Entry #6: Deciding What to Remember
Neural Cube needed a way to decide what was actually worth keeping before it could be trusted with keeping anything at all. Not every moment deserves to last, and a robot that held onto everything indiscriminately would end up with a memory as useless as no memory. The initial worry was less about how rich a memory could become, and more about the plainer problem of what gets remembered in the first place and what gets left to fade.
A moment earns a place in memory by standing out somehow. For example, a name attached to a face for the first time, an instruction meant to hold past the conversation it was given in, or something significant enough that it shouldn't be forgotten. This doesn't mean that routine and repeated occurrences don't make it into memory at all. These end up strengthening or updating existing memories. Memories that don't get reinforced fade in strength over time. This draws inspiration from how remembering and forgetting works in biological systems.
Memory stays fully on-device, the same as everything else the robot experiences. Nothing it remembers about a person is uploaded, synced, or shared elsewhere.