You can never remove internal body sensations even in meditation. The "self" is just a stationary point in this inner space, an element referenced in most situations, but not all, such as when listening to great music we can forget about ourselves.

So my idea was - we have an actual system for representing sensations, a system for imagination and a system for value prediction. We have the hardware for it, and we train our consciousness hardware on environment data.

If you're thinking - how can simple neurons or protein interactions do this? it's the same with a neural net. We don't understand how it can tell cats from dogs, or why it can answer our questions so well. Yet it does. Hard to believe unless you see it and even then it's hard to accept, unless you come up with your own private inputs and test it in a way that is sure not to be memorized.