Emergent Systems

[PurpleYouko]

I would like to point out that while there are a finite number of amino acids used, the length of proteins constructed from these amino acids are thousands of amino acids long.

I wasn't even considering attempting to model real chemistry here.

My idea is more that we would have a finite but large amount of pseudo-proteins that can be made by the DNA bit code. They do not have to look, act or in any way resemble real proteins.
Each pseudo-protein does a different thing. One provides energy for movement, one digests a certain food type etc. Once produced, they can be utilized by other parts of the DNA code.

Everything is tokenized and we never really attempt to model real physics or chemistry.

[PurpleYouko]

Aha but I cleverly never mentioned chemicals merely substances. And a limited number of them at that.

In my head I imaginined there being more 'clumpy' substances like soil, air etc... it might be stupid but it would mean you wouldnt need to calculate chemical processes. You could simply state the rules of interaction instead of having to calculate them.

PRECISELY! Well said Ulc.  :D

[Griz]

great 'brainstorming' thread lads ...
enjoying it a great deal ...
and almost as excited as Greven. ;)
good on ya all ...
keep it up.

[Numsgil]

I'm not understanding the modus operandi of the DNA to the world.  Explain it to me in terms of the current system.

Currently, DNA manipulates packets of information (memory cells) into other packets of information (sysvar commands).

Are we still doign that or comming from a new direction?

[PurpleYouko]

The problem you are describing has apparently gone right over my head; I just don't see one at all. Sure it will take a while to list all commands, assign them codes and group them together, but that's doable.  :blink: Isn't it?

I can see tha nobody else seems to see this as a problem. here is why I do (again)

Sure we can make up a bunch of rules which relate DNA to action.

Sure we can list all commands etc.

Thise things ARE the interpretter.

But doing that puts us right back where everybody who wants simplicity, hates being.
All these rules, commands and stuff are the ARTIFICIAL part of the program that we need to get away from. This is where all the complaints come from.

Command X in DNA results in action Y in the sim. It is a RULE and is always enforced.
If I want action Z in the sim to happen then I need a solid rule that will link a DNA command to that action.

What we really. ideally need is a system where we have no direct control at the interpretter stage. This is where I get stuck. I don't see any way that action Z can ever happen unless the interpretter specifically instructs it to happen in that exact way in response to a specific DNA command.

It is too rigid to allow true emergence.

Please say that somebody understands my point here  :(

[Numsgil]

Please say that somebody understands my point here  :(

:raises hand: I do.

Which is why I've been saying that it's really just an argument over what level of abstraction you want.  I prefer to keep the abstractio just below the level of useful behavior, because it's easier to maintain in code.

Below that I don't much care.  I don't care how the bot moves, but why it decided to move now and not later.  I'm more interested in behavioral evolution than phsyiological evolution.

[Ulciscor]

Aha! I see! :D Well I agree that there are rules imposed when a command is linked to an action, but I don't think you can get around that. Although...

Maybe the command would make the organism attempt to undertake the action. Like (for arguement's sake) .up translated to a bot attempting to fire particles to move forwards. The bot would then try to do so. Whether it would succeed would depend on the factors of the environment, which would vary on the situation and would never be rigid or imposed.

I am thinking of analogies in biology. And I'm starting to think DNA is not the best way of describing bot script any more. 'Neural pathways' seem to be much more accurate, since sensory inputs or prompts are linked to some sort of output.

[PurpleYouko]

I'm not understanding the modus operandi of the DNA to the world.  Explain it to me in terms of the current system.

Currently, DNA manipulates packets of information (memory cells) into other packets of information (sysvar commands).

Are we still doign that or comming from a new direction?

I am thinking along parralell lines to DB. We have to have some kind of register system like the DB memlocs otherwise no variability of commands is possible.

ie. command A could tell the organism to turn left but without some way to specify how far left it would do exactly tehsame thing every time the command was used. That would just make the DNA less flexible.

However I want to get away from the reverse polish notation stuff.

I envision each command to be simply a string of bits (possibly varying length) that contains a whole bunch of instructions that will be either carried out or not depending on whether a previous command has enabled or disabled it.

Certain instructions could (for example say) skip next command IF condition X. But all instructions in the current command would be carried out.

I haven't really fleshed out my entire idea yet, not even fully in my own mind.

[PurpleYouko]

I am thinking of analogies in biology. And I'm starting to think DNA is not the best way of describing bot script any more. 'Neural pathways' seem to be much more accurate, since sensory inputs or prompts are linked to some sort of output.

I was thinking that too.

The present DB DNA is more like a behavioural program than true DNA. It is more what the bot is thinking than what is going on in the real metabolic pathways. There is some crossover though.

Maybe we should have two parallel programs, one for behaviour (the brain) and the other for metabolism and stuff.

[Ulciscor]

Totally agreed! I have recently been thinking that too.

The genetic stuff could handle metabolism, reproduction, and could have effects on the behavioural aspect.

The neural stuff could handle senses and actions, and could have effects on the genetic aspect.

Then there is the actual physical bot that could attempt to 'obey' the other 2 aspects, whilst also affecting them both.


Does this make any sense or is it needlessly complex? I'm fairly sure it would take an eternity to compute for each bot.

[Griz]

But doing that puts us right back where everybody who wants simplicity, hates being.
All these rules, commands and stuff are the ARTIFICIAL part of the program that we need to get away from. This is where all the complaints come from.

they do?
then I think you still misunderstand.

there is no problem with having rules ... they are essential.
let me use a real world example in an attempt to explain.
we have gravity.  
if we do not have something to support us ... we will fall.
it's a rule. it isn't artificial. just how it is.
now ...
those of us who behave in such ways that allow us to fall and die ...
are less likely to survive and reproduce, eh?
and those of us who behave in ways that result in us not 'falling and
not being able to get up' ... those of us who don't jump off bridges ... will.
and as a result ... overall ... the majority of those decended from us are
going to be those who exhibit this particular pattern of behavior.
is this not so?  
and so ... we might say that this is a behavior that 'emerged'.
the rule of gravity and falling was not altered or changed ...
only the behavior that resulted in falling is one that has not continued.
the behavours will evolve to conform to those basic rules ...
and I see no problem with that.
in fact ... it is these rules that will allow us to set up experiments ...
to manipulate the environment ... to alter the context ...
to impose some limitations ... WITHIN which ... behavior may be
observed to emerge.

does this make any sense?
am I saying something other than you are?
perhaps it has appeared so ... but underneath ... I think not.
where does this fit into your vision, PY? if it does.

[attempting to check out the same book and read from the same page.] ;)

[PurpleYouko]

It makes sense and is about as complex as the thing needs to be.

Add to that a bunch of flexible rules and you have something that could work.

I still have issues with limiting the possible actions in the interpretter though.

The way I see it is this.

If the interpretter is simple (say 10 possible actions) then the behaviour HAS to be simple too. After all the organism can only ever do 10 things.

If the interpretter is VERY complex then the organism can do loads of stuff in loads of different combinations.

This is the very heart of my arguments to increase complexity in order to have LESS control over the behaviour.

IMO we need hundreds of possible actions and many of them need to be very similar. then we group them in a graduation from one thing to another and allow mutation to choose which one to use.

[PurpleYouko]

there is no problem with having rules ... they are essential.
let me use a real world example in an attempt to explain.
we have gravity.
if we do not have something to support us ... we will fall.
it's a rule. it isn't artificial. just how it is.

Gravity isn't a RULE in the sense that I am talking about.
Gravity is a physical law.

What I mean by a rule is if memory location 1 has a value in then the bot accelerates by that value.
If memory location 5 has a value in, it rotates by that value.
If memory location 300 has a value in, then reproduce.

Those are RULEs of the interpretter. They are arbitrary, not physical constants.

[Ulciscor]

Well all the movement actions can be grouped together for similarity. In fact as far as I can tell you can group all the functions together because they are based on simple interactions. Move (substance), eject, reproduce, etc

[Griz]

If the interpretter is VERY complex then the organism can do loads of stuff in loads of different combinations.

This is the very heart of my arguments to increase complexity in order to have LESS control over the behaviour.

IMO we need hundreds of possible actions and many of them need to be very similar. then we group them in a graduation from one thing to another and allow mutation to choose which one to use.

ahha!. got ya. I see now your 'more is less'.

providing a great deal to choose from ...
but allowing the 'choosing' to be a result of behavior that
works plus mutations.

the 'gradation' is an interesting idea ...
even within a given action.
perhaps any one possible action could have a 'variable'  ...
so the action could be 'scaled' in magnitude ...
iow ... we have one action defined for  'move' ...
but a variable that is assigned/gets picked up along with it ...
for how far to move ...
that gets passed along at reproduction.
same with a 'turn' ... with a variable assigned for how far to turn.
so as a result, our hundreds of actions ...
could result in 1000's or 10,000's  of actions ...
by adding a magnitude to them.

perhaps that is how it already works, don't know.
or perhaps this is completely off the wall. don't know that either. ;)