3ad9-4be9-86c0-739f507f6db6 # Predictions About Relationships Between Surface and Scene ## Prediction Relationships, Not Surprise You're right to question my emphasis on "surprise." Most picture viewing isn't surprising. What I should have said is that pictures create a specific kind of prediction error that needs resolution. When looking at a Miró drawing, our brain must resolve how these particular marks create this particular depicted content. This resolution process, not surprise, drives our [[aesthetic experience]]. ## Design-Scene Properties and Prediction Rather than saying design-scene properties "emerge" (which is indeed unclear), let me be more direct: design-scene properties are directly represented in our perceptual system through specific predictions about how surface features relate to depicted content. ## The Crucial Point: Relationship Predictions Here's the key idea in more detail: When we perceive pictures, our brain makes three types of predictions: 1. Predictions about surface properties (lines, textures, pigments) 2. Predictions about depicted scene properties (three-dimensional forms, depth) 3. Predictions about how specific surface properties create specific scene properties This third category—relationship predictions—is what explains design-scene properties in [[predictive coding]] terms. Consider Cézanne's pine trees. Our visual system predicts: - Properties of the brushstrokes themselves (their shape, texture, color) - Properties of the depicted pine needles (their three-dimensional arrangement, depth) - **How these specific brushstrokes create these specific pine needles** This third type of prediction doesn't just link surface to scene generally, but links specific surface features to specific scene features. These relationship predictions explain why we can attend to how "this particular brushstroke" creates "this particular pine needle." The [[predictive coding]] framework handles this through hierarchical processing: - Lower levels predict basic visual features - Middle levels predict surface properties - Higher levels predict scene properties - **Cross-level predictions** link specific elements across these levels These cross-level predictions represent precisely [[the design]]-scene properties Nanay describes. When we aesthetically appreciate a picture, we're attending to these cross-level relationships. In naturalistic pictures, these relationship predictions might be fairly straightforward. But in more expressive works, [[the relationship between]] surface and scene becomes more complex and interesting. When thick impasto creates the impression of sunlight, or when quick brushstrokes suggest movement, these relationship predictions become especially salient. [[This approach]] explains why design-scene properties require distributed attention. We must simultaneously process information at multiple levels of the predictive hierarchy to grasp these cross-level relationships. Does this clarification of relationship predictions help explain how design-scene properties fit within the [[predictive coding]] framework?