Creatures Caves

This is something I've been wanting to post for a while, where I explain how the C3/DS brain actually works, in the hopes of seeing more edits of it in the future. I've posted some similar topics before.

I'll try to go over everything even when it's extremely obvious what's happening

Drive Lobe
The drive lobe has 20 neurons, in the default genome the state of these neurons is set to the exact value of the corresponding drive chemical.

The tracts are set up so that if the sum of the navigation drives are higher than 1/10th then all the other drives, aside from wait, are set to 0. Wait is set up in a similar way, except that it targets all the drives except itself.

In this way the driv->driv tracts act the same as the pain overwhelmsion organs, I'm not sure why pain overwhelmsion wasn't done in the same way.

The susceptibility of the drive lobe is set by the response lobe, and controls how much reward/punishment will effect the norn. To explain this i have to go over how it worked in C1 and C2

Basically, a norn, we'll call him Larry. Larry feels hungry, this triggers neurons in his concept lobe, and he eats. The food increases hunger decrease, which will reward Larry, and this will make his hunger neuron stimulate eating more strongly.

Lets take another norn, we'll call her Susan. Susan also feels hungry, but instead she pushes a toy. So she gets a boredom decrease, she's much less bored than she is hungry, but a bit bored, so she gets a reward. This will make the hunger neuron stimulate pushing toys more strongly.

Okay, so in C3, this was changed so that the norn will only get a reward if the action lowered the drive that stimulated the action. And susceptibility is the mechanism by which this happens.

In the drive lobe, susceptibility reflects how much the drives have changed, and in the combination lobe it reflects what actions were taken.

Response Lobe
The response lobe contains how much each of the drives has changed, similar to the change lobe of C2's novasubterras. As far as I can tell, this is not the difference between the aforementioned receptor loci and their previous state.

Instead, the 'silent' checkboxes in the stimuli genes seem to correspond to the response lobe, if the box is checked, then the change in the chemical will not be reflected in the response lobe.

The purpose of the response lobe is to set the susceptibility of the drive lobe.

Noun/Verb Lobes
These are a bit more complex than just the neuron being stimulated when you say a word, the stimulus of you talking lingers a bit longer than that, and how long it lingers is changed by random numbers. However, this is not to say that it is actually random, it would be more accurate to say it was attempted to be made random, what actually happens is that it takes a random sample 4 times a second, so these all tend to average out and be the same every time; it should only do this when the state changes, and add that to the output, then have the tract read from the output; to get the desired effect.

Vision/Smell Lobes
These are grouped together because of the Vision<>Smell Equalizer brain tract. The function of this tract is that if the norn can both see and smell the object then the stimulus of the object is the average of these things.

This gets the wrong answer; because if someone makes an agent and forgets to make it emit CA's then the norn may be able to see an object with a stimulus of .5 and smell it with .01, and then these will be averaged to make it far less stimulating.

However, this does achieve the desired effect of preventing the norn from being overstimulated. All the values in the brain need to be between -1 and +1 for it to function correctly, and this stops it from going over.

Regardless, the correct value of the calculation is the max of smell and vision, not the average.

The input of the vision lobe is a number between -1 and 1. To get how far away the object is from the norn in the X direction, multiply this by the value of the RNGE CAOS command, by default this is 500.

As far as i can tell, norns can only see objects that are in rooms to the left or right of the room they are currently in, and they can't see the room directly above them, even if it has 100% permeability.

The output of the vision lobe is 1 - |input|

I'm not too sure how sensitive the smell lobe is exactly, to be examined.

The output of the smell lobe is (input * 2/3 + 1/3)

Move Lobe
The input in the move lobe is supposed to be the difference between how far away the object was and how far away it is. However it does the calculation wrong.

The issue is that if an object is at position .3 and moves to .-2, the norn perceives it as having moved by .1, instead of .5; this error happens due to move taking the absolute value of the vision input, rather than the raw input.

The output of the move lobe is: output * .0484 + input * .9355

Additionally, the norn will perceive an object coming into, or going out of, view as having suddenly moved.

Stimuli Lobe
The stimuli lobe aggregates the values of noun, move, smell, and vision. The noun lobe is only added in if the norn can smell or see the object.

Aside from that the basic way to understand this lobe is that it does this:
output = (noun * a + move * b + smell * c + vision * d) / (a + b + c + d)

Where by default:
a = .875
b = .125
c = .25
d = .25

This indicates that the noun lobe was originally intended to have a weight of .375, but it was later changed to .875 because the norns were too free thinking.

A suggested brain edit for beginners is to play with these numbers to get norns with poor vision, or who can only see moving objects.

Situation & Detail Lobes
These contain details about the norn's environment like if they are holding an object, or if the norn they are looking at is related to them.

They aren't used for anything, because these aren't hooked up, norns can't learn to not try to drop things they aren't holding, so they quickly learn that dropping usually doesn't work. This is why they have trouble with the idea that they should drop the bee they picked up.

Another suggested brain edit for beginners is to add tracts that inhibits picking things up when the norn is holding something, and inhibit dropping when the norn isn't holding something.

Attention & Decision Lobes
These are both basically WTA lobes from C1, the cell with the highest current input is the current focus of attention/decision. It mostly gets input from the comb lobe, but left/right are instead linked to response. If the norn makes no decision then he executes the quiescent script.

The decision lobe stores the previous decision in var5, this is used by the response lobe to prevent learning from left/right.

Mood Lobes
This appears to add the values of good drives and subtract the values of bad drives, but upon further inspection, the code to subtract the bad drives doesn't actually work. Not sure what the actual function is.

Forf Lobe
This lobe has 36 neurons, but the norn can only learn about 35 creatures. Each neuron corresponds to a different creature, a state of -1 indicates hate, +1 indicates love. It is changed by the values in the drive lobe, but it doesn't seem to be related to if the norn did the given action or not, so if a norn gets stung by a bee, he'll think less favorably of the norns that were around at the time, even though they didn't do anything.

Forf is used to inhibit good actions when the norn hates another norn, and inhibit bad actions when the norn likes another norn. However, it does this uniformly. Meaning, that if a norn is bored and with a norn he hate, he is inhibited from pushing a ball, because the norn he hates is inhibiting his push decision.

Similarly, it never stimulates a norn to kiss a norn he loves, just not hit them.

Comb Lobes
This is the most complex lobe, but it's not all that complex once you get down to the nuts and bolts of it. Essentially it sums the values of drive and verb and multiplies by the value of stim, so norns won't try to eat things that aren't there etc.

The most interesting thing about this is how different it is from the C1 concept lobe. In C1 the concept lobe had 640 neurons, and the norn had to learn to associate decisions with specific neurons, meaning that a norn could functionally forget that he was capable of dropping things, until the language instincts re-ran and reminded him.

In C3 the way learning works is that the norn will learn that doing X to Y will reduce drive Z, all learning does this. Comparatively, in C1 norns were more generalizing, that they should retreat when sad, and so they retreat from carrots because they feel sad, and are generalizing what they do when sad.

This is the basic reason the norns are less interesting in C3, because they aren't capable of generalizing situations in the same way.

Also of note is that in C3 the comb lobe is 40x11 and has 440 neurons (in CA the lobe is 40x5, this appears to be the only difference between the CA and C3 brains). Far less than it has in C1, and in C2 having that few neurons caused OHSS, and it had to be boosted by around 35% in most breeds that fixed it.

To me this indicates that copying the canny genome into C3 wouldn't really be possible, because the most likely reason they did learning this way is that forming new dendrite connections never worked correctly.

I think that's everything, i hope someone learned something!