Quote - Still, I am curious... how do other programs "render" the AO effect of a part of a model not seen by the cam ?

A 'render' is just the process of converting geometry to pixels. For speed reasons, most renderers only use geometry the camera 'sees' for the pixels, but that doesn't mean they can't use whatever part of a scene they like.

For baking the AO onto a texture like the OP wants, the renderer just runs the render process on the entire object and instead of displaying the result on screen, stores it in a texture file instead. The UV map of the object serves as a guide as to where each pixel goes in the file. eg if the UVs of a part of an object are (0.5, 0.5) ie exactly in the middle of the texture, and that part of the object renders black in AO, the renderer copies the pixels to the middle of the texture. This is then repeated for all parts of the texture to get your final 'bake.'

And this can be done for any kind of rendering, AO is just an obvious example. eg you can do it for procedural textures, lighting, GI, bump maps, SSS etc. It's used whenever you want to do the calculation of a particular effect only once and then just reuse it again and again to save rendering time, like in an animation.