creatures caves welcome, guest
downloads   gallery   dev   community   creatchi   forum   mycaves
bookmarks | search | post new topic
Development Forum
old
[C3/DS] brain editing   
Geat_Masta

Geat_Masta
United States  


  5/23/2017  1

So in developing Kreatures i wanted to test some of the ideas i had with brain lobes i knew worked, so I've been having luhrstaap help me by running tests on dragon norn brain edits.

Thus far I have discovered it is possible for a norn in C3/DS to learn "the hand is visible + i hit creature = hand slaps me." However thus far I have not found a way to get this relationship to affect the output of the brain.

I wanted to list some of the things i have learned, if there's a brain edit you want to try, describe it to me and I'll try to tell you how it might be done.

The first thing we learned is that lobe position does absolutely nothing in C3/DS, it's used for rendering like color. So you can put two brain lobes in the same place if you want, it makes no difference.

Each brain lobe seems to have a hypothetical neuron, the spare neuron, that is shared by all neurons for SV rules. E.g. If neuron 5 calls "register as spare neuron", then the spare neuron has all it's values set to the values of neuron 5, and the spare is set as 5. The engine then queries this value for different things. E.g. when decisions are made, decision script 5 doesn't activate because neuron 5 fired, but because neuron 5 registered as the spare neuron.

The way attention/decision works, is that when a neuron in the comb lobe fires it triggers attention and decision neurons to fire, if their firing is stronger than the current firing neuron then the current decision script gets canceled and the new script runs. If it is not stronger nothing happens. When the eat elevator bug happens it seems like the "spare neuron", isn't cleared. However this could be done manually with a genetic edit, where we add "successerase" stimuli, and use that to blank the neurons.

Neuron values are signed, this was referenced by chris double once, but the information was lost into the aether. But in the vision lobe when a neuron has a value of -1 the object is to the left, when it's a value of +1 its to the right, meaning that a value of 0 is the norn's current location. And the stimulation of the neuron is 1 - abs(output)

This is more significant in ForF, the ForF lobe is extremely broken and unfinished, much like the rest of the genome. For example, it handles both feeling angry and feeling friendly as increasing dislike of the target norn.

This is basically how the forf lobe works: neurons start with a value of 0, when something good happens this gets reduced, with -1 being love, and when something bad happens it gets increased, with +1 being hate. From there there are 3 tracts from forf->comb, these correspond to decisions to slap, hit and retreat from creatures. There is no connection between forf and tickling/kissing. Which explains why norns in C3/DS are so slap happy.

So when you see a positive association form, that's the forf lobe feeding negative input into the desire to hit/slap/retreat, so those are getting surpresssed. But kissing the creature you love, doesn't register as a better option than eating/dropping/resting them.

Furthermore, when a norn that is disliked is present, that increases the stimulation to hit everything, not just the norn. Because the forf lobe connects to the entire row of the comb lobe, not just as it applies to creatures.

The comb lobe is similarly broken, e.g. the decision to go east/west is not handled in the comb lobe, because those do not act on an agent, and the comb lobe is about associating agents and drives. That makes sense, but rest also does not act on an agent, and is hooked to the comb lobe. So norns have to learn that resting when sleepy is good as applied to all current focuses of attention individually, rather than as a general thing you do when you're sleepy.

 
Sanglorian

Sanglorian



  5/24/2017

This is very interesting stuff - thanks to you and luhrstaap for digging into this.
 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/24/2017  1

We are making some really fascinating advances, yeah. He's gone back to working on his own game for now, but I'm still tinkering with the brain edits. This year, I WILL have something really cool to release for CCSF ;)

Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 
Wingheart

Wingheart



  5/24/2017

This is all very good! Very interesting.

Proofreading thing: you didn't finish your sentence about the Eat Elevator bug.

 
Malkin

Malkin

Manager


 visit Malkin's website: Malkin's page at CWiki
  5/24/2017

Is that stuff about the forf lobe true for the cfe and allied breeds?

My TCR Norns
 
Geat_Masta

Geat_Masta



  5/24/2017

I haven't examined the brains of the CFE/allied breeds, only the official chici.

On the tracts page of the gentics kit, you'll see forf->comb connects from neuron 40-79, divide by 40 to get the decision this corresponds to. Which is decision 1: deactivate, or slap.

so if there's some that correspond to push/pull, in CFE, it's probably fixed.

Not to spoil anything about the brain edits: but they've started forming social groups.

 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/24/2017

From the work I'm doing on the Dragons, the answer is no for the CFFs. I was going to add a tract to ForF covering Push, but it's already there. I only needed to add Pull.

I've already posted about some of what's going on in my blog if anyone's curious. You wouldn't BELIEVE the Norn soap operas you guys. The kind of stuff that used to crop up mostly by coincidence and sort of resemble a soap opera is happening ALL THE TIME and it looks way way less coincidental and way more purposeful. TBH it's almost unsettling. At the very least, their social processing is significantly more -complicated- and from their perspective the results are a lot better. They seem to truly recognize other individual Norns now. Maybe not to the same degree that you and I as humans recognize individuals, but to a far greater degree than they were capable of before. Serial monogamy is now the MOST COMMON reproductive tactic among my testers, for example, and not because pairs isolate themselves away from other options.

Whether it's -improved- in a subjective gameplay-enjoyment sense... that I'm still working on. XD Because of the aforementioned hit-happiness, they're still pretty destructive creatures. Also - this may or may not be entirely due to that - murders are happening... not A LOT, but more than did in the earlier edits or standard CFFs. They aren't as able to express their like/love as they are dislike/hate, at least not directly. Not yet. If I achieve what I'm trying to, though, their social abilities will be enormously improved. (On the upside, an adult Dragon can kill a C1toDS Grendel in literally one second, two hits, if angry enough and Critical Hits is installed. I actually felt bad for the grendels. They were -wiped out- in this one run by a weirdo Dragon.)

The current step in the project is to give them more positive associations in their ForF lobe. Even the CFF genome seems to only handle Push, Retreat and Hit - the one that did 'slap' in the base genome has apparently, in the CFF, been redone to do 'push', as when I went to add a 'push' tract I found that one of the existing three already handles push, and its SVRules are set to 'inhibit if disliked/hated, don't inhibit if liked/loved'. Now, that's PART of what I wanted to do. But it's not all of it. I'm still trying to figure out how to reverse that SVRule so it says 'if liked/loved, encourage/promote this action. if disliked/hated, don't.' Once I do, though, I'll make some new tracts which do that. The result will be that if they like/love another creature, they will be more likely to push or pull (kiss or tickle), as well as merely less likely to slap or hit.

Last night, I also hooked the Boredom drive to ForF, in the hopes that this will allow them to interpret some individuals as "fun" ( "boredom goes down when I'm around X Norn" ) and "boring" ( "boredom goes up when I'm around Y Norn" ), just the same as how it works for Anger, Fear, Pain, and Sex Drive. Since happiness for Norns is defined as the absence of any high drive, that was the closest I could come up with for now to giving them an actual -positive- possible association with other creatures.

Otherwise, their positive relationships have been entirely either family-based (parents and offspring, siblings) or mating-based, because their only direct positive connection they could make was through Sex Drive ("mating with X Norn reduced Sex Drive and thus made me feel good - I like X Norn";). Their ability to detect familial relationships through the Detail lobe also apparently gave them an indirect ability to develop some sort of connection to their family members, if for no other reason than the non-brain genes that incline them to hang around family members when young, then scatter from them when old enough to breed. In short, they are always aware of who their relatives are, and even though this awareness has no direct connection to ForF the way it's currently set up (that I know of), the fact that that Comb can make connections with itself in our edit seems to allow them to make connections between concepts, allowing them to connect ideas in a way that wasn't previously possible for them. Thus, in turn, allowing them to recognize their family members socially in some small way.

If I'm correct and I did the edit right, this should now allow them to form what we humans would think of as real friendships. Playing with a toy together should create the association of "push toy + Norn X was present + boredom went down", encouraging the Norn to repeat that combined experience. In human terms, that translates to "I played with a toy, and Norn X was there, and it was fun." Especially if Norn X also plays with a toy at the same time and thus has the same experience, the two should increasingly want to be near each other when pushing toy. In short, they should increasingly want to play together. This is what I'd call a real friendship - it isn't based on merely not being annoying (Anger) or not causing you harm (Pain/Fear) or being a good mating partner (Sex Drive). It's valuing someone's company because they were around when you had a pleasant experience, and so you came to associate the two. This is a component of how human friendships work (think about it - it's true!)

The other component will come into place when I get that SVRule figured out. As Geat_Masta described, right now, when a Norn that is disliked or hated is present, the tendency to hit -everything- is increased, not just the rival. IMO this is realistic. Think about it - angry people don't just hit the person making them angry. They often hit whatever objects are around to be hit, and the angrier/more out of control the person is, the more likely they are to vent their wrath on random available objects (and people) as well as the actual person who got them angry to begin with. So IMO this is so by design and is not, per se, inaccurate.

In the brainedits, especially, it also seems (I may be wrong about this) that the presence of disliked Norns causes them to experience everything a bit less pleasantly than they normally would. Things that make them angry make them a bit angrier than usual, things that they enjoy seem to give them less enjoyment than usual. If this is accurate, that's also realistic IMO. I know when people I don't like are around, I'm more irritable, more inclined to be cynical, and less able to enjoy things that I like. XD

Once I get the SVRules figured out, the opposite will become the case as well. That is, when a liked/loved Creature is present, the individual will be more inclined to push/pull everybody and everything, not just the liked/loved creature. In short, they'll be more inclined to be nice to others and to play with toys and other parts of their environment. This, too, strikes me as realistic and appropriate.

Mind you, I'm not disconnecting the older circuits. I just want to add these positive circuits to finish the set. Once all of them are in place, individuals will have a much fuller range of options for social and personal development.


Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 
Malkin

Malkin

Manager


 visit Malkin's website: Malkin's page at CWiki
  5/24/2017

Part of the trouble is the menstrual cycle, I think it causes kisses and pats to be bad at certain times of the month.

My TCR Norns
 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/24/2017

Does it? Interesting. I'll look into that once I have the new tracts up and running.

Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 
Missmysterics

Missmysterics



  5/24/2017

I am pretty sure CFE creatures do form positive associations based on just patting/kissing one another,not neccisarily just the opposite sex, so I would assume loneliness at least has an effect, and otherwise Vampesses' Breeding .cos edit wouldn't have done anything.

Also one problem I've found with creatures getting angry is that if if one creature hits them, they'll hit a random creature in response, they don't really seem to know how to hit back at all.



 
Geat_Masta

Geat_Masta



  5/24/2017  1

the creatures are programmed in a way where they can see one object at a time, which is chosen by the engine, so if a creature gets hit by A, and they're paying attention to B, the forf lobe gets correctly updated because it's hard coded, but they're still paying attention to B, and they don't know A exists. So B get's hit.

Basically it's a problem in the c2e engine.

 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/24/2017

I was able to improve the appropriateness of hitting by changing a few settings in Stimulus genes. For example, "I've been hit" reaction "hit it" helps a little because it improves the odds that they'll hit back right away at the source of the hitting. But even then it's impossible to completely correct for that flaw in the engine.

And actually, no, they don't form attachments in ForF from kissing/tickling. They do experience those as positive and pleasant, but that's not the same as those actions having a direct connection to the ForF lobe. (What's fascinating is that they can form positive associations to some degree in spite of the fact that there is no direct connection, actually.)

Here's how it works, basically. ForF in CFF and presumably CFE takes input from Loneliness, Sex Drive, Pain, Fear, and Anger. That's it. Those are the ONLY drives that have a DIRECT influence on the ForF lobe. By the same token, the CFF genome has three tracts that influence behavior directly in response to the ForF's stored opinions. Those are:

Inhibit hitting when the creature is liked/loved but not when disliked/no opinion.
Inhibit pushing when the creature is disliked/no opinion but not when liked/loved
Inhibit pulling when the creature is disliked/no opinion but not when liked/loved

Again, that's IT. They know to hit people they like less, but not to kiss/tickle people they like *more*. Which is exactly what I'm working to change. :)

Again, I'm not talking about their non-brain genetic reactions to things. Norns definitely do experience reduction in Boredom, Loneliness matters, all that kind of thing. But that's not the same as there being a direct connection to or influence on the ForF lobe. In short, they can know that being tickled feels good, and that being alone feels bad, but they can't make the leap from that to "I like X Norn. I want to be nice to them. So I'll tickle them."


Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 
Geat_Masta

Geat_Masta



  5/24/2017

you could add a chemical ticklease to "i've been pulled" use a receptor to hook that to a neuron, and hook that neuron to the forf lobe. Then they would like norns that tickle them, right?

Probably you should make a stimuli lobe for all the "it does x to me" stims, and connect it to FORF, completely removing the drive connections.


 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/24/2017

Oh hey. That's an idea. If the current setup doesn't work, I may well try that. They need SOME kind of true positive experience. Everything "positive" for Norns is defined in terms of absence or reduction. I've tried to reconcile that with reality, and it is true to some degree, but I keep coming back to the fact that there is such a thing as enjoyment, which is more than merely the absence of boredom/pain/etc. So it is very tempting to try to give them some kind of analogue.

On the other hand, that may not be necessary in terms of just getting them to like each other due to being tickled. If I have things set up correctly, that should happen now due to this pathway:

Tickling makes the tickled Norn feel less bored and lonely
Boredom is now connected to ForF, while Loneliness already was
The tracts to Comb allow connections between concepts
So, just as with mating when they seem to be associating the Norn they first mated with the enjoyment of mating (resulting in semimonogamous behavior and more committed sexual relationships), they should also be able to form similar concepts related to tickling: "I was just tickled. Norn X is here. That felt good. So I like Norn X more."


Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 
Geat_Masta

Geat_Masta



  5/24/2017

If comb were connected to FORF, you could make it so that if comb stimulates a neuron and that neuron doesn't indicate the creature is present a neuron in a lobe fires, with that neuron hooked to a neuroemitter that emits loneliness. But i don't think that completes the connections needed for that sort of behavior.
 
Geat_Masta

Geat_Masta



  5/24/2017

Lobes hooked to neuroemitters are needed generally, you could use the comb->comb tract to stimulate such a neuron, so it fires more when a neuron that is not excited by prior comb firing fires. Thus when something unexpected happens that neuron lights up, and have it reduce boredom, that more approaches real boredom.

I'm not sure about handling proper enjoyment though.

 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/24/2017

Yeah. Enjoyment is a really sticky problem.

It just occurred to me how odd it is that we can easily come up with ways to represent unpleasant states in our virtual animals, but we get to positivity and it's like, whoa. Hmmm.


Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 
Geat_Masta

Geat_Masta



  5/24/2017

Well the unpleasant states aren't actually different to visual stimuli as far as the norns are concerned: it's just a number decisions are based on. If norns have qualia, then pain would be no different feeling than hearing a verb.

Feeling enjoyment and feeling pain are things we can't figure out because they're subjective experiences.

 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/24/2017

That's exactly what qualia are, though - that subjective sense of things "feeling" a given way. And actually I think the comb-comb edit did give them something along those lines, or enhanced whatever teeny bit they may have had before. At a fundamental level, for our brains, too, it's just stronger and weaker electrical pulses, and chemicals moving around. But somehow, out of all that electrochemical activity somehow produces the emergent property of subjective experience. Just like the huge accumulation of water molecules, etc., that is the ocean produces the emergent property of waves, and tides - properties that could never be predicted from the behavior of individual water molecules, or even individual raindrops.

One thing I notice whenever I switch the edits off to do a 'control run' of normal CFF Dragons is that things seem to MATTER less to them. It's much easier to view their behavior as the result of pure number crunching, as you describe, with maybe the tiniest spark of potential for something more. The edits behave more as if their choices mean something to them. Their actions are less pointlessly random-feeling (well, except when they're babies, but I'm pretty sure any baby by definition is like that). Their purposes don't always seem to make a lot of sense in human terms, but there's much more of a feel of something real happening. I could be imagining things, though. We'll see. They're too new to be sure of anything yet.

But just to counterpoint myself, we really don't know much about consciousness or qualia. We know that they're almost certainly emergent properties of complex systems rather than being things that are 'produced' by specific parts of the brain dedicated to that purpose, but we don't know what degree of complexity is the minimum threshold for such traits to begin to emerge. For all we know, it's much lower than we realize. Or higher than we'd like to think (excluding, perhaps, some humans... ) We really, really don't know. So I'm doing my best to keep an open mind in both directions while working with these guys.

But I have to tell you, I know what my gut-neurons think of it. They think I'm looking at animals doing animal things, not programs just crunching numbers and acting out animal-like behavior. Whether that feeling is an illusion or not remains to be seen.


Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 
Geat_Masta

Geat_Masta



  5/24/2017

At a fundamental level, i would describe consciousness as the internalization of motion: the ability of a neural network to build a model of how it's next inputs will look based on it's current inputs and it's current output. If it can successfully build that model, then i would say it is conscious, qualia is basically irrelevant to that (see: blindsight patients). But I think that qualia has to evolve for meta cognition to function, and that is it's purpose evolutionarily.

Of course, I'm also not convinced that a norn is capable of meta-cognition, because it's memories are all stored as that sort of sequence memory i just described, making them more like patient HM, if he had no memories to draw on prior to the accident. And since I've seen no models of how memories aside from sequence memory function, I have no expectation of norns being capable of meta cognition or having qualia.

Hence why i messaged you about the metacognition thing, if we can prove that by experiment, the only conclusion possible is that the creatures do experience qualia.

 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/25/2017

Blindsight patients still experience qualia. Just not visual ones. They feel the warmth of a fire (as opposed to simply knowing their body temperature has increased), they feel pain (as opposed to simply knowing their body has been damaged). But generally speaking, your definition is as good a one as any other I've heard. The only place I'd disagree is that your definition implies absoluteness - it's not conscious before then, it is conscious after. I would tend to suspect that it comes in stages, if the deveopment of that ability comes in stages.

Then there's the argument (which I don't necessarily agree with, but I can't refute it either) that even that much isn't proof of consciousness, because such 'planning' could still happen without the program being aware of its own awareness. That is, there's a difference between being aware and KNOWING that you're aware. That is, obviously that's true (that's the difference between cognition and metacognition, naturally) but that granted, what's the solution? Infinite nested "awareness loops" in the programming, to make it aware of itself being aware of itself being... That just doesn't seem practical. There has to be a limit, some point at which the argument of "it could still be automatic" falls apart.

I suspect we may have nudged the Norns closer to that lower threshold of being aware. Knowing they're aware is a whole other ballgame - that's where your definition comes in IMO, but some would argue that even then.

Gah! This is like the stickiest scientific problem there is. Makes dark matter and black holes look easy.

EDIT: Of course, all of this would be easier if we knew how memory worked in real brains. But since knowledge about that is so limited right now, we're really fumbling around in the dark re: how to interpret the significance of Norn memories.


Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 
Geat_Masta

Geat_Masta



  5/25/2017

Maybe you should tweet to steve grand about your results, he may have some insight.

Also my definition does have stages, think of consciousness like a reflection: it can range from glossy paper to a mirror sheen, but there's still a lowest level of shineyness.

 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/25/2017

You know, that's not a bad idea. He's more active on Twitter than I expected him to be. XD I'm going to give it some more time, so I can be more sure of my observations, and to give Inty and Savannah time to get back to me about their groups. But once I've got a good body of information, I think I'll do just that.

Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 
Elegnaim

Elegnaim



  5/25/2017

I've been reading this and I've got some questions!

1. How does the combination lobe give input to the attention lobe? I'm not familiar with how brain tracts work. I'm visualizing something like maybe the X axis of the comb lobe maps to attention lobe neurons, so anything at, say, column 15 in comb goes to neuron 15 in attention, but I think I'm thinking of this in C1/2 dendrite terms.

2. Could lobe-internal connections be used to represent sequences of states and not just concurrent states?

 
Geat_Masta

Geat_Masta



  5/25/2017

1 you basically have it down

2 if it represented concurrent states then neurons would link to themselves and enter an infiinite loop, so the norns would only ever make the first decision they ever made.

So it's been designed to link neurons to the previously firing neurons, to represent sequences. Essentially when a neuron fires, the neuron that fires next becomes more susceptible to input, and therefore more excitable and likely to fire, but this doesn't stimulate it to fire, it needs external stimuli.

Is this true sequence memory? No. True sequence memory will be possibly in my Kreatures engine, but not in C3/DS. design specs i based the sequence memory in kreatures on

Essentially, say a norn is exposed to sequences A B C D, and A C B D, and is exposed to stimulus A. In the design I came up for for C3/DS the norn predicts B or C follows. In this case B follows, so the norn predicts C or D is next. But it doesn't, only C can follow A, B, but the norn doesn't know that. It's more similar to a markov model than sequence memory for that reason. For true sequence memory, the lobes need to be 3D, which they are in kreatures, so if you give a kreatures norn a lobe with depth 4, then a given input B can be used in 4 unique sequences.

EDIT: just to clarify, in kreatures neurons are programmed so they can't hook to themselves, it's against the rules. But creatures wasn't designed for tracts within lobes to exist.

 
Lurhstaap

Lurhstaap


 visit Lurhstaap's website: Addicted To CAOS
  5/26/2017

I will say that the Creatures system is a fairly good -imitation- of sequence memory, though, in the system we've set up here. But I am very excited to see next-generation ALife's capabiities, such as Kreatures and Grandroids. Those will blow anything we can do with DS out of the water.

Conclude with killer catchphrase.
(Lurhstaap)
"This is not knowledge -
this is information!"
New Model Army, "Courage"

 


downloads
cobs
adoptions
creaturelink
metarooms
breeds
 
gallery
art
wallpaper
screenshots
graphics
promos
sprites
dev
hack shack
script reservations
dev resources
active projects
dev forum
 
community
links
advice
chat
polls
resources
creatchi
 
forum
bookmarks
general
news
help
development
strangeo
survivor
mycaves
log in
register
lost pw
1 online
Papriko
creatures caves is your #1 resource for the creatures artificial life game series: creatures, creatures 2, creatures 3, docking station, and the upcoming creatures family.

contact    help    privacy policy    terms & conditions    rules    donate    wiki