Enviroment/Organisms

[yrwyddfa]

How about this:

(i) Everyone creates a programmatic organism that conforms to a preset interface (Eat, mutate, sleep, shag, die)

(ii) Everyone's organisms are loaded into an environment

(iii) Someone presses go

(iv) the meglomaniac organism that either (i) is in the majority, or (ii) is the only one left after a given piece of time wins.

The object of the exercise is that the environment will measure the efficiency of the organism against the other ones, and the ratio of efficiency will determine how much the organism gets to eat, sleep, shag, or simply dies.

There should be a resource pool which is limited - which should mean the poorer orgranisms will need to mutate or die.

[yrwyddfa]

Further thoughts:

The virtual environment will be a chuck of heap memory organised like a 3d safearray. The organism will need to be able to 'move' through this environment in order to get it's resources to carry on living.

The organism will need to encode behaviour in a binary string (0001010 etc) which will be partially modified by the environment

The binary string must encode for eat, sleeping, killing etc etc. The fact the environment is finite means that there will be competition for resources.

[yrwyddfa]

Even further thoughts (tell me to shut up if you're not interested)

Use IDL to define organism/environment interface (can be implemented in any programming language)

The results will be useful from a scientifc perspective: competing behaviours generated from binary strings implementing different strategies for survival.

All will learn powerful new techniques for solving NP hard problems. Particularly: bayesian intelligence, genetic algorithms, neural networks (for DNA pattern matching: you'll want to fight other organisms, and shag your own)

If we do this then I'll need to enlist the help of those already in the scientific community in order to publish (?) the results. And we'll need proper validation of all the work.

A huge amount of techniques could be employed to make your organism more effective. Sex is probably the best (isn't it always) you'll need to recognise one of your own organisms in order to have sex with it. If you reproduce with another's organism then your resultant DNA will be incompatible and will be useless.

We will probably need to get hold of some powerful servers to run the environment. I'm sure if there was commerical advantange (and/or) publicity Jupiter Media would contribute the loan of a monster of a server.

There will need to be collaboration on the design and execution of the environment - which will probably have to be in VB6 as this is the common denominator of all who frequent this forum.

[yrwyddfa]

Furthermore:

Each and every participant will likely need to learn something new to win.

Once the environment is constructed we can issue the first run, and allow others whose organisms have died to redo their organisms and reenter. There could be some sort of high-score system that tots up the success (or otherwise) of each persons submissions. Run the environment for a month and see who comes out on top.

We can then measure evolution against human wit. ie Can an expert losing human designer create a superior organism faster than the evolution that will be going on inside the winning organism.

Others, and I, if this idea is accepted, will produce a tutorial on evolutionary theory and practice that will be available for everyone to read before commencing.

[yrwyddfa]

Even more:

Why is the a good idea?

(i) It relies on algorithmic correctness
(ii) Each step (eat sleep shag die) will be timed and will be executed on the basis of the the fastest goes first. So if two organisms have decided to eat the same food resource, then the execution speed of that decision will be measured and fastest gets to eat it
(iii) The programming should be fairly easy. Winning will be particulary difficult. Each organism should publish it's DNA. But food, for instance, could be poisonous. The organism should recognise this, and try to avoid it. The poisonous food DNA sequence will be published in advance.

[Phill64]

it sounds a little over the top science wise, as opposed to programming ability wise.

as in, that's more like a scientist game on a prebuilt simulation system, not a program

[yrwyddfa]

I guess: the programming, per se, will not be particularly hard to get it to get it to work. But to make it efficient, and to develop algorithmic strategies (whether already published, or completely new) will require a good modicom of wit, and skill by the programmer.

[dglienna]

Now, if you started everyone off with the basics, and allowed us to optimize, we would both learn, and be able to compete. Kind of start everyone off at basically the same level. It may be biased, but not more than other contests.

[yrwyddfa]

Even the most simplest of organisms has at least an equal chance of winning. Someone may put advanced neural pattern recognition into their software, whilst another would take the risk of cyclic redudancy check for DNA recognition.

The CRC would clearly be able to act faster and that might give some sort of advantage.

We could, for instance, make the requirement for food a function of execution time - so the more complex organisms require more food to survive.

In terms of programming I propose that we all implement an interface (to the environment) and inside that interface is whatever the programmer wishes to put. The interface would obviously need to be agreed in advance.

Should an organism die (likely for all first generations, I might add) we could provide the actual pattern of behaviour the organism displayed (some sort of biography) and the programmer can then adjust the rules for how the organism acted accordingly. In this case it's an evolution vs intelligent designer: Can a human being intelligently resolve it's orgranisms problems quicker than the surviving organisms can evolve?

The questions (and answers) that could possibly come out of this, in my opinion, are huge.

The rules and the process, as they stand, are particularly undefined. If there is sufficient interest (20 or more organisms?) then I can get going on the basic definition of the interface, some sort of tutorial, and the basic design of the environment.

So whose interested, then?

[yrwyddfa]

as in, that's more like a scientist game on a prebuilt simulation system, not a program

I've thought about this comment, and it does have some merit.

But . . .

The programmer will need to decide on whether to use simple IF/THEN's or to use more advanced techniques.

The choice of efficient programming, vs 'intelligent' programmer.

I personally think that this is quite a challenge to anyone who programs. Most programmers don't even think of 'out-of-the-box' techniques to solve NP hard problems.

Entering into this competition will require some real original thinking, efficient programming, and modicom of luck.

Programming is, of course, a tool for achieving something greater. The challenge of this competition is to survive within a framework of rules against other people's code.

Efficiency is not the only goal. You can get the most efficient programmer in the world, but that will not necessarily be enough to win.

[yrwyddfa]

What does the organism have to be able to cope with:


Sight

The organism will be given the complete detail for it’s immediate (two square deep) surrounding squares. The code for this detail will be simple, and will be supplied to each organism.

The organism can request further details from one adjacent square that it might be interested in.



Move

It will need to move to find food, and to find mates. The environment will ask each organism to which adjacent square it wants to move next.

The rules for movement are as follows:

(1) An organism can only move to an adjacent square. This can be in any direction {N,NE,E,SE,S,SW,W,NW)

(2) If two organisms try to move into the same square then neither organism can move

(3) If an illegal move instruction is given, the movement is ignored.


Eat

Each organism will need to eat. They can find food in two ways:

(1) Eat another organism
(2) Eat the environments food.

Because the environment’s food will self generate in a random fashion, the most likely source of food is another’s organism.

An organism can ask to eat another organism if it’s in an adjacent square. If there’s no food there then the order is ignored.

All move order’s are executed first. So an organism may see another’s and move out the way before it is eaten.

[edit] I'm not sure about this: I think that the environment produced food should be a 'static' organism that can breed by itself, and perhaps produce poisons for some organims - we may even see some speciation selection,and defence through this, too[/edit]



Recombine

If the organism finds another organism that it finds attractive then it can elect to recombine. This is the only way in which mutation - and possibly survival - is guaranteed.

Recombination is only allowed when it is mutual. Therefore the scenario could be that an organism might want to recombine with another, but the other is promptly eaten.

Birth of progeny will be immediately placed onto surrounding squares, and will not be allowed to move for n moves. The number of progeny produced will always be a fixed (undecided) number.


Die

Each organism will have a function on it’s interface where it is instructed to die. The environment will enforce this rule. Death occurs in a number of ways:

(1) It gets too old
(2) It gets eaten
(3) It goes too hungry

All actions will cost energy which must be replaced by food. The environment will monitor the energy levels of each organism (in fact it will tell the organism how much energy it has to spend) Once you've spent your energy: you are dead.

[yrwyddfa]

So what (in detail ) is the challenge?

You, as a programmer, must interpret a bit string in order to decide what to do.

You'll need to make many decisions about how long each chromosome of the bit string is (according to your behavioural needs - you may even need to prioritise which behavioural trait that you feel you need the most). You'll need to decide how to deal with meaningless bitstrings.

The efficiency of your code will not depend on speed, but will depend on the 'correct' survival decisions made at the correct times.

I proprose, however, that there is a limit on time for each decision an organism is allowed to make. If the decision takes longer than the designated time then the decision is void and is not carried out.

I believe in the most average of programmer can provide an organism. An outstanding programmer will find ways of heuristically solving what - when all is said and done - an impossible problem.

THe mutation rate, the sexual recombination rate, and even the type of sexual recombination (haploid/diploid?) will probably be of paramount importance.

[yrwyddfa]

All I need now is some feedback.

I need in general - not in detail - the development of the environment is not a quick and dirty thing to be done. It will need to be designed, built, and published before the competition is started.

I really need 'yeas' and 'neas' nothing more.

I reckon I need an absolute minimum of eight competitors. I will not be entering the competion (I will be creating theenvironment, and the environment hazards)

We can sort the real detail out later.

I am actually quite excited about this. Even if you don't fancy the programming of an organism, then I will certainly be happy to enlist your help. Particulary I will need fast VB6 programmers (Merri/Penegate where are you?) I will need good SQL Server designers (szlamany - I know you're listening?)

I think that this could be a great project that we can all get together and work together, and perhaps produce some data that is really useful. I'm thinking of the science of emergence (et al)

Please PLEASE let me know if you're interested (don't PM me - just post here)

[yrwyddfa]

PS: I intend to publish the results (somehow) so the university people can study what we've done. So competitors must be willing to give their real name (in private to me, in public to the publishers)

[yrwyddfa]

Now, if you started everyone off with the basics, and allowed us to optimize, we would both learn, and be able to compete. Kind of start everyone off at basically the same level. It may be biased, but not more than other contests.

Some people will be able to code simple neural nets, some bayesian intelligence (remember this is on a time basis: if you can't complete the task quick enough then you lose your move) The simplest, but most effective, should stand a better chance of winning.

[yrwyddfa]

Simple rules produce very very complex results.

[yrwyddfa]

So that's enough from me. Who's in, then?

[Shaggy Hiker]

Actually, I'm already working on something that is very similar to this. My problem is the GUI (I SUCK as an artist).

The basic program is a lake with 8 organisms. The lowest level organisms feed on three base resource pools that are determined by a chaotic equation. The reason for this is that a paer showed that three populations of plankton in a water body are guaranteed to be in a chaotic equilibrium. However, this may not be important. The original idea behind this program was a population simulation, but the behavioral aspects have come to dominate, and the population sim is now just a basis for encounters.

Originally, I had each organism determining one of 16 different canned behaviors based on 7 yes/no questions. Seven may not seem like many, but this was back in the bad old days of memory models (under DOS), and my design had a limit right around there. Eventually, I scrapped that in favor of the current idea. I had over 1,000 genes/species determining behavior, and behavior was driven at a VERY low level. Unfortunately, somebody broke into my house and stole the computer that had the newest version of that design on it. That left me in a funk for almost a year. I have now redesigned the genes (though not yet coded the new design) in an even more elaborate scheme using as many as 2,000 genes/species.

I am currently hung up on my inability to do graphical programming. I am such a horrid artist that drawing on the screen is mostly trial and error. I knew this shortcoming in advance, so why I thought I would be able to do this is beyond me, but I'm slowly working on it.

I firmly believe that GA's, if given sufficient complexity, can solve most any problem. Having worked with a few of them, I would even go so far as to speculate that the technique I am using would allow for sufficient intelligence to evolve such that a robot could be dropped into a random kitchen, and given the existence of the proper items somewhere in the kitchen, it would be able to create a cheese omelette. The major stumbling block to this would be visual recognition, which is itself a MAJOR issue.

However, I have been sidetracked again. One of the other programs lost in the theft was the newest version of another odd GA program I wrote that evolves predictive equations. This was written in VB6, and it belongs in .NET, so, now that the new version was stolen, rather than fix the existing version, I am re-writing it the way it should be done.

Having said all that, I would sum it up like so: I think that idea is good enough that I have been pursuing it as a hobby for a few years now.

[yrwyddfa]

Actually, I'm already working on something that is very similar to this. My problem is the GUI (I SUCK as an artist).

I know the feeling.

The basic program is a lake with 8 organisms. The lowest level organisms feed on three base resource pools that are determined by a chaotic equation.

Do you use the May's logistic equation? Have you read the softening of the conjecture?

The reason for this is that a paer showed that three populations of plankton in a water body are guaranteed to be in a chaotic equilibrium. However, this may not be important. The original idea behind this program was a population simulation, but the behavioral aspects have come to dominate, and the population sim is now just a basis for encounters.

Are you saying that emergent behaviour rather than environmental factors determine the system, now?

I firmly believe that GA's, if given sufficient complexity, can solve most any problem. Having worked with a few of them, I would even go so far as to speculate that the technique I am using would allow for sufficient intelligence to evolve such that a robot could be dropped into a random kitchen, and given the existence of the proper items somewhere in the kitchen, it would be able to create a cheese omelette. The major stumbling block to this would be visual recognition, which is itself a MAJOR issue.

I think that some recent work in evolving neural nets, and genetic programming (as oppose to genetic algorithms) will make this a reality once the computing power is there. I think that Sony's(?) new 'Cell' processor is a massive step in the right direction.

However, I have been sidetracked again. One of the other programs lost in the theft was the newest version of another odd GA program I wrote that evolves predictive equations.

Cool: genetic programming, eh!

[Merri]

So, if you two want to make the algorithms, I can make it look more cool. I'm not too interested on the subject itself, but could try making some kind of visual display and make an interface. In the other hand, that wouldn't be like a contest. I think this isn't a working idea for a contest, it is too complex and time consuming to get into it (as Shaggy Hiker said: "I have been pursuing it as a hobby for a few years now").

It might work if we had teams. Atleast it sounds like a very time consuming project, so having more programmers on the problem might work if the time limit is one month. We could have team S and team Y, since you two seem to be the most knowleadgeable and interested on it (I hope I wrote that right)

[Shaggy Hiker]

Do you use the May's logistic equation? Have you read the softening of the conjecture?

I don't think so, and No. I use a pretty simple chaotic number generator:

Y = MX(1-X)

I think that's it. This equation (assuming I remember it correctly) exhibits chaotic behavior for certain values of M. I think M has to be around 3.7-3.9 for this to be chaotic.

Are you saying that emergent behaviour rather than environmental factors determine the system, now?

Emergent behavior will be the most interesting part of the result, though it may not be the only part of interest.

So, if you two want to make the algorithms, I can make it look more cool. I'm not too interested on the subject itself, but could try making some kind of visual display and make an interface. In the other hand, that wouldn't be like a contest. I think this isn't a working idea for a contest, it is too complex and time consuming to get into it (as Shaggy Hiker said: "I have been pursuing it as a hobby for a few years now").

It might work if we had teams. Atleast it sounds like a very time consuming project, so having more programmers on the problem might work if the time limit is one month. We could have team S and team Y, since you two seem to be the most knowleadgeable and interested on it (I hope I wrote that right)

It IS too big a task to do completely. Too big a bite to gnaw on at this point. However, there are smaller similar problems that could be worked on. One contest that was run a few times was to create programs that dealt with the Prisoner's Dilemma. Each program would have a strategy, and the different programs would compete against each other. While this particular problem has been dealt with in depth, a similar small world problem would probably be of interest.

[yrwyddfa]

Yeah looks like the same (or a slight variant) of May's thingy:

http://www.bendov.info/cours/chaos/logistic.htm

Very useful for modelling populations, indeed. I intended to use this somewhat to control the food population in the project above.

[yrwyddfa]

If this is not going to be accepted as a competition (shame, really) is anyone interested in some sort of a collaboration.

I reckon we're gonna need the following:

(i) Project manager
(ii) GUI guy
(iii) Systems guy
(iv) Database guy
(v) Science guy

I think the role titles are pretty explanatory. I'm prepared to take any role and am quite happy to be subject to the whims of the group.

[Merri]

I could try the GUI if you keep kicking my bottom head. If someone does something that works, I can probably make it look and work better. A well commented code would be a requirement.

[manavo11]

Since this doesn't look too much like a contest idea, although it is a nice idea, do you want me to split it from here and create a thread in project communication?

[yrwyddfa]

You might as well move it, I guess, but it looks like there's only very limited interest.

[manavo11]

Split and moved to project communication from the Future Contest Suggestions thread.

[wossname]

Arrgh! the number of things I want to get involved in keeps growing

Question: if you want people to design their organisms in IDL then what language are you going to test them in?

I shall follow this with interest, although I'll be too busy with D# to contribute much

[yrwyddfa]

Every language I can think of can consume IDL definitions. So the language choice is arbitrary and down to the choice of the programmer. (the interface will be a single inheritance one: which means it will only derive from IUnknown - this means the scripting languages are out; they can't do vTable binding)