"How does the system acquire the rewards? They would have to be in the same recorded format as the sensory data if you are going to apply a simple +/- ranking schema."
------------------------------------------------------------------------------------------
You install the rewards as senses yes. In our brains the sensory input doesn't pass any rewards because before it gets saved into memory the neurons that it goes through are basically the rewards i.e. input-sensitive-neurons. So in a way we definitely yes have or do have The rewards in there.
Think - if a nerve in your arm to your brain sends a pain signal, and another time a pleasure signal, WHY do you talk and talk about which hurts and which feels soooo good? Neither of the signals are any supior than eachother. IT'S THE REWARDS. The rewards rank and labbel the senses saved into memory. Later they get chosen to do if + and tweaked, or to withhold if - and then guess. And stopping bad rewards is a good reward, or becomes so by becoming a semi-reward by the neutral senses linking to real +/-rewards ex. gravity is so good/win or loss condition.
"I think your describing saliency here. How do you rank the power of the senses?"
"How and when does the linking occur? How does the system know when to link a returned/ ranked sense to the current motor output?"
------------------------------------------------------------------------------------------
The senses are saved as strength in neurons, along with a rank (<which is that< [I'll explain how it ranks higher]), and a label. The more a sense matches the rewards, the more it spikes, that's how it gets ranked higher. The louder or brighter a sense, the stronger the rank. The more a sense is matched at in memory by a strong sense the stronger it gets. Example when you stare at something the input matches to the same memory.
The stronger it is the more faster it charges and fires up to rewards and down to current memory front to be saved near just saved actions&senses and therefore linked to them since so close saved.
The strongest sensory input OR the strongest sense matched at in memory and brought to memory front then goes to search to a match in memory.
"How would the system produce an internal ‘thought’ that’s not linked to any sense? "
------------------------------------------------------------------------------------------
All thoughts are senses and their linked actions being matched at in memory and brought to active memory front of the memory bar. If you want a sense not linked to any sense, like say a video from a month ago that already has sound linked and you want to exchange its sound to a song heard last night, you end up matching it and bringing it up to rewards and down to mem front, and I presume the music you are adding is strong enough matched in your brain that it also comes to mem front, then once the video and your added music is at mem front, it had already went up to rewards to like it, and also now both of them go to match memory to think about (the vid and added music - not the original weaker sound)
Schematic - notice the "backpropagation" is a memory match that goes back-up to rewards and back-down to current active memory front.
http://s1318.photobucket.com/user/coolgodboss/media/New%20AI%20Schematic_zps6r5pz19t.png.html?sort=3&o=0That little backpropagated blue dot ontop a sense is overlayer crops. I'm a little, advanced. I can do right now, I faintly see a little bunny hopping on my table.
The reason I "see" my table FullSense and the bunny clip FaintSense is how much they (at memory front) match a memory at left of the memory bar. If there was nothing to match, you'd see nothing, and be a big 0 thinker, like babies are at first.