When you’re reading a book you modelling situation that author created and most likely feel scared or laughing or what ever is book about.

Hard to tell how exactly it works, maybe through mirroring or empathy, but it works how it works. The question is - should AI imagine it similar way? Let’s imagine we create a robot with LLM inside, that calling some robotised arms or legs via MCP or so, has some memory storage, CV/OCR and more other techy words.

So, the question if tell or draw to that robot some scary image, for example a lot of rust (should robots care about the rust?), should it behaviour changes?

Probably yes as standard program to avoid dangerous situations. Like a low level code that not controlled by reasoning system, instructions are “always avoid being in water”. Unless you are a water life saving robot. But when you tell some story where other robot gets into situation with water around, should it tag that story in memory as one that dangerous, at least for robots?

Ok, maybe it not yet has feelings, but just a story “go to forest, there could be a lake, and lake is water and water is rust” potentially may affect reasoning system to choose other way around the forest.